A water conservancy hidden danger multi-source data dynamic registration and fusion method and system

Abstract

The application relates to the technical field of intelligent monitoring of water conservancy projects, and particularly discloses a water conservancy hidden danger multi-source data dynamic registration and fusion method and system; multi-source monitoring data is acquired and preprocessed, is mapped to a unified engineering space-time reference framework, a unified space-time index key is generated, initial space-time registration is performed through a global basic registration model; a reliability weight is calculated and normalized; abnormal event detection is performed based on the initial space-time registration result and the normalized reliability weight, a local re-registration area is generated; local space-time re-registration is performed with the monitoring data with the largest weight as a spatial reference, a corrected local registration relationship is obtained and written into the global basic registration model; reliable fusion calculation is performed based on the normalized reliability weight and the updated global basic registration model, and a fusion risk result is output. The application solves the problem that multi-source data cannot be accurately aligned and reliably fused in a local hidden danger scene, and improves hidden danger positioning accuracy and the explainability of fusion results.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for water conservancy projects, specifically to a method and system for dynamic registration and fusion of multi-source data on potential water conservancy hazards. Background Technology

[0002] With the continuous advancement of smart water conservancy construction, water conservancy project safety monitoring is gradually shifting from single-device monitoring to multi-source collaborative monitoring. Common data sources in actual projects include at least satellite remote sensing imagery, drone inspection imagery, data from fixed monitoring stations, IoT sensor data, on-site detection data collected by patrol robots, and basic structural data of the project. These data differ significantly in terms of coverage, sampling frequency, coordinate system, observation dimension, and measurement accuracy, but they are also clearly complementary.

[0003] Existing multi-source data processing schemes for water conservancy hazard monitoring typically include the following categories: a one-time global registration method based on a unified coordinate system, which pre-maps various types of data to a unified coordinate system and continues to use this global registration relationship in subsequent monitoring processes; a fusion method based on fixed empirical weights, which pre-sets fixed weights for each data source before calculating risk values; and an independent alarm method based on single-type anomaly thresholds, which sets thresholds for images, sensors, or on-site inspection results and judges them separately.

[0004] However, when seepage, termite activity, or minor surface deformation occurs in a local area of ​​a water conservancy project, the relative spatial relationship of different data sources in that local area will change. Moreover, the reliability of each data source changes dynamically with environmental conditions and equipment status. Existing technologies use globally unified registration relationships and fixed weights for fusion, lacking an adaptive correction mechanism for local registration driven by abnormal events. This makes it difficult to achieve accurate alignment and reliable fusion of multi-source data in local hidden danger scenarios, making it difficult to support hidden danger monitoring and decision-making in the actual application of water conservancy projects. Summary of the Invention

[0005] To address the technical problems in the prior art, this invention provides a method and system for dynamic registration and fusion of multi-source data on water conservancy hazards. It aims to solve the problem that in the prior art, due to the failure of global registration relationships and the dynamic reliability changes of data sources, multi-source monitoring data is difficult to accurately align and reliably fuse, thus making it difficult to support hazard monitoring and decision-making in the actual application of water conservancy projects.

[0006] The technical solution of the present invention is as follows: A method for dynamic registration and fusion of multi-source data on water conservancy hazards, comprising: Acquire multi-source monitoring data of the target water conservancy project area, and preprocess the multi-source monitoring data to obtain a preprocessed multi-source monitoring dataset; The multi-source monitoring dataset is mapped to a unified engineering spatiotemporal reference frame to generate a unified spatiotemporal index key for each monitoring data in the multi-source monitoring dataset. Initial spatiotemporal registration is performed on the monitoring data corresponding to the unified spatiotemporal index key through a global basic registration model, and the initial spatiotemporal registration result is output. The credibility weights of each monitoring data point in different regions and at different times are calculated by weighting, and the credibility weights of each monitoring data point are normalized to output normalized credibility weights. Based on the initial spatiotemporal registration results and the normalized confidence weights, abnormal event detection is performed on the multi-source monitoring dataset in the unified engineering spatiotemporal reference framework; when an abnormal event is detected, a local re-registration region is generated according to the location and time of the abnormal event. Within the local re-registration region, the normalized confidence weights of each monitoring data in the multi-source monitoring dataset are compared. The monitoring data with the largest weight value is used as the spatial reference. Local spatiotemporal re-registration is then performed on the other monitoring data in the local re-registration region to obtain the corrected local registration relationship. The corrected local registration relationship is then written into the global basic registration model to obtain the updated global basic registration model. Based on the normalized credibility weights and the updated global basic registration model, a credible fusion calculation is performed to output the fusion risk results of potential water conservancy project hazards.

[0007] Optionally, the step of mapping the multi-source monitoring dataset to the unified engineering spatiotemporal reference framework, generating a unified spatiotemporal index key for each monitoring data in the multi-source monitoring dataset, performing initial spatiotemporal registration on the monitoring data corresponding to the unified spatiotemporal index key through a global basic registration model, and outputting the initial spatiotemporal registration result specifically includes: Establish a unified engineering spatiotemporal reference framework that includes the WGS coordinate system, the engineering local reference system, the risk analysis grid, and a unified time axis; The original coordinates of each monitoring data in the multi-source monitoring dataset are mapped to the WGS coordinate system through georeferencing to obtain georeferencing coordinates; the georeferencing coordinates are mapped to the engineering local reference system through engineering projection to obtain engineering local coordinates; the engineering local coordinates are mapped to the risk analysis grid through grid operation to obtain grid index; The collection timestamps of each monitoring data are mapped to the unified time axis according to a preset step size to obtain the time slice index; The grid index and the time slice index form a unified spatiotemporal index key; The monitoring data corresponding to the unified spatiotemporal index key is input into the global basic registration model to obtain the registered observation values ​​of the monitoring data in each grid cell of the risk analysis grid and each time slice of the unified time axis. The registered observations, along with the grid index and the time slice index, are used as the initial spatiotemporal registration result.

[0008] Optionally, the global basic registration model is constructed in the following way: Obtain the coordinate transformation rules, grid mapping rules, and time index rules corresponding to each monitoring data in the multi-source monitoring dataset. Based on the coordinate transformation rules, grid mapping rules, and time index rules, construct spatial transformation operators, grid merging operators, and time alignment operators, respectively. The time alignment operator, the spatial transformation operator, and the mesh merging operator are combined to obtain a composite operator; The composite operator is used as the global basic registration model.

[0009] Optionally, the step of calculating the credibility weight of each monitoring data point in different regions and at different times using weighted averages, and normalizing the credibility weight of each monitoring data point to output a normalized credibility weight, specifically includes: Obtain the device accuracy factor, time freshness factor, environmental impact factor, historical stability factor, and neighborhood consistency factor for each monitoring data in the multi-source monitoring dataset; The reliability weight of each monitoring data in different regions and at different times is obtained by weighting and summing the equipment accuracy factor, the time freshness factor, the environmental impact factor, the historical stability factor, and the neighborhood consistency factor. The normalized confidence weight is obtained by dividing the confidence weight of each monitoring data in the multi-source monitoring dataset by the sum of the confidence weights of each monitoring data.

[0010] Optionally, the step of performing anomaly detection on the multi-source monitoring dataset in the unified engineering spatiotemporal reference framework based on the initial spatiotemporal registration result and the normalized confidence weight specifically includes: For each grid cell in the risk analysis grid, obtain the registered observation values ​​of each monitoring data falling into the current grid cell; Calculate the single-source anomaly indication value of each monitoring data based on the registered observation values ​​of each monitoring data in the multi-source monitoring dataset; The anomaly score of the current grid cell is obtained by weighting and summing the normalized confidence weight of each monitoring data with the corresponding single-source anomaly indication value. When the abnormal score exceeds a preset threshold, it is determined that there is an abnormal event in the current grid cell.

[0011] Optionally, when an abnormal event is detected, generating a local reregistration region based on the location and time of the abnormal event specifically includes: Obtain the grid index of the grid cell where the abnormal event is located as the abnormal location, and obtain the time when the abnormal event is detected as the abnormal time. Using the grid cell corresponding to the abnormal location as the center, a predetermined number of grid cells are expanded outward in the plane to generate a spatial boundary; A time window is generated by extending a preset time period forward and backward from the aforementioned abnormal moment. The spatial boundary and the time window are used together as the local reregistration region.

[0012] Optionally, within the local reregistration region, the normalized confidence weights of each monitoring data in the multi-source monitoring dataset are compared. Using the monitoring data with the largest weight as the spatial reference, local spatiotemporal reregistration is performed on the other monitoring data in the local reregistration region to obtain the corrected local registration relationship. Specifically, this includes: Within the local reregistration region, the monitoring data with the highest normalized confidence weight is selected as the local spatial reference. Using the coordinate system of the local spatial reference as the target, at least one of feature matching, iterative nearest point algorithm, or local affine transformation is used to perform local spatial reregistration on other monitoring data within the local reregistration area to obtain local spatial transformation parameters; Local time reregistration is performed on each monitoring data within the local reregistration area to align all monitoring data to a unified time axis, thereby obtaining local time alignment parameters; The local spatial transformation parameters and the obtained local temporal alignment parameters are used together as the corrected local registration relationship.

[0013] Optionally, the step of writing the corrected local registration relationship into the global basic registration model to obtain the updated global basic registration model specifically includes: Obtain the subset of registration parameters corresponding to the local reregistration region in the global basic registration model; The corrected local registration relationship is used to replace the corresponding parameters in the registration parameter subset to obtain the updated global basic registration model.

[0014] Optionally, the step of performing trusted fusion calculation based on the normalized trustworthiness weights and the updated global basic registration model, and outputting the fusion risk result, specifically includes: By using the updated global basic registration model, the re-registered observations of each monitoring data point are obtained on each grid cell and the unified time axis. The re-registered observations of each monitoring data point are converted into risk characterization values ​​through a risk mapping function. The normalized confidence weights of each monitoring data point are weighted and summed with the corresponding risk characterization values ​​to obtain the fused risk value. Calculate the first product of the normalized confidence weight of each monitoring data and the risk characterization value, calculate the sum of the first products of all monitoring data, divide the first product by the sum of the first products, and obtain the source contribution of each monitoring data to the fusion risk value. The sum of the products of the normalized confidence weights of each monitoring data and the neighborhood consistency factor is used as the confidence level of the result. The fusion risk value, the source contribution, and the result confidence level are output as the fusion risk result.

[0015] This invention also provides a dynamic registration and fusion system for multi-source data on water conservancy hazards, comprising: The data acquisition module is used to acquire and preprocess multi-source monitoring data to obtain a multi-source monitoring dataset; The initial registration module is used to map the multi-source monitoring dataset to a unified engineering spatiotemporal reference frame, and to perform initial spatiotemporal registration through a global basic registration model, and output the initial spatiotemporal registration result. The weight calculation module is used to calculate the credibility weight of each monitoring data and perform normalization processing, and output the normalized credibility weight. An anomaly detection module is used to detect anomalies based on the initial spatiotemporal registration result and the normalized confidence weight, and to generate a local re-registration region based on the detected anomalies. The local re-registration module is used to perform local spatiotemporal re-registration within the local re-registration region, using the monitoring data with the largest weight value as the spatial reference, to obtain the corrected local registration relationship and update the global basic registration model. The fusion output module is used to perform trusted fusion calculations based on the normalized trustworthiness weights and the updated global basic registration model, and output the fusion risk results.

[0016] Compared with existing technologies, the present invention provides a method and system for dynamic registration and fusion of multi-source data on water conservancy hazards, which has the following beneficial effects: (1) By constructing a unified engineering spatiotemporal reference framework and performing initial spatiotemporal registration, multi-source monitoring data are mapped to each grid cell of a unified risk analysis grid and each time slice of a unified time axis, which solves the problem that multi-source monitoring data are difficult to align due to different coordinate systems, sampling frequencies and observation dimensions. By triggering a local re-registration mechanism through abnormal events, a local re-registration area is generated with the grid cell where the abnormal location is located as the center when an abnormal event is detected. The monitoring data with the largest weight value is used as the spatial reference, and local spatiotemporal re-registration is performed on other monitoring data in the local re-registration area. The corrected local registration relationship is written into the global basic registration model to update the registration parameters in the local re-registration area. This overcomes the defect of large hazard location error caused by the failure of the global registration relationship in the local abnormal area in the prior art, and realizes accurate alignment of multi-source monitoring data in local hazard scenarios.

[0017] (2) By constructing a weighted model that includes equipment accuracy factor, time freshness factor, environmental impact factor, historical stability factor and neighborhood consistency factor, the credibility weight of each monitoring data in different regions and at different times is calculated and the credibility weight of each monitoring data is normalized, so that the credibility weight of each monitoring data can be adjusted in real time with the changes in environmental conditions and equipment status, avoiding the drawbacks of the existing fixed weight scheme that cannot reflect the dynamic changes in the quality of multi-source monitoring data, and improving the stability and reliability of the fusion risk results.

[0018] (3) By forming a closed-loop linkage between anomaly detection, local spatiotemporal reregistration and trusted fusion calculation, local spatiotemporal reregistration is automatically executed and the global basic registration model is updated after anomaly detection is triggered. The updated global basic registration model is then used for subsequent trusted fusion calculation, breaking the limitation that anomaly detection, spatiotemporal registration and fusion analysis are independent of each other in the existing system. At the same time, by outputting the fusion risk value, the contribution of each monitoring data to the fusion risk value, and the confidence level of the result, interpretable fusion risk results and verification basis are provided for engineers, which facilitates accurate decision-making and on-site investigation.

[0019] (4) By adopting a unified engineering spatiotemporal reference framework and a global basic registration model to perform initial spatiotemporal registration, efficient alignment in the initial registration stage is achieved; by adopting at least one of feature matching, iterative nearest point algorithm or local affine transformation in the local spatiotemporal re-registration stage, local spatial re-registration is performed on other monitoring data in the local re-registration area, and the re-registered observations are converted into risk characterization values ​​through a risk mapping function and then weighted and fused, taking into account both global consistency and local accuracy, and improving the overall efficiency and intelligence level of water conservancy project hidden danger monitoring. Attached Figure Description

[0020] Figure 1This is a flowchart illustrating the steps of a method for dynamic registration and fusion of multi-source data on water conservancy hazards according to the present invention. Figure 2 This is a schematic diagram of the structure of a dynamic registration and fusion system for multi-source data on water conservancy hazards according to the present invention. Detailed Implementation

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

[0022] Please see Figure 1 This invention provides a method for dynamic registration and fusion of multi-source data on water conservancy hazards, including steps S10 to S60.

[0023] S10: Obtain multi-source monitoring data of the target water conservancy project area, and preprocess the multi-source monitoring data to obtain a preprocessed multi-source monitoring dataset.

[0024] In this application, the first step is to acquire multi-source monitoring data for the target water conservancy project area. This multi-source monitoring data includes at least two different types of data: remote sensing imagery data, drone inspection data, IoT sensor data, patrol robot detection data, engineering structure data, and historical inspection records or historical risk labeling data. These data come from a wide range of sources and have varying formats, requiring standardized preprocessing.

[0025] The preprocessing includes timestamp standardization, coordinate information parsing, missing value marking, noise removal, unit unification, and quality label generation. Specifically, timestamp standardization converts time information from different data sources into a unified time format; coordinate information parsing extracts or infers spatial coordinate information from the original data; missing value marking marks the locations of missing data for subsequent processing; noise removal uses filtering or statistical methods to remove obvious abrupt changes; unit unification converts physical quantities in different units to a unified unit; and quality label generation generates an initial quality label for each data point. Through the above preprocessing, the original data is converted into standardized data with a unified format and controllable quality, providing input for subsequent steps.

[0026] S20: Map the multi-source monitoring dataset to a unified engineering spatiotemporal reference frame, generate a unified spatiotemporal index key for each monitoring data in the multi-source monitoring dataset, perform initial spatiotemporal registration on the monitoring data corresponding to the unified spatiotemporal index key through a global basic registration model, and output the initial spatiotemporal registration result.

[0027] In the application, a unified engineering spatiotemporal reference framework is first established, which includes a risk analysis grid and a unified time axis. The risk analysis grid consists of several discrete grid cells for spatial positioning, and the unified time axis consists of multiple discrete time slices for temporal positioning. For each monitoring data, spatial mapping and temporal mapping are performed to obtain the grid index and time slice index corresponding to each monitoring data, which together form a unified spatiotemporal index key. Based on this, initial spatiotemporal registration is performed on each monitoring data to align each monitoring data spatially to each grid cell and temporally to each time slice, and the initial spatiotemporal registration result is output.

[0028] In some embodiments, the global base registration model is constructed in the following manner: Obtain the coordinate transformation rules, grid mapping rules, and time indexing rules corresponding to each monitoring data in the multi-source monitoring dataset. Based on the coordinate transformation rules, grid mapping rules, and time indexing rules, construct a spatial transformation operator, a grid merging operator, and a time alignment operator, respectively. Combine the time alignment operator, the spatial transformation operator, and the grid merging operator to obtain a composite operator. Use the composite operator as the global basic registration model.

[0029] First, for each monitoring data point in the multi-source monitoring dataset, obtain its corresponding coordinate transformation rules, grid mapping rules, and time indexing rules. These rules are derived from the Unified Engineering Spatiotemporal Reference Framework. Based on these rules, three operators are constructed respectively:

[0030] Time alignment operator The time alignment operator is used to resample the original time series of various monitoring data onto a unified time axis; let the i-th monitoring data be in the original time... The observed values ​​on The target time slice index is The time value corresponding to this time slice is .

[0031] In one implementation, linear interpolation is used: ; in, and Encircling time value The adjacent valid sampling times satisfy < < .

[0032] In another implementation, for sparsely sampled data sources, nearest neighbor alignment can be used: ; That is, to take Minimum moment The observed values ​​are used as the alignment result.

[0033] Spatial transformation operator Spatial transformation operators are used to map the spatial observation positions of each monitoring data point to a unified grid coordinate system; in one implementation, initial global registration on the engineering plane or local tangent plane employs a two-dimensional affine transformation. ; in, For rotation and scaling matrices, For translation vectors; parameter matrix Translation vector Composed of no less than three pairs of control points and The solution is obtained by least squares.

[0034] When perspective distortion exists, a 3×3 homography matrix can be used. Homogeneous coordinate form: ; In a three-dimensional implementation, it can be expanded into a rigid body rotation matrix. With translation vector : .

[0035] Grid merge operator The grid merging operator is used to merge indices that fall into the same grid index j and the same time slice index. The set of multiple observations is as follows: Where K is the total number of observations, and the merging rule can be to take the mean: ; Or take the maximum value: ; Alternatively, a representative value can be selected based on prior confidence.

[0036] Combining the three operators above yields a composite operator. : ; With composite operators As a global basic registration model The model will use the original observation sequence This is mapped to a unified representation across each grid cell of the risk analysis grid and each time slice on a unified time axis: ; Where j is the grid index, For time slice index, To unify the spatiotemporal index key.

[0037] In some embodiments, the multi-source monitoring dataset is mapped to the unified engineering spatiotemporal reference frame to generate a unified spatiotemporal index key for each monitoring data in the multi-source monitoring dataset. Initial spatiotemporal registration is then performed on the monitoring data corresponding to the unified spatiotemporal index key using a global basic registration model, and the initial spatiotemporal registration result is output. Specifically, this includes: Establish a unified engineering spatiotemporal reference framework that includes the WGS coordinate system, the engineering local reference system, the risk analysis grid, and a unified time axis; The original coordinates of each monitoring data point in the multi-source monitoring dataset are mapped to the WGS coordinate system through georeferencing to obtain georeferencing coordinates. The georeferencing coordinates are then mapped to the engineering local reference system through engineering projection to obtain engineering local coordinates. The engineering local coordinates are then mapped to the risk analysis grid through lattice operations to obtain a grid index. The collection timestamps of each monitoring data point are mapped to the unified time axis according to a preset step size to obtain a time slice index. The grid index and the time slice index form a unified spatiotemporal index key. The monitoring data corresponding to the unified spatiotemporal index key are input into the global basic registration model to obtain the registered observations of the monitoring data in each grid cell of the risk analysis grid and each time slice of the unified time axis. The registered observations, the grid index, and the time slice index are used together as the initial spatiotemporal registration result.

[0038] In application, a unified engineering spatiotemporal reference framework is first established, which includes the WGS coordinate system, the engineering local reference system, the risk analysis grid, and the unified time axis. The WGS coordinate system serves as a globally unified geographic reference benchmark, while the engineering local reference system is used to adapt to the local coordinate requirements of specific water conservancy projects.

[0039] For the original coordinates are The monitoring data will Mapping to the WGS coordinate system via a seven-parameter / four-parameter plane transformation or homography transformation yields... This step eliminates macroscopic coordinate differences between different data sources; Projected onto the local engineering reference frame ,get Coordinates along the dam axis and dam body zoning can be established. A two-way mapping between geographic and planar coordinates; this step converts geographic coordinates into local coordinates in the engineering coordinate system, facilitating alignment with the engineering structure; The grid index j is obtained by mapping the data onto the risk analysis grid using a logarithmic operation. ; The grid assignment operation adopts the nearest neighbor grid center principle, that is, selecting the grid cell center with the closest Euclidean distance as the assigned grid; if it falls into the grid boundary neighborhood, the nearest grid center or the principle of the largest area proportion can be used to determine a unique j; the collection timestamp will be used. According to the preset step size Mapped onto a unified timeline, resulting in a time slice index. : ; Preset step size It can be set according to monitoring needs. For example, for continuous changes such as seepage pressure, a 15-minute step can be used; for discrete sampling such as drone inspections, a 1-hour step can be used.

[0040] Using grid index j and time slice index Construct a unified spatiotemporal index key Input the monitoring data corresponding to the index key into the global basic registration model constructed in step one. The registered observations were obtained as follows: ; This value represents the grid cell and time slice index corresponding to the i-th monitoring data at grid index j. The registered observations on the corresponding time slices; the registered observations With grid index j, time slice index Together they serve as the initial spatiotemporal registration result output.

[0041] S30: Calculate the credibility weight of each monitoring data in different regions and at different times using weighted average, normalize the credibility weight of each monitoring data, and output the normalized credibility weight.

[0042] In application, the core of this step is to evaluate the reliability of each monitoring data point in the current region and at the current time. For each monitoring data point, a reliability weight is calculated by comprehensively considering factors such as the inherent accuracy of the equipment, the timeliness of the data, the current environmental conditions, the historical operational stability, and the consistency with surrounding multi-source data. This weight reflects the reliability of the monitoring data under the current conditions. To facilitate subsequent fusion calculations, the reliability weights of each monitoring data point are normalized so that the sum of all weights is 1, and the normalized reliability weight is output.

[0043] In some embodiments, a weighted calculation is used to determine the credibility weight of each monitoring data point in different regions and at different times, and the credibility weight of each monitoring data point is normalized to output a normalized credibility weight. Specifically, this includes: Obtain the device accuracy factor, time freshness factor, environmental impact factor, historical stability factor, and neighborhood consistency factor for each monitoring data in the multi-source monitoring dataset; The reliability weight of each monitoring data in different regions and at different times is obtained by weighting and summing the equipment accuracy factor, the time freshness factor, the environmental impact factor, the historical stability factor, and the neighborhood consistency factor. The normalized confidence weight is obtained by dividing the confidence weight of each monitoring data in the multi-source monitoring dataset by the sum of the confidence weights of each monitoring data.

[0044] In this application, this embodiment provides a specific method for calculating the credibility weight, using a five-factor weighted model: ; The correspondence between each weight coefficient and each factor is as follows: Corresponding equipment accuracy factor , Corresponding time freshness factor , Corresponding environmental impact factors , Corresponding historical stability factor , Corresponding neighborhood consistency factor The sum of the ranges of values ​​for each weight coefficient is 1.

[0045] Among them, equipment accuracy factor : Characterizes the inherent measurement accuracy of the equipment used for the i-th monitoring data; the calculation formula is: ; in, The calibration or statistical observation error of the i-th monitoring data is... The maximum allowable error threshold; for example, fixed station equipment. A value of 0.85 to 0.95 is acceptable for drone equipment. A value of 0.75–0.85 is acceptable for satellite imagery. A value of 0.65 to 0.80 is acceptable.

[0046] Time freshness factor : Represents the timeliness of the i-th monitoring data at time t; the calculation formula is: ; in, To monitor the time lag of data i relative to the analysis time t, >0 represents the attenuation coefficient; for example, real-time streaming data A value of 0.90 to 1.0 is acceptable; for data with a delay of less than 1 hour, a value of 0.75 to 0.90 is acceptable.

[0047] Environmental impact factors : Represents the degree to which the i-th monitoring data is affected by environmental factors in the time slice corresponding to grid index j; determined by piecewise functions or table lookup based on environmental conditions such as rainfall, wind speed, humidity, light intensity, visibility, and electromagnetic interference; for example, the values ​​of environmental impact factors can be found in Table 1: When multiple environmental impact items are in effect simultaneously, the minimum value is taken as the comprehensive environmental impact factor. ; in, The coefficients are obtained for the kth environmental impact item according to Table 1; the corresponding values ​​are taken for image-type monitoring data, sensor-type monitoring data, and close-range monitoring data.

[0048] Historical stability factor : Characterizes the historical operational stability of the i-th monitoring data at time t; the calculation formula is: ; in, For drift rate, For packet loss rate, For example, abnormal volatility; during the stable period. A value of 0.80 to 0.95 is acceptable; for significant drift, a value of 0.50 to 0.70 is acceptable.

[0049] Neighborhood Consistency Factor : Represents the consistency between the observation value of the i-th monitoring data in the grid cell corresponding to grid index j and the time slice corresponding to time slice index j, and the multi-source mean; the calculation formula is: ; in, Let i be the observed value of the i-th monitoring data. This represents the mean from multiple sources within the same region, where M is the maximum permissible deviation; for example, when multiple sources are consistent... A value of 0.85 to 1.0 is acceptable; for single-source sources with large deviations, a value of 0.40 to 0.65 is acceptable.

[0050] In one specific embodiment, take =0.25, =0.20, =0.20, =0.20, =0.15; In engineering practice, the above values ​​can be fine-tuned, with the adjustment range not exceeding ±0.05, and the sum of each weight coefficient should be kept at 1; The specific values ​​of the weight coefficients can be determined based on historical sample statistics, engineering type templates, expert initial values, or online feedback.

[0051] Finally, normalization is performed: .

[0052] S40: Based on the initial spatiotemporal registration result and the normalized confidence weight, perform abnormal event detection on the multi-source monitoring dataset in the unified engineering spatiotemporal reference framework; when an abnormal event is detected, generate a local re-registration region according to the location and time of the abnormal event.

[0053] In application, this step first detects abnormal events based on the initial spatiotemporal registration results and normalized confidence weights; for each risk analysis grid cell, it judges whether there are any abnormalities by integrating information from multiple monitoring data; when an abnormal event is detected, it further generates a local reregistration region based on the location and time of the abnormal event, providing spatial and temporal range for subsequent local reregistration.

[0054] In some embodiments, based on the initial spatiotemporal registration result and the normalized confidence weight, abnormal event detection is performed on the multi-source monitoring dataset in the unified engineering spatiotemporal reference framework, specifically including: For each grid cell in the risk analysis grid, the registered observation values ​​of each monitoring data falling into the current grid cell are obtained; the single-source anomaly indication value of each monitoring data is calculated based on the registered observation values ​​of each monitoring data in the multi-source monitoring dataset; the normalized confidence weight of each monitoring data is weighted and summed with the corresponding single-source anomaly indication value to obtain the anomaly score of the current grid cell; when the anomaly score exceeds a preset threshold, it is determined that there is an abnormal event in the current grid cell.

[0055] In the application, for each grid cell j in the risk analysis grid, the registered observations of each monitoring data falling into that grid cell are obtained. For each monitoring data point, based on its registered observation value... Calculate single-source anomaly indication value The single-source anomaly indicator value can be an anomaly score or anomaly probability, with a value range of [0, 1]. The larger the value, the higher the probability of an anomaly.

[0056] For each grid cell, the normalized confidence weights of each monitoring data are... With the corresponding single-source anomaly indication value Perform a weighted summation to obtain the outlier scores. : ; When abnormal scores Exceeding the preset threshold When an abnormal event occurs in the current grid cell, the abnormal score is used only to determine whether local reregistration is triggered. It is used as an internal trigger signal and is not directly output to the user as a fusion result.

[0057] Among them, the preset threshold Take as the baseline standard deviation Multiples of: ; in, The standard deviation of the historical baseline value is given; the coefficient k ranges from 2.5 to 4.0, and the specific value can be adjusted according to the project's requirements for sensitivity and false alarm rate.

[0058] In some embodiments, when an abnormal event is detected, a local reregistration region is generated based on the location and time of the abnormal event, specifically including: The grid index of the grid cell where the abnormal event is located is obtained as the abnormal location, and the time when the abnormal event is detected is obtained as the abnormal time. A predetermined number of grid cells are expanded outward in the plane with the grid cell corresponding to the abnormal location as the center to generate a spatial boundary. A predetermined time period is expanded forward and backward with the abnormal time as the center to generate a time window. The spatial boundary and the time window are used together as the local re-registration region.

[0059] In the application, when an abnormal event is detected, the grid index of the grid cell where the abnormal event occurs is obtained. As the location of the anomaly, obtain the moment when the anomaly event was detected. As an abnormal moment, a predetermined number of grid cells are expanded outward in the plane, centered on the grid cell corresponding to the abnormal location, to generate a spatial boundary.

[0060] Exemplarily, extending in a plane Each grid cell, with radius when isotropic 5 to 10 grids can be selected; in three dimensions, further expansion can be made vertically. Grid, exemplarily 2 to 6 grids can be selected; in abnormal situations Centered on the timeframe, a time window is generated by extending the preset duration forward and backward: ; T- and T+ are preset time durations, both ranging from 30 minutes to 6 hours. In rapidly evolving scenarios, such as sudden changes in seepage pressure during heavy rainfall and rapid development of piping, the abnormal state changes rapidly, requiring a shorter monitoring window to capture the dynamic changes of the anomaly. In this case, a shorter time window, such as 30 minutes to 2 hours, is preferable. In slowly deforming scenarios, such as long-term dam settlement and slow seepage development, the abnormal state changes gradually, requiring a longer time window to observe the development trend of the anomaly. In this case, a longer time window, such as 2 to 6 hours, is preferable. The specific value of the time window can be determined comprehensively based on the sampling frequency of the monitoring data, the type of anomaly, and engineering experience.

[0061] Finally, a local reregistration region is formed, which includes both the spatial boundary and the temporal window, denoted as . .

[0062] S50: Within the local re-registration region, compare the normalized confidence weights of each monitoring data in the multi-source monitoring dataset, take the monitoring data with the largest weight value as the spatial reference, and perform local spatiotemporal re-registration on the other monitoring data in the local re-registration region to obtain the corrected local registration relationship; write the corrected local registration relationship into the global basic registration model to obtain the updated global basic registration model.

[0063] In this application, this step performs local spatiotemporal reregistration on the monitoring data within the local reregistration region. First, the normalized confidence weights of each monitoring data point are compared, and the monitoring data with the largest weight value is selected as the local spatial reference. Then, using the coordinate system of this reference as the target, local spatial reregistration and local temporal reregistration are performed on other monitoring data within the region to obtain the corrected local registration relationship. Finally, this corrected relationship is written into the global base registration model to achieve local updates of the model.

[0064] In some embodiments, within the local reregistration region, the normalized confidence weights of each monitoring data in the multi-source monitoring dataset are compared. Using the monitoring data with the largest weight as the spatial reference, local spatiotemporal reregistration is performed on the other monitoring data in the local reregistration region to obtain a corrected local registration relationship. Specifically, this includes: Within the local reregistration region, the monitoring data with the highest normalized confidence weight is selected as the local spatial reference. Using the coordinate system of the local spatial reference as the target, at least one of feature matching, iterative nearest point algorithm, or local affine transformation is employed to perform local spatial reregistration on other monitoring data within the local reregistration region, obtaining local spatial transformation parameters. Local temporal reregistration is then performed on each monitoring data within the local reregistration region, aligning all monitoring data to a unified time axis, obtaining local temporal alignment parameters. The local spatial transformation parameters and the local temporal alignment parameters are used together as the corrected local registration relationship.

[0065] In applications, in local reregistration regions Internally, based on normalized confidence weights The monitoring data with the largest weight value is selected as the local spatial reference; using the coordinate system of the local spatial reference as the target, local spatial reregistration is performed on the other monitoring data; and the local spatial transformation parameters are solved using at least one of feature matching, iterative nearest neighbor algorithm (ICP), or local affine transformation. This minimizes the local mean square error. ; in, The coordinates of the monitoring data to be registered. Coordinates of a local spatial reference. These are robust kernel functions, such as the Huber kernel, used to reduce the impact of outliers on the registration results.

[0066] Local temporal reregistration is performed on each monitoring data point within the local reregistration region. Specifically, for anomalous evolution segments, cubic spline interpolation or linear interpolation is used to align the monitoring data points to a unified time axis. For sparsely sampled monitoring data, nearest neighbor matching is used to align them to the unified time axis. Preferably, interpolation is used for the main monitoring sequence, and nearest neighbor matching is used for the auxiliary sparse sequences. The identifier used is recorded in the output to obtain the local temporal alignment parameters. ; local space transformation parameters and local time alignment parameters Together they form the corrected local registration relation.

[0067] In some embodiments, the modified local registration relationship is written into the global basic registration model to obtain an updated global basic registration model, specifically including: Obtain the subset of registration parameters corresponding to the local re-registration region in the global basic registration model; replace the corresponding parameters in the subset of registration parameters with the corrected local registration relationship to obtain the updated global basic registration model.

[0068] In applications, obtain local reregistration regions. In the global basic registration model The corresponding registration parameter subset; the current registration parameter subset includes spatial transformation parameters and temporal parameters within the local region; among which A set of grid indices within the spatial boundary. The set of time slice indices within the time window; the local spatial transformation parameters in the corrected local registration relationship. and local time parameters Replace the corresponding parameters in the subset of registration parameters, and smoothly connect the updated local registration result with the global registration result at the region boundary.

[0069] Record the global basic registration model The observed values ​​are as follows The locally corrected observation value is The transition weight is The value is 0 at the spatial boundary and 1 in the anomaly core region, for example, due to distance decay; the fused observation value is: ; Where j is the grid index, For time slice index, It depends only on the position of the grid cell corresponding to grid index j, and not on the time slice index. Irrelevant; the corrected local registration relation, i.e., the local spatial transformation parameters and local time parameters Write into the global base registration model The updated global base registration model is obtained. Updated global base registration model In the local reregistration region The result after updating the parameters retains the original global base registration model outside the local reregistration region. .

[0070] S60: Based on the normalized credibility weights and the updated global basic registration model, perform credible fusion calculations and output the fusion risk results of potential water conservancy project hazards.

[0071] In application, this step is based on normalized confidence weights and the updated global base registration model. Perform trusted fusion computation; obtain the re-registered observations of each monitoring data point through the updated model, convert them into risk characterization values, and then sum them by weight to obtain the fusion risk value. At the same time, calculate the source contribution and result confidence of each monitoring data point as interpretable output results.

[0072] In some embodiments, the step of performing trusted fusion calculation based on the normalized trustworthiness weights and the updated global base registration model, and outputting fusion risk results, specifically includes: Using the updated global registration model, the re-registered observations of each monitoring data point are obtained across each grid cell and on a unified time axis. These re-registered observations are then converted into risk characterization values ​​using a risk mapping function. The normalized confidence weights of each monitoring data point are weighted and summed with their corresponding risk characterization values ​​to obtain the fusion risk value. The first product of the normalized confidence weights and risk characterization values ​​of each monitoring data point is calculated, and the sum of these first products is obtained. The first product is then divided by the sum of the first products to obtain the source contribution of each monitoring data point to the fusion risk value. The sum of the products of the normalized confidence weights and neighborhood consistency factors of each monitoring data point is used as the result confidence level. Finally, the fusion risk value, the source contribution, and the result confidence level are output as the fusion risk result.

[0073] In the application, the updated global base registration model is used. Obtain the re-registered observations of each monitoring data point in each grid cell and time slice. In the local reregistration region Within this range, the observation has been modified by the local spatial transformation parameters. and The description refers to the local spatiotemporal reregistration correction; outside the local reregistration region, the observation retains the initial spatiotemporal registration result, i.e., it is based on the initial global base registration model. Acquire; re-register the observations from each monitoring data point. Through risk mapping function Convert to risk characterization value In one implementation: ; in, and The historical statistical mean and standard deviation of this monitoring data; The Sigmoid function can be used: ; If the monitoring data directly outputs the probability of abnormality Then it is acceptable: ; Normalized confidence weights for each monitoring data point and the corresponding risk characterization value Perform a weighted summation: ; Fusion Risk Value This reflects the grid cell and time slice index corresponding to grid index j. The overall risk level for the corresponding time slice ranges from [0, 1]. This fusion risk value is provided to the user as one of the final output results.

[0074] It is worth noting: Fusion Risk Value The abnormal score calculated above The following differences exist: Anomaly scores use single-source anomaly indicator values. The risk value is integrated using the risk characterization value. Anomaly scores are based on the initial global registration model. The calculated and fused risk values ​​are based on the updated global base registration model. calculate; The anomaly score serves as an internal trigger signal to determine whether to perform local reregistration, and is not directly output to the user; the fusion risk value is provided to the user as the final output result.

[0075] The contribution of a source is calculated as follows: ; in, Small positive numbers, such as 10 -6 This is used to prevent the denominator from being zero; source contribution. This reflects the contribution ratio of each monitoring data point to the fusion risk value, and meets the requirements. =1.

[0076] The confidence level of the result is calculated as follows: ; in, The neighborhood consistency factor characterizes the consistency between the observed values ​​of the monitoring data in region r and time t and the multi-source means; the result confidence level This reflects the reliability of the fusion result, with a value ranging from [0, 1]. A larger value indicates a more reliable fusion result; the fusion risk value... Contribution of sources and result confidence As a result of fusion risk output.

[0077] Please see Figure 2 The present invention also provides a dynamic registration and fusion system for multi-source data on water conservancy hazards, comprising: The data acquisition module is used to acquire and preprocess multi-source monitoring data to obtain a multi-source monitoring dataset; The initial registration module is used to map the multi-source monitoring dataset to a unified engineering spatiotemporal reference frame, and to perform initial spatiotemporal registration through a global basic registration model, and output the initial spatiotemporal registration result. The weight calculation module is used to calculate the credibility weight of each monitoring data and perform normalization processing, and output the normalized credibility weight. An anomaly detection module is used to detect anomalies based on the initial spatiotemporal registration result and the normalized confidence weight, and to generate a local re-registration region based on the detected anomalies. The local re-registration module is used to perform local spatiotemporal re-registration within the local re-registration region, using the monitoring data with the largest weight value as the spatial reference, to obtain the corrected local registration relationship and update the global basic registration model. The fusion output module is used to perform trusted fusion calculations based on the normalized trustworthiness weights and the updated global basic registration model, and output the fusion risk results.

[0078] It is understandable that the data acquisition module, initial registration module, weight calculation module, anomaly detection module, local reregistration module, and fusion output module are sequentially connected to form a complete processing link from multi-source monitoring data acquisition to fusion risk result output. Among them, the anomaly detection module and the local reregistration module form a feedback loop: after the anomaly detection module detects an anomaly event, it triggers the local reregistration module to perform local spatiotemporal reregistration and update the global basic registration model. The updated global basic registration model is fed back to the fusion output module for subsequent trusted fusion calculation, thereby realizing anomaly event-driven adaptive correction of local registration.

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic registration and fusion of multi-source data on water conservancy hazards, characterized in that, include: Acquire multi-source monitoring data of the target water conservancy project area, and preprocess the multi-source monitoring data to obtain a preprocessed multi-source monitoring dataset; The multi-source monitoring dataset is mapped to a unified engineering spatiotemporal reference frame to generate a unified spatiotemporal index key for each monitoring data in the multi-source monitoring dataset. Initial spatiotemporal registration is performed on the monitoring data corresponding to the unified spatiotemporal index key through a global basic registration model, and the initial spatiotemporal registration result is output. The credibility weights of each monitoring data point in different regions and at different times are calculated by weighting, and the credibility weights of each monitoring data point are normalized to output normalized credibility weights. Based on the initial spatiotemporal registration results and the normalized confidence weights, anomaly event detection is performed on the multi-source monitoring dataset in the unified engineering spatiotemporal reference framework; when an anomaly event is detected, a local re-registration region is generated according to the location and time of the anomaly event. Within the local re-registration region, the normalized confidence weights of each monitoring data in the multi-source monitoring dataset are compared. The monitoring data with the largest weight value is used as the spatial reference. Local spatiotemporal re-registration is then performed on the other monitoring data in the local re-registration region to obtain the corrected local registration relationship. The corrected local registration relationship is then written into the global basic registration model to obtain the updated global basic registration model. Based on the normalized credibility weights and the updated global basic registration model, a credible fusion calculation is performed to output the fusion risk results of potential water conservancy project hazards.

2. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 1, characterized in that, The process of mapping the multi-source monitoring dataset to the unified engineering spatiotemporal reference framework, generating a unified spatiotemporal index key for each monitoring data in the multi-source monitoring dataset, performing initial spatiotemporal registration on the monitoring data corresponding to the unified spatiotemporal index key through a global basic registration model, and outputting the initial spatiotemporal registration result specifically includes: Establish a unified engineering spatiotemporal reference framework that includes the WGS coordinate system, the engineering local reference system, the risk analysis grid, and a unified time axis; The original coordinates of each monitoring data in the multi-source monitoring dataset are mapped to the WGS coordinate system through georeferencing to obtain georeferencing coordinates; the georeferencing coordinates are mapped to the engineering local reference system through engineering projection to obtain engineering local coordinates; the engineering local coordinates are mapped to the risk analysis grid through grid operation to obtain grid index; The collection timestamps of each monitoring data are mapped to the unified time axis according to a preset step size to obtain the time slice index; The grid index and the time slice index form a unified spatiotemporal index key; The monitoring data corresponding to the unified spatiotemporal index key is input into the global basic registration model to obtain the registered observation values ​​of the monitoring data in each grid cell of the risk analysis grid and each time slice of the unified time axis. The registered observations, along with the grid index and the time slice index, are used as the initial spatiotemporal registration result.

3. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 2, characterized in that, The global basic registration model is constructed in the following way: Obtain the coordinate transformation rules, grid mapping rules, and time index rules corresponding to each monitoring data in the multi-source monitoring dataset. Based on the coordinate transformation rules, grid mapping rules, and time index rules, construct spatial transformation operators, grid merging operators, and time alignment operators, respectively. The time alignment operator, the spatial transformation operator, and the mesh merging operator are combined to obtain a composite operator; The composite operator is used as the global basic registration model.

4. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 2, characterized in that, The method involves calculating the credibility weight of each monitoring data point in different regions and at different times using a weighted approach, and then normalizing the credibility weight of each monitoring data point to output a normalized credibility weight. Specifically, this includes: Obtain the device accuracy factor, time freshness factor, environmental impact factor, historical stability factor, and neighborhood consistency factor for each monitoring data in the multi-source monitoring dataset; The reliability weight of each monitoring data in different regions and at different times is obtained by weighting and summing the equipment accuracy factor, the time freshness factor, the environmental impact factor, the historical stability factor, and the neighborhood consistency factor. The normalized confidence weight is obtained by dividing the confidence weight of each monitoring data in the multi-source monitoring dataset by the sum of the confidence weights of each monitoring data.

5. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 2, characterized in that, The step of performing abnormal event detection on the multi-source monitoring dataset in the unified engineering spatiotemporal reference framework based on the initial spatiotemporal registration result and the normalized confidence weight specifically includes: For each grid cell in the risk analysis grid, obtain the registered observation values ​​of each monitoring data falling into the current grid cell; Calculate the single-source anomaly indication value of each monitoring data based on the registered observation values ​​of each monitoring data in the multi-source monitoring dataset; The anomaly score of the current grid cell is obtained by weighting and summing the normalized confidence weight of each monitoring data with the corresponding single-source anomaly indication value. When the abnormal score exceeds a preset threshold, it is determined that there is an abnormal event in the current grid cell.

6. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 5, characterized in that, When an abnormal event is detected, generating a local reregistration region based on the location and time of the abnormal event specifically includes: Obtain the grid index of the grid cell where the abnormal event is located as the abnormal location, and obtain the time when the abnormal event is detected as the abnormal time. Using the grid cell corresponding to the abnormal location as the center, a predetermined number of grid cells are expanded outward in the plane to generate a spatial boundary; A time window is generated by extending a preset time period forward and backward from the aforementioned abnormal moment; The spatial boundary and the time window are used together as the local reregistration region.

7. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 2, characterized in that, Within the local reregistration region, the normalized confidence weights of each monitoring data point in the multi-source monitoring dataset are compared. Using the monitoring data with the largest weight as the spatial reference, local spatiotemporal reregistration is performed on the other monitoring data within the local reregistration region to obtain the corrected local registration relationship. Specifically, this includes: Within the local reregistration region, the monitoring data with the highest normalized confidence weight is selected as the local spatial reference. Using the coordinate system of the local spatial reference as the target, at least one of feature matching, iterative nearest point algorithm, or local affine transformation is used to perform local spatial reregistration on other monitoring data within the local reregistration area to obtain local spatial transformation parameters; Local time reregistration is performed on each monitoring data within the local reregistration area to align all monitoring data to a unified time axis, thereby obtaining local time alignment parameters; The local spatial transformation parameters and the obtained local temporal alignment parameters are used together as the corrected local registration relationship.

8. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 7, characterized in that, The step of writing the corrected local registration relationship into the global basic registration model to obtain the updated global basic registration model specifically includes: Obtain the subset of registration parameters corresponding to the local reregistration region in the global basic registration model; The corrected local registration relationship is used to replace the corresponding parameters in the registration parameter subset to obtain the updated global basic registration model.

9. The method for dynamic registration and fusion of multi-source data on water conservancy hazards according to claim 2, characterized in that, The process of performing trusted fusion calculations based on the normalized trustworthiness weights and the updated global basic registration model, and outputting fusion risk results, specifically includes: By using the updated global basic registration model, the re-registered observations of each monitoring data point are obtained on each grid cell and the unified time axis. The re-registered observations of each monitoring data point are converted into risk characterization values ​​through a risk mapping function. The normalized confidence weights of each monitoring data point are weighted and summed with the corresponding risk characterization values ​​to obtain the fused risk value. Calculate the first product of the normalized confidence weight of each monitoring data and the risk characterization value, calculate the sum of the first products of all monitoring data, divide the first product by the sum of the first products, and obtain the source contribution of each monitoring data to the fusion risk value. The sum of the products of the normalized confidence weights of each monitoring data and the neighborhood consistency factor is used as the confidence level of the result. The fusion risk value, the source contribution, and the result confidence level are output as the fusion risk result.

10. A dynamic registration and fusion system for multi-source data on water conservancy hazards, characterized in that, include: The data acquisition module is used to acquire and preprocess multi-source monitoring data to obtain a multi-source monitoring dataset; The initial registration module is used to map the multi-source monitoring dataset to a unified engineering spatiotemporal reference frame, and to perform initial spatiotemporal registration through a global basic registration model, and output the initial spatiotemporal registration result. The weight calculation module is used to calculate the credibility weight of each monitoring data and perform normalization processing, and output the normalized credibility weight. An anomaly detection module is used to detect anomalies based on the initial spatiotemporal registration result and the normalized confidence weight, and to generate a local re-registration region based on the detected anomalies. The local re-registration module is used to perform local spatiotemporal re-registration within the local re-registration region, using the monitoring data with the largest weight value as the spatial reference, to obtain the corrected local registration relationship and update the global basic registration model. The fusion output module is used to perform trusted fusion calculations based on the normalized trustworthiness weights and the updated global basic registration model, and output the fusion risk results.

Images