River and reservoir hydrological safety production management system and method

Abstract

The present application relates to river and reservoir hydrological safety production risk grading management and control and dynamic visualization monitoring technical field, especially relates to river and reservoir hydrological safety production management system and method, including spatial semantic analysis module utilizes river and reservoir hydrology geographic ontology library and converts unstructured dangerous source data into vector space relation tree;Probability field mapping module extracts absolute space reference anchor point, and the theoretical center coordinate is input direction adaptive probability distribution algorithm through direction adaptive probability distribution algorithm, and the output space probability density field;Risk quantitative evaluation module calculates the three-dimensional integral interference quantity of the space probability density field to the water conservancy fixed building model structure vulnerability field;Twin rendering early warning module converts the space probability density field into voxel cloud rendering, and triggers the alarm according to the preset threshold value.The present application solves the heterogeneous problem between the ubiquitous description of field dangerous source and the accurate reference of fixed building, and improves the risk fall map precision and the visual monitoring effectiveness.

Description

Technical Field

[0001] This invention relates to the field of risk classification and control and dynamic visualization monitoring of river and reservoir hydrological safety production, and particularly to a river and reservoir hydrological safety production management system and method. Background Technology

[0002] The risk classification, control, and monitoring of river and reservoir hydrological safety production constitutes a comprehensive management framework in water conservancy projects. Rivers and reservoirs refer to rivers and reservoirs, serving as core carriers for water resource regulation, flood control, drought relief, and hydropower generation. Hydrology involves the systematic observation and analysis of natural elements such as water level, flow, and precipitation, providing basic data support for river and reservoir operation. Safety production emphasizes reducing accidents through preventive measures and ensuring the stable operation of personnel, facilities, and the environment. Risk classification and control categorizes risks into different levels based on the severity and likelihood of potential hazards and adopts corresponding management strategies to optimize resource allocation. Monitoring utilizes monitoring equipment and technologies to continuously track the state and hydrological parameters of rivers and reservoirs to identify anomalies and support decision-making adjustments, ultimately forming a structured safety management process.

[0003] The existing technologies for risk classification and dynamic visualization monitoring of river and reservoir hydrological safety production have the following technical pain points: the spatial information description of field hazards generally has ubiquitous characteristics, and the positioning mostly relies on broad geographical orientation or unstructured textual descriptions, lacking high-precision absolute coordinate constraints. On the other hand, fixed building models are built based on precise spatial benchmarks, with strict geometric topological relationships and millimeter-level coordinate systems. The two types of data are heterogeneous in data structure and coordinate reference, which makes it impossible to accurately map non-fixed point risks in the field on a unified geographic information base map, thus causing spatial position offsets or overlaps in the dynamic visualization monitoring process. Taking hydrological field survey application scenarios as an example, when surveyors discover a landslide risk point on a riverbank, they can usually only make a spatial qualitative description based on the nearby river section number or the approximate bank location. This ubiquitous location information is difficult to directly map into a high-precision 3D model of the flood control dike built based on digital twin technology. As a result, the monitoring platform cannot accurately determine the real-time relative positional relationship between the landslide risk point and the dike engineering structure, and ultimately cannot accurately locate, plot, and associate the dynamically evolving risk in the visualization scenario. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a river and reservoir hydrological safety production management system and method. This invention solves the technical problem that the heterogeneity between the ubiquity of spatial information description of field hazards and the accuracy of spatial benchmarks of fixed building models makes it impossible to accurately map and dynamically monitor non-fixed location risks on a unified geographic information base map.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: In a first aspect, the river and reservoir hydrological safety production management system provided by the present invention includes: The spatial semantic parsing module is used to receive unstructured hazard source data including ubiquitous location descriptions, extract spatial entity words, distance measures, and water system features from the ubiquitous location descriptions using a river and reservoir hydrogeographic ontology, and construct a vector spatial relationship tree based on the spatial entity words, distance measures, and water system features. The probability field mapping module is used to receive the vector space relationship tree, extract the corresponding absolute spatial reference anchor point from the spatial entity words in the vector space relationship tree in the preset standard geographic information base map, calculate the theoretical center coordinates according to the absolute spatial reference anchor point, the distance measure and the water system feature, input the theoretical center coordinates into the direction adaptive probability distribution algorithm, and output the spatial probability density field. The risk quantification and assessment module is used to receive the spatial probability density field, obtain the hazard attributes bound to the unstructured hazard source data, input the hazard attributes into the risk assessment model to output a four-color status identifier, the four-color status identifier including a red status identifier, load the structural vulnerability field corresponding to the hydraulic fixed building model, perform three-dimensional multiplication calculation between the spatial probability density field and the structural vulnerability field, and output a three-dimensional integral interference quantity. The twin rendering early warning module is used to receive the four-color status identifier, the spatial probability density field, and the three-dimensional integral interference quantity, convert the spatial probability density field into a voxel cloud, map the four-color status identifier to the rendering parameters of the voxel cloud, output the voxel cloud with the rendering parameters in the three-dimensional visualization base map of security monitoring, obtain a preset security interference threshold, determine whether the three-dimensional integral interference quantity is greater than the preset security interference threshold, and trigger a layer highlighting alarm command when the three-dimensional integral interference quantity is greater than the preset security interference threshold.

[0006] Furthermore, in the river and reservoir hydrological safety production management system of the present invention, the spatial semantic parsing module includes: The entity extraction unit is used to receive the unstructured hazard source data, use the knowledge graph dictionary in the river and reservoir hydrogeographic ontology database, call the natural language processing model to perform word segmentation and sequence labeling on the unstructured hazard source data, extract the spatial entity words, the distance measure and the water system features, and combine the spatial entity words, the distance measure and the water system features to output a discrete data set. The topology constraint unit is used to receive the discrete data set, input the discrete data set into the river and reservoir hydrological geographic ontology database for verification, establish directed connection edges for the spatial entity words, distance measures and water system features in the discrete data set according to the river network and water system flow direction rules and water level fluctuation topology in the river and reservoir hydrological geographic ontology database, extract the water flow velocity and flow rate values ​​from the water system features and combine them into hydrological dynamic weights, assign the hydrological dynamic weights to the directed connection edges, and output the vector space relationship tree carrying the directed connection edges and the hydrological dynamic weights.

[0007] Furthermore, in the river and reservoir hydrological safety production management system of the present invention, the probability field mapping module includes: The reference anchoring unit is used to receive the vector space relationship tree, extract the spatial entity words in the vector space relationship tree, search for matching spatial entity words in the preset standard geographic information base map, and obtain the three-dimensional absolute coordinates corresponding to the spatial entity words as the absolute spatial reference anchor point. The affine transformation unit is used to receive the absolute space reference anchor point, take the absolute space reference anchor point as the origin, and establish a dynamic affine transformation matrix from the relative topological space to the absolute coordinate system according to the water system characteristics, the distance measurement and the hydrological dynamic weight in the vector space relationship tree. The distance measurement in the vector space relationship tree is converted into a relative vector, and the relative vector is substituted into the dynamic affine transformation matrix for rotation and translation calculation, and the theoretical center coordinates are output.

[0008] Furthermore, in the river and reservoir hydrological safety production management system of the present invention, the probability field mapping module further includes: The probabilistic dimensionality reduction unit receives the theoretical center coordinates, extracts digital elevation model data from the preset standard geographic information base map according to the theoretical center coordinates, calculates the terrain slope, river direction, and bank slope soil erosion tendency index based on the digital elevation model data, combines the terrain slope, river direction, and bank slope soil erosion tendency index into adaptive bias weights, sets the initial distribution boundary according to the distance metric, uses the theoretical center coordinates as the center origin, extracts the Gaussian kernel function corresponding to the directional adaptive probability distribution algorithm, and substitutes the adaptive bias weights and the initial distribution boundary into the Gaussian kernel function to perform Gaussian kernel deformation operation, outputting the spatial probability density field presenting a three-dimensional continuous grid distribution.

[0009] Furthermore, in the river and reservoir hydrological safety production management system of the present invention, the risk quantification assessment module includes: The quantization and coding unit is used to receive the hazard source attributes, extract the accident cause attributes and accident type attributes from the hazard source attributes, access the external meteorological interface to obtain meteorological and hydrological forecast data, extract the rainfall value and duration value from the meteorological and hydrological forecast data, perform logarithmic conversion to output the rainfall time decay coefficient, perform weighted summation of the accident cause attributes, the accident type attributes, and the rainfall time decay coefficient, input the weighted summation result into the risk assessment model for multiplication of the probability of occurrence and severity, perform calculation and comparison within a preset risk threshold range, output the risk level, and bind the corresponding four-color status identifier to the risk level.

[0010] Furthermore, in the river and reservoir hydrological safety production management system of the present invention, the risk quantification assessment module further includes: An interferometric calculation unit is used to receive the spatial probability density field, load the structural vulnerability field corresponding to the hydraulic fixed building model, calculate the angle between the diffusion normal vector of the spatial probability density field and the force tangent of the structural vulnerability field, use the cosine value of the angle between the force tangents as the angle collision coefficient, extract the probability density value of the spatial probability density field and the structural vulnerability value of the structural vulnerability field within the same three-dimensional coordinate system voxels, perform voxel-level three-dimensional multiplication and accumulation calculation on the probability density value, the structural vulnerability value and the angle collision coefficient and perform global summation integration, and output the three-dimensional integral interferometric quantity.

[0011] Furthermore, in the river and reservoir hydrological safety production management system of the present invention, the twin rendering early warning module includes: The gradient rendering unit is used to receive the four-color status identifier and the spatial probability density field, obtain the spatial occlusion perspective parameters of the hydraulic fixed building model in the 3D visualization base map of the safety monitoring, substitute the probability density value of the spatial probability density field into the 3D graphics rendering algorithm, multiply the probability density value by the reciprocal of the spatial occlusion perspective parameter, output the transparency gradient corresponding to the spatial occlusion perspective parameter, map the four-color status identifier and the transparency gradient to the transparency channel rendering parameters of the voxel cloud, use the transparency channel rendering parameters as the rendering parameters, transform the spatial probability density field into the voxel cloud, and output the voxel cloud with the rendering parameters in the 3D visualization base map of the safety monitoring.

[0012] Furthermore, in the river and reservoir hydrological safety production management system of the present invention, the twin rendering early warning module further includes: The threshold triggering unit is used to receive the three-dimensional integral interference quantity, obtain the fatigue reduction coefficient and preset initial safety threshold of the hydraulic fixed structure model, multiply the fatigue reduction coefficient and the preset initial safety threshold to calculate and output the preset safety interference threshold, determine whether the three-dimensional integral interference quantity is greater than the preset safety interference threshold, and trigger the layer red alarm command when the three-dimensional integral interference quantity is greater than the preset safety threshold, and output the closed-loop control command including the interference coordinates.

[0013] Furthermore, the river and reservoir hydrological safety production management system of the present invention also includes: The closed-loop scheduling module is used to receive the closed-loop control command, parse the closed-loop control command to extract the interference coordinates, send the interference coordinates to the UAV equipment to instruct the UAV equipment to perform three-dimensional laser scanning, receive the field point cloud data returned by the UAV equipment, extract the bounding box geometric volume of the field point cloud data as the true volume, extract the probability coverage boundary of the spatial probability density field to calculate the predicted volume, perform volume comparison calculation by intersecting the true volume and the predicted volume based on the overlap, output the cross-validation result including the volume overlap rate, and send the cross-validation result to the safety production scheduling terminal.

[0014] Secondly, the river and reservoir hydrological safety production management method provided by the present invention is applied to the aforementioned river and reservoir hydrological safety production management system, including: Step 1: The spatial semantic parsing module receives unstructured hazard source data including ubiquitous location descriptions, extracts spatial entity words, distance measures, and water system features from the ubiquitous location descriptions using the river and reservoir hydrogeographic ontology, and constructs a vector spatial relationship tree based on the spatial entity words, the distance measures, and the water system features. Step 2: The probability field mapping module receives the vector space relationship tree, extracts the corresponding absolute spatial reference anchor point from the spatial entity words in the vector space relationship tree in the preset standard geographic information base map, calculates the theoretical center coordinates according to the absolute spatial reference anchor point, the distance measure and the water system feature, inputs the theoretical center coordinates into the direction adaptive probability distribution algorithm, and outputs the spatial probability density field. Step 3: The risk quantification assessment module receives the spatial probability density field, obtains the hazard attributes bound to the unstructured hazard source data, inputs the hazard attributes into the risk assessment model to output a four-color status identifier, the four-color status identifier includes a red status identifier, loads the structural vulnerability field corresponding to the hydraulic fixed building model, performs a three-dimensional multiplication calculation between the spatial probability density field and the structural vulnerability field, and outputs a three-dimensional integral interference quantity. Step 4: The twin rendering early warning module receives the four-color status identifier, the spatial probability density field, and the three-dimensional integral interference quantity. It converts the spatial probability density field into a voxel cloud, maps the four-color status identifier to the rendering parameters of the voxel cloud, outputs the voxel cloud with the rendering parameters in the three-dimensional visualization base map of security monitoring, obtains a preset security interference threshold, determines whether the three-dimensional integral interference quantity is greater than the preset security interference threshold, and triggers a layer highlighting alarm command when the three-dimensional integral interference quantity is greater than the preset security interference threshold.

[0015] Beneficial effects of this invention: The river and reservoir hydrological safety production management system and method provided by this invention performs deep semantic mining on unstructured hazard source data, including ubiquitous location descriptions, through a spatial semantic parsing module. It extracts spatial entity words, distance measures, and water system features using a river and reservoir hydrological geographic ontology database and constructs a vector spatial relationship tree, effectively transforming colloquial field hazard source descriptions into structured topological relationships. This overcomes the technical obstacle of mapping field survey data due to the lack of absolute coordinates. The probability field mapping module extracts absolute spatial reference anchor points and constructs a dynamic affine transformation matrix. Using a direction-adaptive probability distribution algorithm, it maps uncertain textual descriptions into a three-dimensional continuous grid-distributed spatial probability density field. This not only compensates for hydrodynamic drift errors but also accurately simulates the asymmetric diffusion pattern of hazard sources constrained by topographic undulations and water erosion. The risk quantification and assessment module deeply couples hazard source attributes and meteorological and hydrological forecast data with a hydraulic fixed structure model loaded with a structural vulnerability field. Through voxel-level three-dimensional multiplication and accumulation calculation, it outputs a three-dimensional integral interference quantity, quantifying the structural interference strength between the dynamic risk field and the fixed structure entity. This elevates risk assessment beyond qualitative levels to quantitative analysis based on three-dimensional collision attributes. The twin rendering and early warning module, combined with a voxel cloud rendering scheme generated from spatial occlusion perspective parameters, solves the problem of physical occlusion of risk lines by hydraulic structures. Coupled with a preset safety interference threshold corrected in real-time based on fatigue reduction coefficients, it forms a full-link monitoring system covering identification, mapping, assessment, and closed-loop scheduling. This significantly improves the positioning accuracy and effectiveness of associated early warnings of dynamic river and reservoir hydrological risks within the digital twin framework. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the method for managing hydrological safety in rivers and reservoirs according to the present invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0019] In a first aspect, the river and reservoir hydrological safety production management system provided by the present invention includes: The spatial semantic parsing module is used to receive unstructured hazard source data including ubiquitous location descriptions, extract spatial entity words, distance measures, and water system features from the ubiquitous location descriptions using a river and reservoir hydrogeographic ontology, and construct a vector spatial relationship tree based on the spatial entity words, distance measures, and water system features. The probability field mapping module is used to receive the vector space relationship tree, extract the corresponding absolute spatial reference anchor point from the spatial entity words in the vector space relationship tree in the preset standard geographic information base map, calculate the theoretical center coordinates according to the absolute spatial reference anchor point, the distance measure and the water system feature, input the theoretical center coordinates into the direction adaptive probability distribution algorithm, and output the spatial probability density field. The risk quantification and assessment module is used to receive the spatial probability density field, obtain the hazard attributes bound to the unstructured hazard source data, input the hazard attributes into the risk assessment model to output a four-color status identifier, the four-color status identifier including a red status identifier, load the structural vulnerability field corresponding to the hydraulic fixed building model, perform three-dimensional multiplication calculation between the spatial probability density field and the structural vulnerability field, and output a three-dimensional integral interference quantity. The twin rendering early warning module is used to receive the four-color status identifier, the spatial probability density field, and the three-dimensional integral interference quantity, convert the spatial probability density field into a voxel cloud, map the four-color status identifier to the rendering parameters of the voxel cloud, output the voxel cloud with the rendering parameters in the three-dimensional visualization base map of security monitoring, obtain a preset security interference threshold, determine whether the three-dimensional integral interference quantity is greater than the preset security interference threshold, and trigger a layer highlighting alarm command when the three-dimensional integral interference quantity is greater than the preset security interference threshold.

[0020] Field surveys in river and reservoir hydrological environments generate a large amount of colloquial hazard reporting text. Surveyors typically rely on manual visual observation and reference to surrounding landmarks for textual recording. The spatial semantic analysis module receives unstructured hazard data, including ubiquitous location descriptions. Since ubiquitous location descriptions generally lack absolute coordinate references, the system calls upon a river and reservoir hydrological geographic ontology database to perform semantic segmentation and part-of-speech tagging on these descriptions. This database embeds a knowledge graph dictionary covering hydrological station names, levee station numbers, and hydraulic structure names. The spatial semantic analysis module combines this knowledge graph dictionary to extract spatial entity words, distance measures, and river system features. Spatial entity words represent objectively existing fixed features, while river system features include the flow velocity and direction of river sections. The system uses spatial entity words as root nodes and transforms distance measures and river system features into directed edges with direction and length attributes. Based on these directed edges and the root node, the spatial semantic analysis module constructs a vector spatial relationship tree, establishing a data structure with spatial hierarchical relationships and flow direction topological constraints.

[0021] After generating the vector space relationship tree, the probability field mapping module receives the vector space relationship tree and retrieves matching spatial entity words on a pre-set standard geographic information base map. The standard geographic information base map covers a high-resolution three-dimensional coordinate grid. The system extracts the absolute spatial reference anchor points corresponding to the spatial entity words, which serve as the central origin for subsequent coordinate system transformations. Conventional calculation models use simple buffer circles to define the risk range. Due to the undulating terrain of the river channel and the scouring effect of water flow, the diffusion of hazard sources usually exhibits an asymmetrical pattern. The probability field mapping module sets the initial distribution boundary according to distance measurements. The terrain slope and river direction covered by the water system features are converted into adaptive bias weights by the system. The probability field mapping module synchronously inputs the central origin, the initial distribution boundary, and the adaptive bias weights into the direction adaptive probability distribution algorithm. The direction adaptive probability distribution algorithm performs Gaussian kernel deformation operations, stretching the covariance matrix of the Gaussian distribution outward from the central origin along the water flow direction and downhill direction, outputting an irregular spatial probability density field that conforms to the terrain trend.

[0022] The risk quantification assessment module receives the spatial probability density field and simultaneously acquires the hazard source attributes bound to the unstructured hazard source data. Hazard source attributes include accident cause attributes and accident type attributes. The risk quantification assessment module inputs the hazard source attributes into the risk assessment model. The risk assessment model embeds an accident occurrence probability scoring matrix and a severity bias parameter. The system performs a multiplication calculation of the occurrence probability score matrix and the severity bias parameter, outputting a four-color status indicator including four levels: major, significant, moderate, and low. To calculate the degree of collision between dynamically evolving risks and fixed hydraulic engineering spatial entities, the risk quantification assessment module loads the structural vulnerability field corresponding to the fixed hydraulic building model. The structural vulnerability field pre-assigns values ​​to the building components in a three-dimensional mesh based on their material resistance to damage. The system imports the spatial probability density field and the structural vulnerability field into the same three-dimensional coordinate system, performs voxel-level three-dimensional multiplication and accumulation calculations on the probability density values ​​and structural vulnerability values ​​within the same coordinate voxel, and outputs a three-dimensional integral interferometry quantity with spatial entity collision intensity.

[0023] The twin rendering early warning module receives four-color status indicators, a spatial probability density field, and a three-dimensional integral interference quantity. It transforms the spatial probability density field into a voxel cloud. The voxel cloud is formed by assembling a tiny three-dimensional cube array with independent coordinates and material properties. The system maps the four-color status indicators to rendering parameters for the voxel cloud. These parameters include color ratio and transparency channel values. High-risk areas at the center are assigned lower transparency channel values ​​to present a rich color, while low-risk areas at the edges have gradually increased transparency channel values ​​to create a blurred transparency effect. The system outputs a voxel cloud with rendering parameters on the 3D visualization base map for safety monitoring. Simultaneously, the twin rendering early warning module acquires a preset safety interference threshold. This threshold represents the maximum structural safety limit that the hydraulic facility can withstand. The system continuously checks whether the three-dimensional integral interference quantity exceeds the preset safety interference threshold, and if it does, triggers a layer-highlighting alarm command.

[0024] Field surveys in river and reservoir hydrological environments generate a large amount of colloquial hazard reporting text. Survey personnel typically rely on manual visual observation and reference to surrounding landmarks for textual recording. The spatial semantic analysis module receives unstructured hazard data including ubiquitous location descriptions, and simultaneously acquires the accident cause attributes and accident type attributes recorded in the unstructured hazard data, collectively referred to as hazard attributes. Since ubiquitous location descriptions generally lack absolute coordinate references, the system calls upon a river and reservoir hydrological geographic ontology database to perform semantic segmentation and part-of-speech tagging on the ubiquitous location descriptions. The river and reservoir hydrological geographic ontology database embeds a knowledge graph dictionary covering hydrological station names, levee station numbers, and hydraulic structure names. The spatial semantic analysis module combines the knowledge graph dictionary to extract spatial entity words, distance measurements, and water system features. Spatial entity words represent objectively existing fixed features, and water system features include the flow velocity and direction attributes of river sections. The system uses spatial entity words as root nodes and transforms distance measurements and water system features into directed edges with direction and length attributes. The spatial semantic parsing module constructs a vector space relation tree based on directed connection edges and root nodes, establishing a data structure with spatial hierarchical relationships and flow topological constraints.

[0025] After generating the vector space relationship tree, the probability field mapping module receives the vector space relationship tree and retrieves matching spatial entity words on a pre-set standard geographic information base map. The standard geographic information base map covers a high-resolution three-dimensional coordinate grid. The system extracts the absolute spatial reference anchor points corresponding to the spatial entity words, which serve as the central origin for subsequent coordinate system transformations. Conventional calculation models use simple buffer circles to define the risk range. Due to the undulating terrain of the river channel and the scouring effect of water flow, the diffusion of hazard sources usually exhibits an asymmetrical pattern. The probability field mapping module sets the initial distribution boundary according to distance measurements. The terrain slope and river direction covered by the water system features are converted into adaptive bias weights by the system. The probability field mapping module synchronously inputs the central origin, the initial distribution boundary, and the adaptive bias weights into the direction adaptive probability distribution algorithm. The direction adaptive probability distribution algorithm performs Gaussian kernel deformation operations, stretching the covariance matrix of the Gaussian distribution outward from the central origin along the water flow direction and downhill direction, outputting an irregular spatial probability density field that conforms to the terrain trend.

[0026] The risk quantification assessment module receives the spatial probability density field and simultaneously acquires the hazard source attributes. The module inputs these attributes into the risk assessment model. The risk assessment model embeds an accident occurrence probability scoring matrix and a severity bias parameter. The system performs a multiplication of the occurrence probability score matrix and the severity bias parameter, outputting a four-color status indicator with four levels: major, significant, moderate, and low, corresponding to red, orange, yellow, and blue, respectively. To calculate the degree of collision between dynamically evolving risks and the spatial entities of a fixed hydraulic engineering project, the risk quantification assessment module loads the structural vulnerability field corresponding to the fixed hydraulic building model. The structural vulnerability field pre-assigns values ​​to the building components using a three-dimensional mesh based on their material resistance to damage. The system imports the spatial probability density field and the structural vulnerability field into the same three-dimensional coordinate system, performs voxel-level three-dimensional multiplication and accumulation calculations on the probability density values ​​and structural vulnerability values ​​within the same coordinate voxel, and outputs a three-dimensional integral interferometer with spatial entity collision intensity.

[0027] The twin rendering early warning module receives four-color status indicators, a spatial probability density field, and a three-dimensional integral interference quantity. It transforms the spatial probability density field into a voxel cloud. The voxel cloud is formed by assembling a tiny three-dimensional cube array with independent coordinates and material properties. The system maps the four-color status indicators to rendering parameters for the voxel cloud. These parameters include color ratio and transparency channel values. High-risk areas at the center are assigned lower transparency channel values ​​to present an opaque, deep color, while low-risk areas at the edges have gradually increased transparency channel values ​​to present a blurred, transparent effect. The system outputs a voxel cloud with rendering parameters on the 3D visualization base map for safety monitoring. Simultaneously, the twin rendering early warning module acquires a preset safety interference threshold. This threshold represents the maximum structural safety limit that the hydraulic facility can withstand. The system continuously checks whether the three-dimensional integral interference quantity exceeds the preset safety interference threshold, and if it does, triggers a layer-highlighting alarm command.

[0028] Field hydrological surveys are conducted in complex and ever-changing natural environments, resulting in textual information reported by surveyors containing a large amount of non-standardized geographical descriptive features. The entity extraction unit directly connects to the river and reservoir hydrological geographic ontology database after receiving unstructured hazard source data. To handle these non-standardized geographical descriptive features, the entity extraction unit utilizes the knowledge graph dictionary within the river and reservoir hydrological geographic ontology database and calls a natural language processing model to perform word segmentation and sequence labeling operations on the unstructured hazard source data. The knowledge graph dictionary internalizes a massive amount of hydrological terminology and a mapping table of historical survey colloquialisms, enabling the natural language processing model to extract spatial entity words, distance measures, and river system features from the text through sequence labeling. Subsequently, the entity extraction unit combines and encapsulates the spatial entity words, distance measures, and river system features, outputting a discrete dataset. The topology constraint unit receives the discrete dataset, merges it, and inputs the discrete dataset into the river and reservoir hydrological geographic ontology database for validity verification. In the verification phase, the topology constraint unit establishes directed connections reflecting spatial logical relationships for spatial entities, distance measures, and river system features in the discrete dataset, based on the fixed river network flow direction rules and water level fluctuation topology relationships in the river and reservoir hydrogeographic ontology. Considering the driving and exacerbating effect of water flow on the evolution of hazard sources, the topology constraint unit extracts the recorded water flow velocity and discharge values ​​from the river system features, combines them to generate hydrological dynamic weights, and directly assigns these hydrological dynamic weights to the directed connections, outputting a vector space relation tree carrying topological structure information.

[0029] The vector space relation tree transforms textual descriptions into computer-readable spatial logical graphs, enabling the reference anchoring unit to receive the vector space relation tree and perform spatial coordinate system anchoring. The reference anchoring unit extracts spatial entity words from the vector space relation tree, retrieves the corresponding three-dimensional absolute coordinates of the spatial entity words in a pre-set standard geographic information base map, and sets these three-dimensional absolute coordinates as absolute spatial reference anchor points to provide an origin reference for subsequent coordinate transformations. The affine transformation unit receives the absolute spatial reference anchor points and uses them as the origin of the spatial transformation. Based on the water system characteristics, distance measurements, and hydrodynamic weights carried in the vector space relation tree, it constructs a dynamic affine transformation matrix that dynamically reflects the displacement trend of water flow scouring. To perform coordinate transformation, the affine transformation unit converts the distance measurements in the vector space relation tree into relative vectors with directional attributes and substitutes these relative vectors into the dynamic affine transformation matrix to perform rotation and translation calculations. By performing rotation and translation calculations to compensate for spatial drift errors caused by hydrodynamics, the affine transformation unit outputs the theoretical center coordinates of the hazard source on the standard geographic information base map.

[0030] The spread of hydrological hazards is often severely constrained by topographic relief. Therefore, the probabilistic dimensionality reduction unit receives the theoretical center coordinates and extracts digital elevation model (DEM) data around the theoretical center coordinates from a standard geographic information base map. Based on the DEM data, the probabilistic dimensionality reduction unit calculates the topographic slope, river course, and bank soil erosion tendency index for the current area. Specifically, topographic slope determines the tendency for gravity landslides, river course restricts the boundary of water erosion, and the bank soil erosion tendency index reflects the fragility of the geological structure. The probabilistic dimensionality reduction unit combines the topographic slope, river course, and bank soil erosion tendency index into adaptive bias weights, and simultaneously sets an initial distribution boundary around the theoretical center coordinates according to distance measurements. The probabilistic dimensionality reduction unit uses the theoretical center coordinates as the origin of the Gaussian kernel function and substitutes the adaptive bias weights and the initial distribution boundary into the Gaussian kernel function of the directional adaptive probability distribution algorithm to perform Gaussian kernel deformation operations. Guided by the adaptive bias weights, the Gaussian kernel function undergoes asymmetric stretching deformation towards areas with steep slopes and severe erosion tendencies, outputting a spatial probability density field with a three-dimensional continuous grid distribution.

[0031] The evolution of risk is closely related to meteorological conditions. The quantification and coding unit receives hazard source attributes and specifically extracts accident trigger attributes and accident type attributes from these attributes. To integrate meteorological data, the quantification and coding unit accesses an external meteorological interface to obtain meteorological and hydrological forecast data. This forecast data is acquired periodically at high frequency through an external meteorological API interface. Based on the reporting time marker and absolute spatial reference anchor point of the unstructured hazard source data, spatiotemporal data alignment processing is performed. The unit also extracts the rainfall and duration values ​​recorded in the meteorological and hydrological forecast data and performs logarithmic conversion to output a rainfall time decay coefficient that reflects the cumulative destructive effect of rainfall. The quantification and coding unit inputs the accident trigger attributes, accident type attributes, and rainfall time decay coefficient into a weighted algorithm model for weighted summation calculation. The weighted summation result is then input into a risk assessment model for multiplication of probability and severity. The risk assessment model compares the multiplication result within a preset risk threshold range, outputs a matching risk level, and the quantification and coding unit assigns a corresponding four-color status identifier to each risk level.

[0032] To objectively assess the structural collision threat posed by hazardous sources to hydraulic facilities, the interferometric computation unit receives the spatial probability density field and simultaneously loads the structural vulnerability field corresponding to the fixed hydraulic structure model. The destructive force of the hazardous source depends on the distance and impact direction, prompting the interferometric computation unit to calculate the angle between the diffusion normal vector of the spatial probability density field and the force-bearing tangent of the structural vulnerability field. The interferometric computation unit extracts the cosine of the force-bearing tangent angle as the angle collision coefficient, ensuring that a perpendicular impact yields the largest angle collision coefficient while a parallel glide yields the smallest. The interferometric computation unit extracts the probability density values ​​of the spatial probability density field and the structural vulnerability values ​​of the structural vulnerability field within the same three-dimensional coordinate system voxels. It then incorporates the probability density values, structural vulnerability values, and angle collision coefficient into the mathematical model for voxel-level three-dimensional multiplication and accumulation, followed by global summation and integration, outputting a three-dimensional integral interferometric quantity that quantifies the collision intensity of spatial entities.

[0033] In the construction phase of the digital twin scenario, the gradient rendering unit receives four-color status indicators and a spatial probability density field. To prevent hazard sources from being obscured by hydraulic structures due to line-of-sight occlusion in 3D space, the gradient rendering unit obtains the spatial occlusion perspective parameters of the hydraulic fixed structure model on the safety monitoring 3D visualization base map. The gradient rendering unit substitutes the probability density value of the spatial probability density field into the 3D graphics rendering algorithm, multiplies the probability density value by the reciprocal of the spatial occlusion perspective parameter, and dynamically amplifies the visual penetration of severely occluded areas through the reciprocal operation, outputting the transparency gradient corresponding to the spatial occlusion perspective parameter. The gradient rendering unit maps the four-color status indicators and transparency gradient to the transparency channel rendering parameters of the voxel cloud, uses the transparency channel rendering parameters as rendering parameters, transforms the spatial probability density field into a volumetric voxel cloud, and outputs the voxel cloud with rendering parameters on the safety monitoring 3D visualization base map.

[0034] The resilience of water conservancy facilities degrades over time. The threshold triggering unit receives the 3D integral interferometry and obtains the fatigue reduction coefficient of the water conservancy fixed structure model over time, as well as a preset initial safety threshold. The threshold triggering unit multiplies the fatigue reduction coefficient by the preset initial safety threshold and outputs a real-time updated preset safety interference threshold, thereby determining whether the 3D integral interferometry exceeds the preset safety interference threshold. When the 3D integral interferometry exceeds the preset safety interference threshold, the threshold triggering unit triggers a layer-highlighting alarm and outputs a closed-loop control command including the interference coordinates. The closed-loop scheduling module receives the closed-loop control command, parses it to extract the interference coordinates, encapsulates the interference coordinates, and sends them to the UAV equipment to control the UAV equipment to fly to the target airspace for 3D laser scanning. The closed-loop scheduling module receives the on-site point cloud data returned by the UAV equipment, extracts the bounding box geometric volume from the on-site point cloud data as the true volume, and simultaneously extracts the probability coverage boundary of the spatial probability density field to calculate the predicted volume. The closed-loop scheduling module puts the actual volume and the predicted volume into the volume comparison algorithm to calculate the overlap and intersection, outputs the cross-validation results including the volume overlap rate, and sends the cross-validation results to the safety production scheduling terminal.

[0035] Secondly, please refer to Figure 1 The river and reservoir hydrological safety production management method provided by the present invention is applied to the aforementioned river and reservoir hydrological safety production management system, comprising: Step 1: The spatial semantic parsing module receives unstructured hazard source data including ubiquitous location descriptions, extracts spatial entity words, distance measures, and water system features from the ubiquitous location descriptions using the river and reservoir hydrogeographic ontology, and constructs a vector spatial relationship tree based on the spatial entity words, the distance measures, and the water system features. Step 2: The probability field mapping module receives the vector space relationship tree, extracts the corresponding absolute spatial reference anchor point from the spatial entity words in the vector space relationship tree in the preset standard geographic information base map, calculates the theoretical center coordinates according to the absolute spatial reference anchor point, the distance measure and the water system feature, inputs the theoretical center coordinates into the direction adaptive probability distribution algorithm, and outputs the spatial probability density field. Step 3: The risk quantification assessment module receives the spatial probability density field, obtains the hazard attributes bound to the unstructured hazard source data, inputs the hazard attributes into the risk assessment model to output a four-color status identifier, the four-color status identifier includes a red status identifier, loads the structural vulnerability field corresponding to the hydraulic fixed building model, performs a three-dimensional multiplication calculation between the spatial probability density field and the structural vulnerability field, and outputs a three-dimensional integral interference quantity. Step 4: The twin rendering early warning module receives the four-color status identifier, the spatial probability density field, and the three-dimensional integral interference quantity. It converts the spatial probability density field into a voxel cloud, maps the four-color status identifier to the rendering parameters of the voxel cloud, outputs the voxel cloud with the rendering parameters in the three-dimensional visualization base map of security monitoring, obtains a preset security interference threshold, determines whether the three-dimensional integral interference quantity is greater than the preset security interference threshold, and triggers a layer highlighting alarm command when the three-dimensional integral interference quantity is greater than the preset security interference threshold.

[0036] Field hydrological surveys are conducted in complex and ever-changing natural environments. The textual information reported by surveyors often contains a large number of non-standardized geographical descriptive features, leading to ubiquitous obstacles in the spatial location of hazard sources. The spatial semantic parsing module, after receiving unstructured hazard source data, directly connects to the river and reservoir hydrological geographic ontology database. Utilizing the knowledge graph dictionary within this database, it calls a natural language processing (NLP) model to perform word segmentation and sequence labeling on the unstructured hazard source data. When the entity extraction unit calls the NLP model to perform word segmentation and sequence labeling, the NLP model specifically employs a bidirectional long short-term memory network combined with a conditional random field (BiLSTM-CRF) architecture. In the feature extraction stage, the embedding layer of the NLP model receives the knowledge graph dictionary as a prior domain knowledge base for feature initialization, mapping hydrological terminology to a high-dimensional vector space. By extracting bidirectional semantic features from the text through a bidirectional long short-term memory network layer, the problem of long-distance semantic dependence in the description of hydrological colloquialisms can be effectively solved. Subsequently, the output sequence is decoded using a conditional random field layer to achieve the global optimal path. Combined with the labeling rules of hydrological and geographical entities, the boundary ambiguity and ambiguity in non-standard place names such as "Lao Wang Jia Zha Kou" can be resolved, thereby accurately identifying and segmenting spatial entity words, distance measures, and water system features.

[0037] The knowledge graph dictionary internalizes a massive amount of hydrological terminology and a mapping table of historical survey slang. Through sequence labeling, it can accurately segment spatial entities, distance measures, and water system features within the text. Spatial entities represent objectively existing fixed features, distance measures reflect relative displacement vectors, and water system features cover the river's flow velocity and direction attributes. The spatial semantic parsing module combines and encapsulates spatial entities, distance measures, and water system features, outputting a discrete dataset. The topology constraint unit receives the discrete dataset and merges it into the river and reservoir hydrological geographic ontology database for validity verification. During the verification process, the topology constraint unit establishes directed connections reflecting spatial logical relationships for spatial entities, distance measures, and water system features in the discrete dataset, based on the fixed river network flow direction rules and water level fluctuation topology relationships in the river and reservoir hydrological geographic ontology database. Considering the driving and exacerbating effect of water flow on the evolution of hazard sources, the topological constraint unit specifically extracts the water flow velocity and flow rate values ​​recorded in the water system features, combines them to generate hydrological dynamic weights, and assigns the hydrological dynamic weights to the directed connecting edges, outputting a vector space relation tree carrying topological structure information.

[0038] The vector space relation tree transforms ambiguous textual descriptions into computer-readable spatial logical graphs, enabling the probability field mapping module to receive the vector space relation tree and perform spatial coordinate system anchoring. The reference anchoring unit extracts spatial entity words from the vector space relation tree, retrieves the corresponding three-dimensional absolute coordinates of the spatial entity words in a pre-set standard geographic information base map, and sets these three-dimensional absolute coordinates as the absolute spatial reference anchor point, providing a reference origin for subsequent coordinate transformations. The affine transformation unit receives the absolute spatial reference anchor point and uses it as the spatial transformation center. Based on the water system characteristics, distance measurements, and hydrodynamic weights carried in the vector space relation tree, it constructs a dynamic affine transformation matrix that dynamically reflects the displacement trend of water flow scouring. To perform spatial positioning, the affine transformation unit converts the distance measurements in the vector space relation tree into relative vectors with directional attributes and substitutes these relative vectors into the dynamic affine transformation matrix to perform rotation and translation calculations. By compensating for spatial drift errors caused by hydrodynamics through rotation and translation calculations, the affine transformation unit outputs the theoretical center coordinates of the hazard source on the standard geographic information base map.

[0039] The spread of hydrological hazards is typically constrained by topographic relief. Therefore, the probabilistic dimensionality reduction unit receives the theoretical center coordinates and extracts digital elevation model (DEM) data around these coordinates from a standard geographic information base map. Based on the DEM data, the unit calculates the topographic slope, river course, and bank erosion tendency index for the current area. Topographic slope determines the tendency for gravity landslides, river course limits the boundary of water erosion, and the bank erosion tendency index reflects the fragility of the geological structure. The unit combines the topographic slope, river course, and bank erosion tendency index into adaptive bias weights and sets initial distribution boundaries around the theoretical center coordinates according to distance measurements. Using the theoretical center coordinates as the origin of the Gaussian kernel function, the unit substitutes the adaptive bias weights and initial distribution boundaries into the Gaussian kernel function of the directional adaptive probability distribution algorithm to perform Gaussian kernel deformation. Guided by the adaptive bias weights, the Gaussian kernel function undergoes asymmetric stretching deformation towards areas with steep slopes and severe erosion, outputting a spatial probability density field with a three-dimensional continuous grid distribution.

[0040] The evolution of risk is closely related to meteorological conditions. The risk quantification and assessment module receives the spatial probability density field and acquires the hazard source attributes bound to unstructured hazard source data. Hazard source attributes include accident trigger attributes and accident type attributes. The quantification and coding unit receives the hazard source attributes and specifically extracts accident trigger attributes and accident type attributes, then connects to an external meteorological interface to obtain meteorological and hydrological forecast data. The quantification and coding unit extracts the rainfall and duration values ​​recorded in the meteorological and hydrological forecast data, performs a logarithmic conversion operation, and outputs a rainfall time decay coefficient that reflects the cumulative destructive effect of rainfall. The quantification and coding unit inputs the accident trigger attributes, accident type attributes, and rainfall time decay coefficient into a weighted algorithm model for weighted summation calculation. The weighted summation result is then input into the risk assessment model for multiplication of probability and severity. The risk assessment model compares the multiplication result within a preset risk threshold range, outputs the matching risk level, and the quantification and coding unit assigns a corresponding four-color status identifier to the risk level.

[0041] To objectively assess the structural collision threat posed by hazardous sources to water conservancy facilities, the risk quantification assessment module loads the structural vulnerability field corresponding to the fixed structure model of the water conservancy facility. The destructive force of the hazardous source depends on the distance and impact direction, prompting the interferometric calculation unit to calculate the angle between the diffusion normal vector of the spatial probability density field and the force-bearing tangent of the structural vulnerability field. The interferometric calculation unit extracts the cosine value of the force-bearing tangent angle as the angle collision coefficient; a perpendicular impact yields a larger angle collision coefficient, while a parallel glide yields a smaller one. The interferometric calculation unit extracts the probability density values ​​of the spatial probability density field and the structural vulnerability values ​​of the structural vulnerability field within the same three-dimensional coordinate system voxels. It then incorporates the probability density values, structural vulnerability values, and angle collision coefficient into the mathematical model for voxel-level three-dimensional multiplication and accumulation calculations, followed by global summation and integration, outputting a three-dimensional integral interferometric quantity that quantifies the physical threat intensity.

[0042] In the construction phase of the digital twin scenario, the twin rendering early warning module receives four-color status indicators, a spatial probability density field, and three-dimensional integral interference. To prevent hazard sources from being obscured by hydraulic structures due to line-of-sight occlusion in three-dimensional space, the gradient rendering unit obtains the spatial occlusion perspective parameters of the hydraulic fixed structure model on the safety monitoring 3D visualization base map. The gradient rendering unit substitutes the probability density value of the spatial probability density field into the 3D graphics rendering algorithm, multiplies the probability density value by the reciprocal of the spatial occlusion perspective parameter, and dynamically amplifies the visual penetration of severely occluded areas through the reciprocal operation, outputting a transparency gradient corresponding to the spatial occlusion perspective parameter. The gradient rendering unit maps the four-color status indicators and transparency gradient to the rendering parameters of the voxel cloud. The voxel cloud is formed by assembling a tiny three-dimensional cube array with independent coordinates and material properties. The rendering parameters include color ratio and transparency channel values. The central high-risk area is given a lower transparency channel value to present an opaque, deep color, while the transparency channel value of the edge low-risk area gradually increases to present a blurred transparency effect. The twin rendering early warning module outputs a voxel cloud with rendering parameters on the safety monitoring 3D visualization base map.

[0043] The resilience of water conservancy facilities degrades over time. The threshold triggering unit receives the 3D integral interferometry and obtains the fatigue reduction coefficient of the water conservancy fixed structure model over time, as well as a preset initial safety threshold. The threshold triggering unit multiplies the fatigue reduction coefficient by the preset initial safety threshold and outputs a real-time updated preset safety interference threshold, thereby determining whether the 3D integral interferometry exceeds the preset safety interference threshold. When the 3D integral interferometry exceeds the preset safety interference threshold, the threshold triggering unit triggers a layer-highlighting alarm and outputs a closed-loop control command including the interference coordinates. The closed-loop scheduling module receives the closed-loop control command, parses it to extract the interference coordinates, packages the interference coordinates, and sends them to the UAV to drive the UAV to the target airspace for 3D laser scanning. The closed-loop scheduling module receives the on-site point cloud data returned by the UAV, extracts the bounding box geometric volume from the on-site point cloud data as the true volume, and simultaneously extracts the probability coverage boundary of the spatial probability density field to calculate the predicted volume. The closed-loop scheduling module puts the actual volume and the predicted volume into the volume comparison algorithm to calculate the overlap and intersection, outputs the cross-validation results including the volume overlap rate, and sends the cross-validation results to the safety production scheduling terminal.

[0044] Field hydrological surveys face exceptionally complex geological and hydrological environments, leading surveyors to often rely on visual observation and surrounding landmarks to report descriptive information when discovering landslides or piping along riverbanks. This results in unstructured hazard data, including ubiquitous location descriptions. The spatial semantic parsing module, upon receiving this unstructured hazard data, invokes a natural language processing model to segment and label the data. To parse non-standardized geographic descriptive features, the entity extraction unit utilizes a knowledge graph dictionary from the river and reservoir hydrological geographic ontology database to assist the natural language processing model in labeling. This extracts spatial entity words, distance measures, and river system features from the unstructured hazard data and combines them to output a discrete dataset. The topology constraint unit receives the discrete dataset and inputs it into the river and reservoir hydrological geographic ontology database for validation. Based on the river network flow direction rules and water level fluctuation topology in the database, it establishes directed connections for spatial entity words, distance measures, and river system features in the discrete dataset. Considering that water erosion can accelerate the evolution of risks, the topological constraint unit specifically extracts the water flow velocity and flow rate values ​​from the water system features and combines them into hydrodynamic weights. These hydrodynamic weights are then assigned to directed connecting edges to output a vector space relation tree carrying topological structure information.

[0045] After the vector space relation tree is constructed, the probability field mapping module immediately receives the vector space relation tree and inputs it into the coordinate transformation process. The benchmark anchoring unit extracts spatial entity words from the vector space relation tree, searches for matching spatial entity words in the preset standard geographic information base map, and then obtains the three-dimensional absolute coordinates of the spatial entity words as absolute spatial benchmark anchor points. The affine transformation unit receives the absolute spatial benchmark anchor points and uses them as the origin to establish a dynamic affine transformation matrix according to the water system characteristics, distance measures, and hydrodynamic weights in the vector space relation tree. The affine transformation unit converts the distance measures in the vector space relation tree into relative vectors, substitutes the relative vectors into the dynamic affine transformation matrix for rotation and translation calculations, and finally outputs the theoretical center coordinates of the hazard source on the standard geographic information base map. The probability dimensionality reduction unit receives the theoretical center coordinates, extracts digital elevation model data from the standard geographic information base map according to the theoretical center coordinates, and calculates the terrain slope, river channel direction, and bank slope soil erosion tendency index based on the digital elevation model data. The probabilistic dimensionality reduction unit combines terrain slope, river channel orientation, and bank soil erosion tendency index into adaptive bias weights, while setting the initial distribution boundary according to distance metrics. Using the theoretical center coordinates as the origin, the probabilistic dimensionality reduction unit substitutes the adaptive bias weights and the initial distribution boundary into the Gaussian kernel function of the direction-adaptive probability distribution algorithm for Gaussian kernel deformation operations, resulting in a spatial probability density field with a three-dimensional continuous grid distribution as the output of the Gaussian kernel function.

[0046] Given that the evolution of risk status is influenced by multiple environmental factors, the risk quantification assessment module simultaneously acquires the hazard source attributes bound to unstructured hazard source data while receiving the spatial probability density field. The quantification coding unit receives the hazard source attributes, extracts the accident trigger attributes and accident type attributes from them, and combines them with the acquired meteorological and hydrological forecast data to extract rainfall and duration values, performing logarithmic conversion to output a rainfall time decay coefficient. The quantification coding unit performs a weighted summation of the accident trigger attributes, accident type attributes, and rainfall time decay coefficient, inputting the weighted summation result into the risk assessment model for multiplication of probability and severity. The quantification coding unit performs calculations and comparisons within a preset risk threshold range to output the risk level and assigns a corresponding four-color status identifier to the risk level. To measure the degree of physical damage, the interference calculation unit receives the spatial probability density field and loads the structural vulnerability field corresponding to the hydraulic fixed building model, calculating the angle between the diffusion normal vector of the spatial probability density field and the force tangent of the structural vulnerability field, using the cosine of the force tangent angle as the angle collision coefficient. The interferometric calculation unit extracts the probability density value of the spatial probability density field and the structural vulnerability value of the structural vulnerability field within the same three-dimensional coordinate system voxels. It then performs voxel-level three-dimensional multiplication and accumulation calculations on the probability density value, structural vulnerability value, and angular collision coefficient, and performs global summation and integration to output the three-dimensional integral interferometric quantity that quantifies the physical threat.

[0047] Entering the twin rendering and closed-loop early warning stage, the twin rendering early warning module receives four-color status indicators, a spatial probability density field, and a three-dimensional integral interference quantity. The gradient rendering unit receives the four-color status indicators and the spatial probability density field, and obtains the spatial occlusion perspective parameters of the hydraulic fixed building model on the safety monitoring 3D visualization base map for the observation view. Since the spatial occlusion perspective parameters reflect the degree of obstruction of the view by the hydraulic structure, the gradient rendering unit substitutes the probability density value of the spatial probability density field into the 3D graphics rendering algorithm, multiplies the probability density value by the reciprocal of the spatial occlusion perspective parameter, and outputs the transparency gradient. The transparency gradient, combined with the four-color status indicators, is mapped by the gradient rendering unit to the transparency channel rendering parameters of the voxel cloud, enabling the gradient rendering unit to use the transparency channel rendering parameters as rendering parameters to transform the spatial probability density field into a voxel cloud, and output the voxel cloud with rendering parameters on the safety monitoring 3D visualization base map. At the same time, the threshold triggering unit receives the three-dimensional integral interference quantity, synchronously obtains the fatigue reduction coefficient of the hydraulic fixed building model and the preset initial safety threshold, multiplies the fatigue reduction coefficient and the preset initial safety threshold to calculate and output the preset safety interference threshold that dynamically changes over time. The threshold triggering unit determines whether the 3D integral interferometry exceeds a preset safety interferometry threshold. If the value exceeds the limit, it triggers a layer-highlighting alarm command, followed by outputting a closed-loop control command including the interferometry coordinates. The closed-loop scheduling module receives and parses the closed-loop control command to extract the interferometry coordinates, then sends these coordinates to the UAV for 3D laser scanning. The closed-loop scheduling module receives the field point cloud data returned by the UAV, extracts the bounding box geometric volume from the field point cloud data as the true volume, and simultaneously extracts the probability coverage boundary of the spatial probability density field to calculate the predicted volume. The closed-loop scheduling module performs a volume comparison calculation by intersecting the true and predicted volumes based on their overlap, and sends the cross-validation results, including the volume overlap rate, to the safety production scheduling terminal.

[0048] Unstructured hazard source data generated from field hydrological surveys represents the raw text records reported by surveyors via handheld terminals. Because the text content includes non-standardized geographical descriptions such as "not far downstream of the dam" and vague location descriptions, the system requires that unstructured hazard source data be bound to hazard source attributes to provide multi-dimensional analytical support. Hazard source attributes encompass accident-inducing attributes that could trigger hydrological safety accidents, as well as accident type attributes that may be induced. Accident-inducing attributes record direct driving factors that cause risks, such as continuous heavy rainfall or illegal sand mining, while accident type attributes define specific disaster manifestations such as piping, landslides, or hydrological station foundation settlement. The system packages accident-inducing attributes and accident type attributes together and temporarily caches them in system memory for the risk assessment model to access at any time.

[0049] To meet the needs of non-standardized geographic description feature parsing, the system has specifically constructed a river and reservoir hydrological geographic ontology database. This database deeply integrates common knowledge of water conservancy operations with accumulated hydrological monitoring and reporting place names, forming a massive geographic semantic retrieval engine. Because the database internally incorporates a knowledge graph dictionary that includes all standardized hydrological station names, dike flood control station numbers, fixed hydraulic structure names, and historical colloquial aliases within the jurisdiction, and establishes spatial mapping relationships between entities, the natural language processing model can directly map the colloquial description of the "Old Wangjia Sluice Gate" by surveyors to the specific sluice gate node recorded in official documentation when retrieving the knowledge graph dictionary for sequence labeling. After mapping, the system sends the matched specific sluice gate node name along with its contextual semantic structure to the topological constraint unit for validity verification.

[0050] Unstructured hazard source data reported by field surveyors often contains intricate spatial location descriptions. Traditional literal matching algorithms are prone to semantic ambiguity, prompting natural language processing (NLP) models to specifically cascade word embedding layers and one-dimensional convolutional neural network (CNN) layers. The NLP model receives unstructured hazard source data and uses a knowledge graph dictionary from a river and reservoir hydrogeographic ontology to transform it into a high-dimensional sequence of word feature vectors. The word embedding layer then passes the output word feature vector sequence to the CNN layer. The CNN layer internally uses local feature extraction convolutional kernels covering different receptive field widths. These kernels continuously perform sliding window multiplication and addition operations on the word feature vector sequence, specifically capturing the local spatial location combinations inherent in colloquial phrases such as "downstream of the back slope" and "right bank of the dam." The CNN layer calls an activation function to perform a non-linear mapping on the multiplication and addition results, outputting a shallow semantic vector matrix carrying hydrological local phrase features.

[0051] Shallow semantic vector matrices only represent isolated phrase fragments. Constructing spatial reference logic across long sentence segments requires a natural language processing model to connect a bidirectional long short-term memory (LSTM) network layer after a one-dimensional convolutional neural network layer. The LSTM layer receives the shallow semantic vector matrix and inputs it into the forward and backward hidden nodes respectively. The forward hidden node extracts semantic dependencies word-by-word along the reading order of the text, while the backward hidden node traces back the spatial location reference conditions along the reverse direction of the text. The forgetting gate within the LSTM layer is responsible for calculating retention weights, attenuating and filtering redundant word vectors such as "maybe" and "probably," which have no practical mapping value. The input gate is responsible for extracting distance values ​​and orientation vectors to update the long-term cell state. The LSTM layer concatenates and fuses the hidden states output by the forward and backward hidden nodes, outputting a deep spatial state vector that incorporates the global hydrological spatial topological context.

[0052] The deep spatial state vector, carrying complete spatial reference inference information, is then fed into the Conditional Random Field (CRF) decoding layer of the natural language processing model to perform the final entity classification. The CRF decoding layer receives the deep spatial state vector and, combined with its internally pre-set entity label state transition matrix, calculates the globally optimal probability sequence for each entity label combination corresponding to the deep spatial state vector. Because the entity label state transition matrix restricts the topological constraint that distance metrics cannot exist independently of spatial entities, the CRF decoding layer performs decoding and segmentation on the globally optimal probability sequence, directly separating the spatial entities, distance metrics, and water system features hidden within the text.

[0053] Before being deployed in spatial semantic parsing operations, the natural language processing (NLP) model requires pre-training. The system server continuously collects historical hydrological field survey records from rivers and reservoirs, constructing a raw corpus using these records. Annotators retrieve standard place names and hydrological terminology from the river and reservoir hydrological geographic ontology database, manually delineating boundaries and labeling categories on the text sequences within the raw corpus according to these standards. Labeling includes spatial entity tags, distance metric tags, and river system feature tags. The system server transforms the manually labeled raw corpus into a labeled standard training set. This standard training set is then divided into training and validation sets. The training set is input into the NLP model for forward propagation network computation. After feature mapping at the embedding layer, semantic extraction at the bidirectional long short-term memory (LSTM) layer, and path decoding at the conditional random field (CRF) layer, the NLP model outputs a predicted label sequence corresponding to the training set. The system server then uses the cross-entropy loss function to calculate the error gradient between the predicted label sequence and the labeled standard training set. The system server employs an adaptive moment estimation optimizer. It propagates the error gradient back to the hidden layers of the natural language processing model via backpropagation to update the network weights. The system server iteratively executes the forward propagation network computation and backpropagation weight update steps until the loss value output by the cross-entropy loss function falls below a pre-set convergence threshold. Once the loss value falls below the convergence threshold, the system server stops updating network weights and outputs a natural language processing model with fixed parameters for the spatial semantic parsing module to use.

[0054] In the training parameter setting stage of the natural language processing model, the system server is configured with a feature mapping dimension of 256 for the word embedding layer, and the number of neurons in both the forward and backward hidden nodes inside the bidirectional long short-term memory network layer structure is configured to be 128.

[0055] Meanwhile, the system server sets the initial learning rate of the adaptive moment estimation optimizer to 0.001 and the batch size of the model training cycle to 64 samples. To limit overfitting caused by numerical inflation of the network weight matrix, the system server sets the weight decay rate to 0.0001 and uses the weight decay rate to perform weight numerical boundary constraints.

[0056] To establish the intrinsic relationship between the algorithm's input and output data and the hydrological survey scenario, the system server constructs the input sequence feature matrix required by the algorithm. The row dimension of the input sequence feature matrix is ​​equal to the fixed truncation length of the hydrological text sequence, and the column dimension of the input sequence feature matrix is ​​equal to the feature map dimension of the word embedding layer.

[0057] The system server extracts unstructured hazard source data generated from river and reservoir hydrological operations, maps and fills the data structure of the input sequence feature matrix with this unstructured hazard source data. Subsequently, the bidirectional long short-term memory network layer receives the filled input sequence feature matrix, performs temporal feature extraction operations on the input sequence feature matrix, and then outputs a hidden layer state matrix containing contextual features.

[0058] After temporal feature extraction, the Conditional Random Field (CRF) decoding layer receives the hidden layer state matrix and performs probability decoding calculations, thus outputting a conditional probability matrix. The system server extracts the conditional probability matrix, performs Viterbi decoding on it, and finally outputs a predicted label sequence corresponding to the hydrological text sequence and carrying the entity classification results.

[0059] In the spatial constraint logic construction phase, the system server extracts the river flow direction rules and water level fluctuation topology from the river and reservoir hydrogeographic ontology database, transforming these rules into a mathematical feature constraint matrix. The row and column dimensions of the feature constraint matrix are completely consistent with those of the entity label state transition matrix within the conditional random field decoding layer algorithm architecture.

[0060] The system server extracts the feature constraint matrix and the entity label state transition matrix, and then performs a Hadamard product calculation on these matrices. By performing the Hadamard product calculation, the system server forcibly corrects label transition probabilities that do not conform to the hydrological spatial topology to 0. Through matrix multiplication, the system server establishes matrix-level constraint relationships between hydrological business logic and model output data within the algorithm's computational chain.

[0061] Sequence labeling operations extract spatial entity words, distance measures, and water system features from text, forming the foundation of spatial topology analysis. Spatial entity words represent fixed hydraulic facilities or geographical landmarks with clearly defined coordinates in the real physical world. Distance measures, by extracting numbers and length units from the text, objectively reflect the relative displacement vector of the hazard source from the spatial entity words. Water system features encompass the water flow velocity, perennial flow direction, and water level fluctuations of the target river section. Combining the extracted elements, the system uses spatial entity words as the starting point for relationships, while transforming distance measures and water system features into directed edges with direction and length attributes. After combination and splicing, a vector spatial relationship tree is output and stored in a graph database for downstream model access.

[0062] The standard geographic information base map integrates high-resolution satellite orthophotos and data from the National Surveying and Mapping Baseline Network to construct a seamlessly stitched three-dimensional coordinate grid, providing a unified absolute three-dimensional coordinate system framework for the hydrological safety monitoring system. When the system retrieves and matches spatial entity words within the standard geographic information base map, it extracts the corresponding three-dimensional absolute coordinates. These three-dimensional absolute coordinates are then set as absolute spatial reference anchor points, which the system then directly inputs as the origin of the spatial transformation into the subsequent affine transformation module.

[0063] The actual evolution of hazards on riverbanks is often accompanied by the impact and erosion of water flow. The combination of water velocity and discharge values ​​recorded in the river system characteristics forms the hydrodynamic weight. The hydrodynamic weight reflects the magnitude of the thrust exerted by the hydrological environment on the hazard, prompting the system to establish a dynamic affine transformation matrix based on the hydrodynamic weight and distance measurements. The dynamic affine transformation matrix, as a mathematical calculation matrix including rotation, translation, and scaling factors, is specifically used to simulate the spatial coordinate drift trajectory of the hazard under the impact of water flow. To achieve spatial position correction, the affine transformation unit converts the distance measurements into relative vectors and substitutes them into the dynamic affine transformation matrix to perform calculations, outputting theoretical center coordinates that compensate for hydrodynamic drift errors.

[0064] Considering that the spread of hazards is often irregular, the system extracts data from digital elevation models to calculate topographic slope, river channel direction, and bank soil erosion tendency index. Topographic slope indicates the direction of gravitational potential energy release, while the bank soil erosion tendency index reveals the vulnerability of the soil structure to damage. The system combines topographic slope, river channel direction, and bank soil erosion tendency index into adaptive bias weights, which are then input into a direction-adaptive probability distribution algorithm. The introduction of adaptive bias weights causes the Gaussian kernel function within the algorithm to deform, stretching the covariance boundary towards low-lying areas and soft soil, resulting in a spatial probability density field with a three-dimensional continuous grid distribution.

[0065] Changes in the meteorological environment directly affect the outbreak cycle of hydrological safety risks. The system extracts rainfall and duration values ​​from meteorological and hydrological forecast data and performs logarithmic conversion to generate a rainfall time decay coefficient. The rainfall time decay coefficient reflects the phenomenon of a sharp drop in safety redundancy as rainfall continues to accumulate. The system then weights and sums the rainfall time decay coefficient with accident causation attributes and accident type attributes, and inputs this sum into the risk assessment model to calculate the probability and severity of occurrence. After the risk assessment model outputs a matching risk level, it assigns a corresponding four-color status label (red, orange, yellow, and blue) to the risk level so that the system can use the four-color status label as a visual classification tag to be stored and retrieved by the 3D visualization engine.

[0066] Given that the resistance of fixed hydraulic structures to external damage is not uniform, the structural vulnerability field corresponding to the fixed hydraulic building model pre-defines the dam or hydrological station into a 3D mesh. The structural vulnerability field specifically assigns higher vulnerability values ​​to critical components such as load-bearing columns and water-facing retaining walls, while assigning lower values ​​to ordinary non-load-bearing areas. In the interference calculation stage, the system calculates the angle between the diffusion normal vector of the spatial probability density field and the force-bearing tangent of the structural vulnerability field, extracting the cosine value of the angle as the angle collision coefficient to reflect the geometric collision efficiency between the impact force of the hazard source and the stress-bearing surface of the building. The system then extracts the probability density value and structural vulnerability value within the same voxel, performs voxel-level 3D multiplication and accumulation calculations combined with the angle collision coefficient, and performs global summation integration, outputting the 3D integral interference quantity to the rendering and early warning module.

[0067] To recreate the spread of risk in a 3D visualization scene, microscopic rendering units are needed. The system uses this to transform the spatial probability density field into a voxel cloud composed of multiple tiny 3D cube arrays with independent coordinates. To prevent the solid walls of the hydrological station from obscuring the risk warnings behind it, the system calculates the spatial occlusion perspective parameters of the fixed hydraulic structure model from the observation perspective. The system multiplies the probability density value of the spatial probability density field by the reciprocal of the spatial occlusion perspective parameters to output a transparency gradient, and combines this transparency gradient with four-color status indicators to form the transparency channel rendering parameters for the voxel cloud. Based on these transparency channel rendering parameters, the 3D rendering engine performs pixel-by-pixel coloring on the 3D base map.

[0068] Considering that the structural health of water conservancy projects deteriorates with increasing service life, the system acquires the fatigue reduction coefficient of the fixed building model and uses this coefficient to downwardly adjust a fixed initial safety threshold, outputting a dynamically evolving preset safety interference threshold. When the 3D integral interference exceeds the preset safety interference threshold, the system immediately triggers a closed-loop control command to dispatch a drone to the site to perform fixed-point 3D laser scanning. The point cloud data transmitted by the drone is used to extract the actual volume reflecting the collapse or expansion of the terrain through bounding box extraction. Simultaneously, the system extracts the probability coverage boundary of the spatial probability density field to calculate the predicted volume. The system performs a volume comparison calculation by intersecting the actual volume and the predicted volume using overlap calculation, and finally outputs the cross-validation results, including the volume overlap rate, and pushes them to the safety production scheduling terminal.

[0069] The risk quantification and assessment module has a pre-defined risk threshold range, with the numerical range of the preset risk threshold range set to 0 to 100. In order to distinguish between states, the risk assessment model divides the preset risk threshold range into four numerical stages at equal intervals.

[0070] When the risk assessment score output by the risk assessment model is greater than or equal to 0 and less than 25, the quantification coding unit is bound to a blue status icon; when the risk assessment score is greater than or equal to 25 and less than 50, the quantification coding unit is bound to a yellow status icon; when the risk assessment score is greater than or equal to 50 and less than 75, the quantification coding unit is bound to an orange status icon; when the risk assessment score is greater than or equal to 75 and less than or equal to 100, the quantification coding unit is bound to a red status icon.

[0071] As the limiting parameter that a hydraulic fixed structure model can withstand under ideal conditions in terms of structural interference strength, the preset initial safety threshold retrieved by the threshold triggering unit is specifically set to a standard interference unit between 800 and 1200. Meanwhile, the fatigue reduction factor obtained by the threshold triggering unit is a dimensionless decimal that gradually decreases with the service life of the hydraulic fixed structure model.

[0072] Based on the service life of the fixed hydraulic structure model, the system dynamically assigns a fatigue reduction factor. When the service life is between 0 and 5 years, the system sets the fatigue reduction factor to 1.0; when the service life is between 5 and 20 years, the system sets the fatigue reduction factor between 0.85 and 0.95; when the service life exceeds 20 years, the system sets the fatigue reduction factor between 0.60 and 0.85.

[0073] After obtaining the specific value, the threshold triggering unit multiplies the fatigue reduction coefficient by the preset initial safety threshold. After the product operation, the value of the preset safety interference threshold output by the threshold triggering unit dynamically evolves into a calculation result between 480 and 1200 standard interference units.

[0074] In the spatial region extraction process, the closed-loop scheduling module relies on a safety cutoff boundary to provide a numerical filtering benchmark. The safety cutoff boundary represents the lowest probability density threshold in the spatial probability density field that poses a destructive threat to the surrounding spatial environment. The system sets the safety cutoff boundary values ​​to be small probability values ​​between 0.15 and 0.25.

[0075] The closed-loop scheduling module extracts the probability density values ​​encompassed in the three-dimensional coordinate system of the spatial probability density field and compares these values ​​with the safe truncation boundary. After numerical comparison, the closed-loop scheduling module filters out three-dimensional mesh voxels with probability density values ​​greater than the safe truncation boundary. Finally, the closed-loop scheduling module performs outer contour wrapping calculations on the filtered three-dimensional mesh voxels and outputs the predicted volume for volume comparison.

[0076] The affine transformation unit extracts distance measures from the vector space relationship tree and transforms them into relative vectors with spatial direction attributes. It also extracts the absolute spatial reference anchor point from the standard geographic information base map as the origin for the 3D coordinate transformation. Furthermore, the affine transformation unit extracts hydrodynamic weights from the vector space relationship tree. Based on the river system characteristics, it constructs a rotation and translation matrix that reflects the displacement trend of water flow scouring. Finally, the affine transformation unit substitutes the relative vectors, hydrodynamic weights, rotation and translation matrix, and absolute spatial reference anchor point into the coordinate translation calculation formula, outputting the theoretical center coordinates of the hazard source on the standard geographic information base map.

[0077] The specific formula for calculating coordinate translation is as follows:

[0078] In the formula for calculating coordinate translation, Represents the central coordinates of the theory. Represents the rotation and translation matrix. Represents a relative vector. Represents hydrodynamic weights, This represents the absolute spatial reference anchor point. The probabilistic dimensionality reduction unit calculates the topographic slope, river course, and bank soil erosion tendency index for the current region. The probabilistic dimensionality reduction unit combines the topographic slope, river course, and bank soil erosion tendency index into adaptive bias weights. Based on these adaptive bias weights, the probabilistic dimensionality reduction unit numerically reconstructs the covariance matrix of the Gaussian distribution model, generating a deformed covariance matrix that can control the spatial diffusion pattern.

[0079] The probabilistic dimensionality reduction unit sets the theoretical center coordinates as the probability distribution center. It substitutes the spatial sampling point coordinates, theoretical center coordinates, and deformation covariance matrix included within the standard geographic information base map into the Gaussian kernel deformation calculation formula to perform matrix mapping calculations, outputting a spatial probability density field with a continuous grid distribution.

[0080] The specific formula for Gaussian kernel deformation calculation is as follows:

[0081] In the Gaussian kernel deformation calculation formula It represents the probability density values ​​covered by the spatial probability density field; The symbol represents an exponential function with the natural constant as its base; Represents constant coefficients; Represents the coordinates of the spatial sampling point; Represents the central coordinates of the theory; A transpose matrix representing the difference between the coordinates of spatial sampling points and the coordinates of the theoretical center; The inverse matrix representing the transformed covariance matrix; This represents the difference matrix between the coordinates of the spatial sampling points and the coordinates of the theoretical center. Wherein, and A column vector in a three-dimensional Cartesian coordinate system. Constrained by topographic slope and river course Spatial covariance matrix. The quantization and coding unit acquires meteorological and hydrological forecast data and extracts the rainfall and duration values ​​recorded in the data. The quantization and coding unit substitutes the rainfall and duration values ​​into the logarithmic conversion formula. Before performing weighted summation, because the accident cause score, accident type score, and rainfall time decay coefficient belong to different physical dimensions and numerical magnitudes, the system uses a max-min standardization strategy to uniformly map these three types of values ​​to a unified standard. Within a continuous dimensionless mathematical domain. Subsequently, the risk assessment model retrieves pre-set severity bias parameters. Substituting the normalized values ​​into the risk product formula, the output is a risk assessment score used to match the risk level. .

[0082] The specific formula for the risk product is as follows:

[0083] In the formula, This represents the risk assessment score; Represents the weight of the incentive; This represents the normalized causation score. Represents type weight; This represents the normalized type score value; Represents rainfall weight; This represents the normalized rainfall duration decay coefficient; This represents the severity bias parameter. The quantization and coding unit then extracts the hazard source attributes, converting the accident precipitating attributes included within the hazard source attributes into precipitating score values, and converting the accident type attributes included within the hazard source attributes into type score values. The quantization and coding unit then performs a weighted summation of the precipitating score values, type score values, and rainfall time attenuation coefficient, outputting an occurrence probability value.

[0084] The risk assessment model retrieves pre-set severity bias parameters. It then substitutes the probability of occurrence and the severity bias parameters into the risk product formula, outputting a risk assessment score used to match the risk level.

[0085] The specific formula for the risk product is as follows:

[0086] In the risk product formula, Represents the risk assessment score. Represents the weight of the incentive. This represents the score of the trigger. Represents type weight, Representative type rating value, Represents rainfall weight, Represents the rainfall time decay coefficient. This represents the severity bias parameter. The interferometric computation unit loads the structural vulnerability field corresponding to the hydraulic fixed structure model. The interferometric computation unit extracts the diffusion normal vector from the surface of the spatial probability density field. The interferometric computation unit extracts the angle of the force-bearing tangent at the periphery of the structural vulnerability field. The interferometric computation unit calculates the geometric angle between the diffusion normal vector and the angle of the force-bearing tangent, and extracts the cosine value corresponding to the geometric angle as the angle collision coefficient. The interferometric computation unit divides the same three-dimensional coordinate system into multiple volume elements with completely equal volumes. The interferometric computation unit extracts the probability density value and structural vulnerability value within the same volume element. The interferometric computation unit substitutes the probability density value, structural vulnerability value, angle collision coefficient, and unit volume of the volume element into a voxel-level three-dimensional multiplication and accumulation formula to perform a global summation integral operation, outputting the three-dimensional integral interferometric quantity.

[0087] To achieve accurate coupling assessment of dynamic risk fields and fixed physical entities in three-dimensional space, the system first performs voxel-based segmentation of the target region using the three-dimensional coordinate system, and sets the unit volume of each volume element. The preset spatial integration step size is used. The interference calculation unit uses the bounding box of the hydraulic fixed structure model as the integration search boundary, and traverses all active volume elements within the boundary.

[0088] The specific formula for voxel-level three-dimensional multiplication and summation is as follows:

[0089] In the voxel-level three-dimensional multiplication and accumulation formula Represents the three-dimensional integral interference quantity; Represents the integral search boundary for the first... To the The global summation integral sign performed on each volume element; Representing the The probability density values ​​corresponding to each volume element; Representing the The structural fragility value corresponding to each volume element; Representing the The included angle collision coefficient corresponding to each volume element; Representing the The unit volume of a volume element.

[0090] The gradient rendering unit acquires the spatial occlusion perspective parameters of the hydraulic fixed structure model on the 3D visualization base map of the safety monitoring system, reflecting the resistance of the line of sight through the solid walls of the hydraulic structure. The gradient rendering unit extracts the probability density values ​​encompassed by the spatial probability density field. It then substitutes these probability density values ​​and the spatial occlusion perspective parameters into the inverse formula for line-of-sight penetration to perform numerical conversion, outputting the transparency gradient.

[0091] The formula for the reciprocal of line-of-sight penetration is as follows:

[0092] In the reciprocal formula for line of sight penetration Represents the transparency gradient. Represents the probability density value. This represents the perspective parameters of spatial occlusion. The threshold trigger unit obtains the fatigue reduction coefficient of the hydraulic fixed structure model over time. The fatigue reduction coefficient reflects the attenuation law of the pressure resistance of hydraulic structures as the service life increases. The threshold trigger unit retrieves the preset initial safety threshold fixed in the system. The threshold trigger unit substitutes the fatigue reduction coefficient and the preset initial safety threshold into the threshold dynamic reduction formula to perform a product calculation and outputs the preset safety interference threshold that is dynamically updated in real time.

[0093] The specific formula for dynamic threshold reduction is as follows:

[0094] In the dynamic threshold reduction formula This represents the preset safety interference threshold. Represents the fatigue reduction factor. This represents a preset initial safety threshold. The closed-loop scheduling module receives the on-site point cloud data returned by the UAV equipment. The closed-loop scheduling module performs 3D boundary contour extraction on the on-site point cloud data, outputting the true volume that reflects the actual collapse morphology. The closed-loop scheduling module extracts spatial regions within the spatial probability density field whose probability density values ​​are greater than the safety cutoff boundary, and combines them to generate a predicted volume for pre-simulation. The closed-loop scheduling module substitutes the true volume and the predicted volume into the overlap intersection formula to perform a 3D mesh comparison operation, outputting cross-validation results including the volume overlap rate.

[0095] The specific formula for calculating the overlap is as follows:

[0096] In the formula for finding the intersection of overlaps Represents volume overlap rate. Represents actual volume, Represents the volume intersection operator. Represents the predicted volume. The volume union operator is represented by this symbol.

[0097] Embodiment 1 of this invention: In a safety inspection scenario of a large reservoir's embankment, inspectors discovered a significant seepage on the downstream slope of the dam and reported unstructured hazard data via a mobile terminal. The unstructured hazard data included ubiquitous location descriptions such as "a small amount of seepage occurred on the downstream slope approximately 50 meters downstream of the No. 2 spillway gate." After receiving the unstructured hazard data, the spatial semantic parsing module performed sequence labeling on the ubiquitous location description using a river and reservoir hydrological geographic ontology database, extracting the spatial entity words as "No. 2 spillway gate," a distance measurement of 50 meters, and the water system feature as the downstream orientation of the downstream slope. The topological constraint unit, combined with a knowledge graph dictionary, used the spatial entity words as logical root nodes, established directed connections based on the water level fluctuation topology, and incorporated the hydrological dynamic weights caused by the current reservoir water level rise to construct a vector spatial relationship tree. The probability field mapping module received the vector spatial relationship tree and matched it with the three-dimensional absolute coordinates of the No. 2 spillway gate on a standard geographic information base map, setting it as the absolute spatial reference anchor point. The affine transformation unit constructs a dynamic affine transformation matrix based on the absolute spatial reference anchor point, substitutes the 50-meter relative vector into the matrix to perform rotation and translation calculations, and outputs the theoretical center coordinates of the hazard source. The probability dimensionality reduction unit extracts digital elevation model data around the theoretical center coordinates and calculates the terrain slope and the erosion tendency index of the bank slope at the location of the seepage point. The Gaussian kernel function distribution algorithm transforms the terrain slope into an adaptive bias weight, stretches the covariance boundary in the downward direction of the slope from the central origin, and outputs a spatial probability density field with a three-dimensional continuous grid distribution. The risk quantification assessment module obtains the seepage accident induction factors from the hazard source attributes, calculates the rainfall time decay coefficient by combining meteorological and hydrological forecast data, and binds the risk level to a red status indicator after calculation by the risk assessment model. The interference calculation unit extracts the angle between the diffusion normal vector of the spatial probability density field and the stress section of the dam's three-dimensional model and calculates and generates a three-dimensional integral interference quantity. The twin rendering early warning module transforms the spatial probability density field into a voxel cloud and outputs a red cloud with extremely high opacity on the three-dimensional visualization base map according to the rendering parameters. The system determines that the three-dimensional integral interference exceeds the preset safe interference threshold and immediately triggers a layer highlighting alarm command.

[0098] Embodiment 2 of this invention: During a field survey of a river channel in a certain watershed, staff discovered a localized bank collapse on the left bank. Location information, "a collapse occurred on the left bank approximately 100 meters upstream of a certain cross section," was reported using unstructured hazard source data. The spatial semantic parsing module performed semantic segmentation on the unstructured hazard source data, identifying the spatial entity as the "certain cross section" and the distance measurement as 100 meters. The probability field mapping module received the vector space relationship tree and retrieved the absolute spatial reference anchor point on the standard geographic information base map. It then used an affine transformation unit to perform coordinate translation in the opposite direction of water flow to locate the theoretical center coordinates. The probability dimensionality reduction unit calculated the river channel orientation and the bank slope soil erosion tendency index in the collapsed area. The direction-adaptive probability distribution algorithm received adaptive bias weights and generated an anisotropic spatial probability density field along the river channel scour section direction. The risk quantification assessment module determined that the collapse was caused by continuous heavy rainfall. The quantification coding unit calculated the rainfall time decay coefficient and input it into the risk assessment model, outputting an orange status marker indicating a higher risk. The interferometric calculation unit loads the structural vulnerability field corresponding to the 3D mesh model of the embankment, performs voxel-level 3D multiplication and accumulation calculation on the spatial probability density field and the structural vulnerability field, and outputs the 3D integral interferometry. The gradient rendering unit obtains the spatial occlusion perspective parameters of the embankment model from the observation perspective, multiplies the probability density value by the reciprocal of the spatial occlusion perspective parameters, and outputs the transparency gradient. The twin rendering early warning module renders an orange voxel cloud closely attached to the inner side of the embankment on the 3D visualization base map of safety monitoring. The threshold triggering unit extracts the fatigue reduction coefficient of the embankment model as its service life increases, corrects the preset initial safety threshold downward, and outputs a closed-loop control command including the interferometric coordinates after determining that the 3D integral interferometry is in the danger zone. The closed-loop scheduling module drives the UAV to perform fixed-point 3D laser scanning, calculates the overlap intersection between the actual volume generated by the on-site point cloud data and the predicted volume generated by the probability field, outputs the volume overlap rate cross-validation result, and pushes it to the scheduling terminal.

[0099] Embodiment 3 of this invention: During the management of hydrological station buildings and observation fields, management personnel discovered that the foundation of the observation field perimeter wall had been eroded due to a sudden rise in river water. The spatial semantic parsing module receives unstructured hazard source data and extracts spatial entity words for the station observation field. The probability field mapping module locates the observation field coordinates on a standard geographic information base map. The affine transformation unit generates hydrological dynamic weights based on water flow velocity and flow rate values, compensates for water level drift errors, and outputs the theoretical center coordinates. The probability dimensionality reduction unit calculates the topographic slope and soil erosion tendency index at the erosion site, generating a spatial probability density field conforming to the water flow scour trajectory. The risk quantification assessment module receives the risk cause as water flow scour and outputs a yellow status indicator representing general risk. The interferometry calculation unit extracts the angle between the probability field diffusion normal vector and the stress section of the station foundation, calculating a three-dimensional integral interferometry reflecting the intensity of the physical threat. The twin rendering early warning module converts the spatial probability density field into a voxel cloud and maps it to yellow rendering parameters. The gradient rendering unit adjusts the voxel cloud transparency based on spatial occlusion perspective parameters. The system monitors the relative interference between the boundary of the hazard source and the basic assets of the station building through dynamic integral calculation. The threshold trigger unit determines that the three-dimensional integral interference has not reached the major risk threshold but shows an evolving trend, maintains the yellow alarm status, and continuously updates the three-dimensional integral interference.

Claims

1. A river reservoir hydrological safety production management system, characterized in that, include: The spatial semantic parsing module is used to receive unstructured hazard source data including ubiquitous location descriptions, extract spatial entity words, distance measures, and water system features from the ubiquitous location descriptions using a river and reservoir hydrogeographic ontology, and construct a vector spatial relationship tree based on the spatial entity words, distance measures, and water system features. The probability field mapping module is used to receive the vector space relationship tree, extract the corresponding absolute spatial reference anchor point from the spatial entity words in the vector space relationship tree in the preset standard geographic information base map, calculate the theoretical center coordinates according to the absolute spatial reference anchor point, the distance measure and the water system feature, input the theoretical center coordinates into the direction adaptive probability distribution algorithm, and output the spatial probability density field. The risk quantification and assessment module is used to receive the spatial probability density field, obtain the hazard attributes bound to the unstructured hazard source data, input the hazard attributes into the risk assessment model to output a four-color status identifier, the four-color status identifier including a red status identifier, load the structural vulnerability field corresponding to the hydraulic fixed building model, perform three-dimensional multiplication calculation between the spatial probability density field and the structural vulnerability field, and output a three-dimensional integral interference quantity. The twin rendering early warning module is used to receive the four-color status identifier, the spatial probability density field, and the three-dimensional integral interference quantity, convert the spatial probability density field into a voxel cloud, map the four-color status identifier to the rendering parameters of the voxel cloud, output the voxel cloud with the rendering parameters in the three-dimensional visualization base map of security monitoring, obtain a preset security interference threshold, determine whether the three-dimensional integral interference quantity is greater than the preset security interference threshold, and trigger a layer highlighting alarm command when the three-dimensional integral interference quantity is greater than the preset security interference threshold.

2. The river reservoir hydrological safety production management system according to claim 1, characterized in that, The spatial semantic parsing module includes: The entity extraction unit is used to receive the unstructured hazard source data, use the knowledge graph dictionary in the river and reservoir hydrogeographic ontology database, call the natural language processing model to perform word segmentation and sequence labeling on the unstructured hazard source data, extract the spatial entity words, the distance measure and the water system features, and combine the spatial entity words, the distance measure and the water system features to output a discrete data set. The topology constraint unit is used to receive the discrete data set, input the discrete data set into the river and reservoir hydrological geographic ontology database for verification, establish directed connection edges for the spatial entity words, distance measures and water system features in the discrete data set according to the river network and water system flow direction rules and water level fluctuation topology in the river and reservoir hydrological geographic ontology database, extract the water flow velocity and flow rate values ​​from the water system features and combine them into hydrological dynamic weights, assign the hydrological dynamic weights to the directed connection edges, and output the vector space relationship tree carrying the directed connection edges and the hydrological dynamic weights.

3. The river reservoir hydrological safety production management system according to claim 2, characterized in that, The probability field mapping module includes: The reference anchoring unit is used to receive the vector space relationship tree, extract the spatial entity words in the vector space relationship tree, search for matching spatial entity words in the preset standard geographic information base map, and obtain the three-dimensional absolute coordinates corresponding to the spatial entity words as the absolute spatial reference anchor point. The affine transformation unit is used to receive the absolute space reference anchor point, take the absolute space reference anchor point as the origin, and establish a dynamic affine transformation matrix from the relative topological space to the absolute coordinate system according to the water system characteristics, the distance measurement and the hydrological dynamic weight in the vector space relationship tree. The distance measurement in the vector space relationship tree is converted into a relative vector, and the relative vector is substituted into the dynamic affine transformation matrix for rotation and translation calculation, and the theoretical center coordinates are output.

4. The river and reservoir hydrological safety production management system according to claim 3, characterized in that, The probability field mapping module also includes: The probabilistic dimensionality reduction unit receives the theoretical center coordinates, extracts digital elevation model data from the preset standard geographic information base map according to the theoretical center coordinates, calculates the terrain slope, river direction, and bank slope soil erosion tendency index based on the digital elevation model data, combines the terrain slope, river direction, and bank slope soil erosion tendency index into adaptive bias weights, sets the initial distribution boundary according to the distance metric, uses the theoretical center coordinates as the center origin, extracts the Gaussian kernel function corresponding to the directional adaptive probability distribution algorithm, and substitutes the adaptive bias weights and the initial distribution boundary into the Gaussian kernel function to perform Gaussian kernel deformation operation, outputting the spatial probability density field presenting a three-dimensional continuous grid distribution.

5. The river and reservoir hydrological safety production management system according to claim 4, characterized in that, The risk quantification assessment module includes: The quantization and coding unit is used to receive the hazard source attributes, extract the accident cause attributes and accident type attributes from the hazard source attributes, access the external meteorological interface to obtain meteorological and hydrological forecast data, extract the rainfall value and duration value from the meteorological and hydrological forecast data, perform logarithmic conversion to output the rainfall time decay coefficient, perform weighted summation of the accident cause attributes, the accident type attributes, and the rainfall time decay coefficient, input the weighted summation result into the risk assessment model for multiplication of the probability of occurrence and severity, perform calculation and comparison within a preset risk threshold range, output the risk level, and bind the corresponding four-color status identifier to the risk level.

6. The river and reservoir hydrological safety production management system according to claim 5, characterized in that, The risk quantification assessment module also includes: An interferometric calculation unit is used to receive the spatial probability density field, load the structural vulnerability field corresponding to the hydraulic fixed building model, calculate the angle between the diffusion normal vector of the spatial probability density field and the force tangent of the structural vulnerability field, use the cosine value of the angle between the force tangents as the angle collision coefficient, extract the probability density value of the spatial probability density field and the structural vulnerability value of the structural vulnerability field within the same three-dimensional coordinate system voxels, perform voxel-level three-dimensional multiplication and accumulation calculation on the probability density value, the structural vulnerability value and the angle collision coefficient and perform global summation integration, and output the three-dimensional integral interferometric quantity.

7. The river and reservoir hydrological safety production management system according to claim 6, characterized in that, The twin rendering early warning module includes: The gradient rendering unit is used to receive the four-color status identifier and the spatial probability density field, obtain the spatial occlusion perspective parameters of the hydraulic fixed building model in the 3D visualization base map of the safety monitoring, substitute the probability density value of the spatial probability density field into the 3D graphics rendering algorithm, multiply the probability density value by the reciprocal of the spatial occlusion perspective parameter, output the transparency gradient corresponding to the spatial occlusion perspective parameter, map the four-color status identifier and the transparency gradient to the transparency channel rendering parameters of the voxel cloud, use the transparency channel rendering parameters as the rendering parameters, transform the spatial probability density field into the voxel cloud, and output the voxel cloud with the rendering parameters in the 3D visualization base map of the safety monitoring.

8. The river and reservoir hydrological safety production management system according to claim 7, characterized in that, The twin rendering early warning module also includes: The threshold triggering unit is used to receive the three-dimensional integral interference quantity, obtain the fatigue reduction coefficient and preset initial safety threshold of the hydraulic fixed structure model, multiply the fatigue reduction coefficient and the preset initial safety threshold to calculate and output the preset safety interference threshold, determine whether the three-dimensional integral interference quantity is greater than the preset safety interference threshold, and trigger the layer red alarm command when the three-dimensional integral interference quantity is greater than the preset safety threshold, and output the closed-loop control command including the interference coordinates.

9. The river and reservoir hydrological safety production management system according to claim 8, characterized in that, Also includes: The closed-loop scheduling module is used to receive the closed-loop control command, parse the closed-loop control command to extract the interference coordinates, send the interference coordinates to the UAV equipment to instruct the UAV equipment to perform three-dimensional laser scanning, receive the field point cloud data returned by the UAV equipment, extract the bounding box geometric volume of the field point cloud data as the true volume, extract the probability coverage boundary of the spatial probability density field to calculate the predicted volume, perform volume comparison calculation by intersecting the true volume and the predicted volume based on the overlap, output the cross-validation result including the volume overlap rate, and send the cross-validation result to the safety production scheduling terminal.

10. A method for managing hydrological safety in rivers and reservoirs, applied to the river and reservoir hydrological safety management system as described in any one of claims 1 to 9, characterized in that, include: Step 1: The spatial semantic parsing module receives unstructured hazard source data including ubiquitous location descriptions, extracts spatial entity words, distance measures, and water system features from the ubiquitous location descriptions using the river and reservoir hydrogeographic ontology, and constructs a vector spatial relationship tree based on the spatial entity words, the distance measures, and the water system features. Step 2: The probability field mapping module receives the vector space relationship tree, extracts the corresponding absolute spatial reference anchor point from the spatial entity words in the vector space relationship tree in the preset standard geographic information base map, calculates the theoretical center coordinates according to the absolute spatial reference anchor point, the distance measure and the water system feature, inputs the theoretical center coordinates into the direction adaptive probability distribution algorithm, and outputs the spatial probability density field. Step 3: The risk quantification assessment module receives the spatial probability density field, obtains the hazard attributes bound to the unstructured hazard source data, inputs the hazard attributes into the risk assessment model to output a four-color status identifier, the four-color status identifier includes a red status identifier, loads the structural vulnerability field corresponding to the hydraulic fixed building model, performs a three-dimensional multiplication calculation between the spatial probability density field and the structural vulnerability field, and outputs a three-dimensional integral interference quantity. Step 4: The twin rendering early warning module receives the four-color status identifier, the spatial probability density field, and the three-dimensional integral interference quantity. It converts the spatial probability density field into a voxel cloud, maps the four-color status identifier to the rendering parameters of the voxel cloud, outputs the voxel cloud with the rendering parameters in the three-dimensional visualization base map of security monitoring, obtains a preset security interference threshold, determines whether the three-dimensional integral interference quantity is greater than the preset security interference threshold, and triggers a layer highlighting alarm command when the three-dimensional integral interference quantity is greater than the preset security interference threshold.

Images