Multi-scale channel surveying and mapping geospatial database integrated modeling method

Abstract

The application discloses a multi-scale channel surveying and mapping geographic information spatio-temporal database integrated modeling method, and particularly relates to the field of channel surveying and mapping spatio-temporal database, and is used for solving the problems that under the condition that high-frequency dynamic monitoring data continuously enters, the local update range of multi-scale channel maps is difficult to accurately limit, the topological relationship is easy to be unstable during local reconstruction, and the incremental result is difficult to be stably published; the method comprises the following steps: performing clustering and noise reduction on the heterogeneous observation signals transmitted by the channel edge monitoring equipment, and forming a surveying and mapping incremental feature tensor; mapping the surveying and mapping incremental feature tensor to a spatio-temporal topological graph network; combining the hydrodynamic physical constraint adjustment of a target region to drive the stop threshold of an edge weight of a graph network to extract a target node subgraph; and under the navigation isolation state, performing adaptive geometric sparse reconstruction, data persistent writing, multi-scale incremental rendering slice generation and memory page swapping linkage process control on the target node subgraph.

Description

Technical Field

[0001] This invention relates to the field of spatiotemporal databases for waterway surveying, and more specifically, to an integrated modeling method for spatiotemporal databases of multi-scale waterway surveying geographic information. Background Technology

[0002] With the increasing frequency of data acquisition for inland waterway and coastal navigation channels, electronic navigation chart support platforms need to simultaneously handle changes in bed sedimentation and scouring, local dredging adjustments, and navigation boundary updates across multiple scales. Existing platforms typically rely on static tile pyramid slices or simplified feature methods based on geometric distances to organize spatial data. When local changes occur in the underlying mapping data, it is often necessary to recalculate layer relationships synchronously over a large area, making it difficult to simultaneously handle local incremental writing, maintain topological relationships, and ensure consistent cross-scale publishing. In navigation channel mapping scenarios, if the area of ​​local change cannot be accurately defined, the boundary displacement at the micro level can easily cause inconsistencies between the main channel boundary, navigation markers, and related nodes when transmitted to the macro view.

[0003] Patent document CN102722854A discloses a database-based method for producing and applying electronic navigation charts for the Yangtze River, which employs map sheet cropping and layer slicing for online base map publishing. This type of solution is suitable for relatively stable base map organization and publishing. However, under conditions of continuous inflow of high-frequency dynamic monitoring data, it still mainly revolves around layer slicing and geometric processing, lacking a processing chain that links local mapping increments, the propagation range of the spatiotemporal topology map network, isolated state reconstruction, and incremental rendering slice linkage control. When local siltation, scour, or boundary changes only occur in certain navigation sections, continuing to use an update method oriented towards the entire layer can easily lead to problems such as an excessively large calculation range, exposure of intermediate update states, and unstable connection between multi-scale results.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an integrated modeling method for spatiotemporal databases of multi-scale waterway mapping geographic information. This method involves performing clustering and noise reduction on heterogeneous observation signals transmitted from waterway edge monitoring equipment to form a mapping incremental feature tensor. This incremental feature tensor is then mapped to a spatiotemporal topology network. The stopping threshold of the traversal algorithm, driven by the edge weights of the target region's hydrodynamic physical constraints, is adjusted to extract the target node subgraph. Finally, under navigation isolation conditions, adaptive geometric thinning reconstruction, persistent data writing, multi-scale incremental rendering slice generation, and memory paging linkage processing flow control are performed on the target node subgraph to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: S1: Collect heterogeneous observation signals from channel edge monitoring equipment within a preset time window, perform density-based spatiotemporal clustering to remove surface noise, and extract the mapping incremental feature tensor containing spatial coordinate change sequences and geometric evolution envelopes. S2: Receive the incremental feature tensor of the mapping and map it to the pre-constructed spatiotemporal topology graph network. Combine the hydrodynamic physical constraints of the target area to adjust the stopping threshold of the graph network edge weight-driven traversal algorithm and extract the target node subgraph corresponding to the incremental feature tensor of the mapping. S3: Apply an asynchronous consistent topology lock with historical snapshot bypass routing function to the target node subgraph, mark the target node subgraph as a navigation isolation state, and perform adaptive geometric thinning reconstruction of the target node subgraph in combination with the airway control line under the navigation isolation state; S4: Persistently write data to the target node subgraph after adaptive geometric thinning reconstruction, release the asynchronous consistency topology lock, generate multi-scale incremental rendering slices based on the reconstructed node spatial morphology, and trigger the memory paging mechanism based on the query heat distribution of graph network nodes.

[0007] Furthermore, in step S1, a unified timestamp is first added to the original point cloud of the multibeam sonar and the echo of the shore-based radar and converted to the same waterway mapping coordinate reference. The spatiotemporal observation sequence is formed in ascending order of the acquisition time. Then, density-based spatiotemporal clustering is performed to mark the observation points that do not meet the density reachability relationship as water surface noise, and the observation points that meet the continuous density support are merged into a set of retained observation clusters.

[0008] Furthermore, in step S1, the geometric outer boundary coordinate set is extracted one by one for the retained observation cluster set, and the geometric outer boundary coordinate set confirmed in the previous preset time window is read from the local cache. The current geometric outer boundary coordinate set and the geometric outer boundary coordinate set of the previous preset time window are resampled and organized into a sequence of boundary points of the same length according to the same arc length. The coordinate difference is calculated point by point along the circumferential direction of the boundary to form a spatial coordinate change sequence and a geometric evolution envelope, and the two are encapsulated into a mapping incremental feature tensor.

[0009] Furthermore, in step S2, the geometric evolution envelope is projected onto the spatiotemporal topology graph network, and an overlay search is performed on the geometric evolution envelope and the spatial boundaries of each node. Nodes that overlap, cross, or cover the geometric evolution envelope are identified as seed node sets, and the spatial coordinate change sequence is written into the corresponding seed nodes according to the projection relationship of the nearest point of the boundary. The target region is determined by the connected regions covered by the seed node set.

[0010] Furthermore, in step S2, the flow velocity and hydraulic radius of the corresponding cross section of the target area are extracted from the hydrodynamic model base to calculate the sediment carrying force. The bottom scattering intensity is analyzed on the mapping sonar echo corresponding to the target area to invert the median sediment particle size and calculate the bed shear resistance. The sediment carrying force and the bed shear resistance are then fed into the pre-trained radial basis neural network as two-dimensional real-time feature points to output the evolution sensitivity coefficient.

[0011] Furthermore, in step S2, the graph network edge weight-driven traversal algorithm is started with the seed node set as the starting point. The graph network edge weights that have been traversed on the diffusion path are accumulated as path energy dissipation. The stopping threshold of the traversal algorithm is corrected with the evolution sensitivity coefficient. When the path energy dissipation reaches the corrected stopping threshold, the current diffusion path is truncated. After all diffusion paths have completed the truncation determination, the nodes included in the diffusion range and their connected edges are summarized to generate the target node subgraph.

[0012] Furthermore, in step S3, the set of nodes and the set of connected edges in the target node subgraph are read, the spatial range covered by the target node subgraph is registered as the locking range, and the locking range is managed through a state machine in the stages of pending locking, locked, and pending release. When there is a node intersection with the existing locking range, the locking range is updated according to the node union and spatial union. When there is no node intersection, a new locking range is established with the current target node subgraph, and all nodes in the target node subgraph are written into the navigation isolation state.

[0013] Furthermore, in step S3, the locking range is matched with the historical snapshot slices in the memory cache to establish a historical snapshot slice pointer. After the application publishing layer interface receives a cross-level read request, the read request that overlaps with the locking range is split into a locked sub-request and an unlocked sub-request. The historical snapshot slice corresponding to the locked sub-request is returned according to the historical snapshot slice pointer, and the current node data corresponding to the unlocked sub-request is returned according to the underlying spatial database. The data is then merged and output according to the original request scale and spatial splicing relationship.

[0014] Furthermore, in step S3, continuous boundary segments that fall into the geometric evolution envelope or intersect with the boundary of the geometric evolution envelope are extracted from the target node subgraph as candidate boundary segments for reconstruction. Each boundary point is projected onto the channel control line. Shared edge endpoints, geometric evolution envelope intersections, and the nearest projection point of the channel control line are fixed as retained points. Constrained Douglas-Peucker recursive thinning is performed between adjacent retained points. The updated boundary point sequence is written back to the target node subgraph to form the reconstructed node space morphology.

[0015] Furthermore, in step S4, the reconstructed node spatial morphology is written back to the underlying spatial database and a closure check is performed. After the closure check passes, it is registered as a publishable version. Then, the set of slices to be updated is determined based on the overlap relationship between the locked range and the slice grids of each scale. Multi-scale incremental rendering slices are generated and slice index replacement is completed. At the same time, the ship's AIS activity trajectory and the scaling operation of the dispatch center's large screen view are mapped to the graph network node branches to form a query heat distribution. Based on this, inactive node branches are serialized and transferred to the disk medium.

[0016] The technical effects and advantages of the multi-scale waterway surveying geographic information spatiotemporal database integrated modeling method of this invention are as follows: This invention first performs clustering noise reduction and boundary compression on heterogeneous observation signals at the data ingress link front end, converging the original point cloud and echo data into a mapping incremental feature tensor, allowing subsequent processing to focus on the locally changing regions. Compared to processing methods that repeatedly recalculate around the full layer, this organization method makes the spatial coordinate change sequence and geometric evolution envelope a unified incremental expression object, and the input boundaries of local changes are clearer.

[0017] Secondly, this invention maps the incremental feature tensor of the mapping to a spatiotemporal topological graph network, and uses the hydrodynamic physical constraints of the target region to adjust the edge weights of the graph network to drive the stopping threshold of the traversal algorithm, so that the extraction of the target node subgraph no longer depends solely on geometric proximity. Therefore, the propagation range of local changes can expand around the actually affected nodes, and the subsequent geometric reconstruction processing range remains within the already defined local topological region.

[0018] Furthermore, this invention applies an asynchronous consistent topology lock with historical snapshot bypass routing functionality to the target node subgraph, and completes adaptive geometric thinning reconstruction, persistent data writing, and multi-scale incremental rendering slice generation in a navigational isolation state, ensuring the separation of read paths, reconstruction paths, and publish paths during local updates. Combined with the memory paging mechanism triggered by the query popularity distribution of graph network nodes, high-frequency access branches and inactive node branches can be organized according to different storage locations. Attached Figure Description

[0019] Figure 1 This is the overall flowchart of the multi-scale waterway surveying geographic information spatiotemporal database integrated modeling method of the present invention; Figure 2 This is a schematic diagram illustrating the process of generating incremental feature tensors in this invention. Figure 3 This is a schematic diagram of the target node subgraph extraction process of the present invention; Figure 4 This is a schematic diagram of the asynchronous consistent topology lock and historical snapshot bypass routing process of the present invention; Figure 5This is a schematic diagram of the adaptive geometric thinning reconstruction and incremental release process of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 - Figure 5 This invention provides an integrated modeling method for spatiotemporal databases of multi-scale waterway surveying geographic information, comprising: Step S1: Collect heterogeneous observation signals from the channel edge monitoring equipment within a preset time window, perform density-based spatiotemporal clustering to remove surface noise, and extract the mapping incremental feature tensor containing spatial coordinate change sequences and geometric evolution envelopes. Step S2: Receive the mapping incremental feature tensor and map it to a pre-constructed spatiotemporal topology graph network. Combine the hydrodynamic physical constraints of the target region with the stopping threshold of the graph network edge weight-driven traversal algorithm to extract the target node subgraph corresponding to the mapping incremental feature tensor. Step S3: Apply an asynchronous consistent topology lock with historical snapshot bypass routing function to the target node subgraph, mark the target node subgraph as a navigation isolation state, and perform adaptive geometric thinning reconstruction on the target node subgraph in the navigation isolation state in conjunction with the airway control line; Step S4: Persistently write data to the target node subgraph after the adaptive geometric thinning reconstruction, release the asynchronous consistency topology lock, generate multi-scale incremental rendering slices based on the reconstructed node spatial morphology, and trigger the memory paging mechanism based on the query heat distribution of the graph network nodes.

[0022] Specifically, this invention focuses on processing heterogeneous observation signals generated by waterway edge monitoring equipment within a preset time window. First, at the edge computing gateway, spatiotemporal observation sequences are constructed, density-based spatiotemporal clustering is performed, and geometric boundaries are extracted to obtain a mapping incremental feature tensor containing spatial coordinate change sequences and geometric evolution envelopes. Then, this mapping incremental feature tensor is mapped to a spatiotemporal topological graph network. The propagation range of the graph network edge weight-driven traversal algorithm is limited by the hydrodynamic physical constraints of the target region, forming a target node subgraph. Subsequently, an asynchronous consistent topological lock with historical snapshot bypass routing function is applied to the target node subgraph, and adaptive geometric thinning reconstruction is performed in conjunction with the waterway control line under navigation isolation conditions. Finally, persistent data writing, multi-scale incremental rendering slice generation, and memory paging processing based on the graph network node query heat distribution are completed.

[0023] In the integrated modeling method of spatiotemporal database of geographic information for multi-scale waterway mapping, the waterway edge monitoring equipment continuously generates multibeam sonar raw point clouds and shore-based radar echoes. These heterogeneous observation signals directly reflect the local waterbed morphology and surface disturbance state. However, the raw data also contains a large number of unstable observations formed by wave undulations, floating object reflections and occasional flying points, which do not yet meet the conditions for directly participating in waterway topography determination. Therefore, step S1 needs to first uniformly organize, density screen and extract the heterogeneous observation signals within the same time window in the edge computing gateway to form incremental results that can characterize the local real topographic changes.

[0024] S101: Merging the time windows of heterogeneous observation signals into a unified sequence.

[0025] After the channel edge monitoring equipment has continuously uploaded observation data, the edge computing gateway first extracts the raw point cloud of multibeam sonar and the echo of shore-based radar within the same target area within a preset time window. Instead of directly making boundary judgments on single frames of data, it first forms a unified input that can be processed for clustering. The following sub-steps are all executed by the edge computing gateway deployed on the ship or navigation mark.

[0026] The edge computing gateway adds a unified timestamp to various types of observation data arriving within a preset time window, and converts the spatial coordinates in the raw point cloud of multibeam sonar and the echo landing points in shore-based radar echoes that can locate the channel edge area to the same channel mapping coordinate reference. These are then arranged in ascending order of acquisition time to form a spatiotemporal observation sequence. Each observation point in the spatiotemporal observation sequence retains at least its spatial location and acquisition time for subsequent density neighborhood establishment. The raw point cloud of multibeam sonar provides a continuous observation basis for seabed undulations, while shore-based radar echoes provide a spatial distribution reference for surface disturbances within the same time window. Both types of observation points participate in subsequent neighborhood determination under a unified coordinate reference.

[0027] For example, in a shallow water survey scenario along a curved waterway, the shipborne multibeam sonar continuously outputs water depth points, while the radar on the shore navigation mark synchronously outputs wave crest reflection points. The edge computing gateway captures both within the same preset time window and converts them into a unified spatiotemporal observation sequence in the same channel cross-section coordinate system. At this point, the heterogeneous observation signals have been transformed from parallel streaming input into a sequential observation basis that can be used by the DBSCAN spatial density clustering algorithm. Subsequent processing all begins with this spatiotemporal observation sequence.

[0028] S102: Water surface noise removal based on DBSCAN spatial density clustering algorithm.

[0029] After the spatiotemporal observation sequence is constructed, the edge computing gateway uses this spatiotemporal observation sequence as the only clustering input and begins to perform density-based spatiotemporal clustering processing; at this time, it begins to identify real observation clusters that can maintain stable spatial clustering relationships within a preset time window.

[0030] In specific processing, the edge computing gateway uses a spatial radius threshold. Time radius threshold and minimum neighborhood number Construct the decision boundary for the DBSCAN spatial density clustering algorithm, where the spatial radius threshold is... The initial value is determined by the larger of the average point distance of the multibeam sonar point cloud and the positioning resolution of the shore-based radar within the current preset time window, and then corrected by combining the upper quartile of the neighborhood distance of the labeled surface noise samples in the most recent three preset time windows; time radius threshold. The sampling period is taken as twice the duration of two consecutive sampling cycles of the channel edge monitoring equipment, and when asynchronous sampling exists between multibeam sonar and shore-based radar, the larger of the two sampling periods is used; minimum neighborhood points. Take the floor value of the median number of neighborhood points in the stable bed zone within the three most recent preset time windows, and make it no less than 4. For any observation point in the spatiotemporal observation sequence, the edge computing gateway must first be within a spatial distance of no more than [missing value]. And the time difference in data collection is no greater than Under the dual constraints, the neighborhood point set is retrieved, and then the number of points in the neighborhood point set is counted; when the number of points reaches... When the number of observation points reaches a certain threshold, the observation point is designated as the core observation point, and an observation cluster is generated by expanding outward based on the density reachability relationship. If the density of a point does not fall within the reach of any core observation point, the observation point is directly marked as surface noise. The resulting retained observation cluster represents a local bed change region with continuous density support within a preset time window.

[0031] Within the same processing chain, the edge computing gateway no longer extracts observation points marked as surface noise from the boundary, but only retains observation clusters verified by spatiotemporal density as input for subsequent geometric processing. For example, when a survey vessel passes through narrow waters with large waves, the instantaneous echoes caused by wave crests may appear in a single sample, but because they cannot form a continuous density neighborhood in adjacent sampling times and do not satisfy the density reachability relationship with stable bottom points, they will be classified as surface noise. Conversely, sonar anomalies appearing consecutively at the shoal front will remain clustered across multiple adjacent times and will ultimately be retained as true observation clusters. After this sub-step, the spatiotemporal observation sequence is compressed into a set of retained observation clusters that can be used for geometric boundary extraction.

[0032] S103: Geometric evolution envelope extraction, spatial coordinate transformation generation, and mapping incremental feature tensor output.

[0033] After the set of retained observation clusters is formed, the edge computing gateway continues to extract geometric results using each retained observation cluster as a processing unit, and organizes the results into a mapping incremental feature tensor that can be directly received in step S2. The geometric evolution envelope referred to here is the closed outer boundary formed by tracing along the spatial adjacency relationship for all valid observation points within the same retained observation cluster, used to characterize the actual outward expansion range of local bed bulges or depressions within the current preset time window.

[0034] The edge computing gateway first performs an outer boundary tracking algorithm on each retained observation cluster. This algorithm employs alpha shape boundary extraction based on Delaunay triangulation. It first constructs a triangulation network using valid observation points from the retained clusters, then deletes observations whose circumcircle radius is greater than a parameter. The triangle, parameters The coordinates of the remaining outermost edges of the triangles are sorted circumferentially, taking twice the average point distance of the retained observation cluster, to form the corresponding set of geometric outer boundary coordinates. Then, the set of geometric outer boundary coordinates confirmed in the previous preset time window is read from the local cache. The current set of geometric outer boundary coordinates and the set of geometric outer boundary coordinates from the previous preset time window are resampled using an equal arc length resampling method to form a sequence of boundary points of the same length. The number of resampled points is the larger of the current number of geometric outer boundary points and the number of baseline boundary points from the previous preset time window. In the initial acquisition cycle, the current number of geometric outer boundary points is directly used as the number of resampled points. After resampling, the coordinate difference is calculated point by point in the circumferential order of the boundary, forming a spatial coordinate change sequence. The spatial coordinate change sequence reflects the local geometric displacement of the current preset time window relative to the previous preset time window. The geometric evolution envelope maintains the closed boundary shape of the current effective change area. Together, they constitute the main content of the incremental expression. Upon entering the initial acquisition cycle, the edge computing gateway directly writes the current set of geometric outer boundary coordinates into the local cache as the baseline boundary and simultaneously generates a zero-displacement spatial coordinate change sequence, allowing subsequent cycles to use the same comparison rules.

[0035] After completing the boundary extraction and displacement calculation for each retained observation cluster, the edge computing gateway, according to the generation order of the observation clusters within a preset time window, stacks and encapsulates the corresponding spatial coordinate change sequence and geometric evolution envelope into a mapping incremental feature tensor, and outputs this mapping incremental feature tensor to the spatiotemporal topology graph network mapping interface in step S2. For example, when local siltation occurs on the inner side of a bend, the retained observation cluster will have a closed boundary that slightly expands outward towards the center of the channel extracted in the edge computing gateway, and a boundary displacement sequence compared to the previous preset time window will be generated. After being encapsulated, the two become the direct input for locating the target node subgraph in step S2, without the need to upload the full map sheet or the full original point cloud.

[0036] In step S1, the edge computing gateway has merged the heterogeneous observation signals from the channel edge monitoring equipment into a spatiotemporal observation sequence. After removing surface noise using the DBSCAN spatial density clustering algorithm, a set of retained observation clusters representing the actual local bed changes is obtained. Furthermore, the geometric evolution envelope and spatial coordinate change sequence are extracted, and finally, a mapping incremental feature tensor is output. This mapping incremental feature tensor serves as the direct input to step S2, providing both the closed boundary of the local change region and the geometric displacement relative to the previous preset time window. Simultaneously, it retains the current set of geometric outer boundary coordinates for further comparison in the next acquisition cycle.

[0037] Step S1 has completed the denoising and geometric compression of heterogeneous observation signals, forming a mapping incremental feature tensor containing spatial coordinate change sequences and geometric evolution envelopes. At this point, local terrain changes have clear geometric boundaries and incremental displacement expressions. However, the incremental results are still at the local mapping level and have not yet been implemented in the specific propagation range of the spatiotemporal topology network, nor have they distinguished which nodes belong to the real affected areas. Therefore, step S2 needs to complete network mapping, physical risk assessment, and diffusion boundary delineation based on the existing mapping incremental feature tensor.

[0038] S201: Mapping incremental feature tensors to spatiotemporal topological graph networks and determining seed nodes.

[0039] Step S1 has compressed the original observation data into a mapping incremental feature tensor. After receiving the mapping incremental feature tensor, the data processing server does not immediately perform a search across the entire map; instead, it first locates local geometric changes on traversable topological nodes. The following sub-steps are all executed by the data processing server.

[0040] The data processing server projects the geometric evolution envelope onto a pre-constructed spatiotemporal topology graph network. This network uses basic spatial units for waterway mapping as nodes and spatial adjacency and temporal continuity relationships between nodes as edges. The edge weights are pre-written during graph construction. The edge weights for spatial adjacency edges are determined by the normalized ratio of the distance between adjacent node centers to the length of the shared edge. The edge weights for temporal continuity edges are determined by the normalized ratio of the interval between adjacent temporal snapshots to the standard sampling interval. Both are uniformly converted into dimensionless propagation costs and stored. The data processing server performs an overlay search on the geometric evolution envelope and the spatial boundaries of each node. Nodes that overlap, cross, or enclose the geometric evolution envelope are identified as seed nodes. The spatial coordinate change sequence is written into the corresponding seed nodes according to the nearest boundary point projection relationship, forming the initial message source for the traversal algorithm. The connected region covered by the seed node set is identified as the target region for subsequent extraction of water flow velocity, hydraulic radius, and sonar echo mapping.

[0041] In an embodiment of shoal uplift in a bend, the geometric evolution envelope output by the edge computing gateway falls near the right boundary of the main channel. After the data processing server overlays the geometric evolution envelope onto the spatiotemporal topology network, it obtains multiple continuously distributed seed nodes. These seed nodes collectively cover the local segment where the shoal uplift occurs. The spatial coordinate change sequence is then written into the corresponding seed nodes, causing the traversal algorithm to start spreading from nodes with actual geometric displacements, rather than starting from arbitrary nodes across the entire segment. After completing this sub-step, the data processing server obtains the target region and the set of seed nodes. Subsequent physical quantity calculations and diffusion determinations all use the target region and the set of seed nodes as input.

[0042] S202: Calculation of hydrodynamic physical quantities in the target area and generation of sensitivity coefficients for evolution.

[0043] After the seed node set and target region are determined, the data processing server begins to convert local geometric changes into physical features that can participate in the diffusion truncation determination. At this time, the target region output by S201 is referenced, and the cross-sectional parameters in the mapping sonar echo and hydrodynamic model base corresponding to the same time window as the target region are also called. The hydrodynamic model base is a pre-constructed parameterized computational unit for calculating water flow and sediment transport in waterway cross-sections. Its construction process includes: First, segmenting the target waterway according to measured waterway data to form a set of waterway category classes consistent with the waterway category; second, for each waterway category, collecting the average cross-sectional velocity, cross-sectional area, wetted perimeter, measured sediment concentration, and corresponding scouring and deposition records for the corresponding historical period, and calculating the hydraulic radius of each cross-section based on the cross-sectional area and wetted perimeter; third, using the average cross-sectional velocity and hydraulic radius as inputs and the measured sediment concentration as the target, fitting the empirical coefficients in Zhang Ruijin's formula to each waterway category to obtain a pre-set empirical coefficient set corresponding to each waterway category; finally, writing the empirical coefficient set corresponding to each waterway category, along with the cross-sectional identifier, average cross-sectional velocity sequence, and hydraulic radius sequence for that waterway category, into the hydrodynamic model base.

[0044] The data processing server first extracts the water flow velocity of the corresponding cross section of the target area from the hydrodynamic model base. and hydraulic radius The sediment carrying capacity was then calculated using Zhang Ruijin's formula. : , Among them, the sand-carrying capacity ratio coefficient Flow velocity index coefficient and hydraulic radius index coefficient The parameters are read from the built-in parameter table of the hydrodynamic model base according to the category of the target area and the type of seabed sediment. This parameter table is calibrated and stored after being used in the hydrodynamic model base, based on historical cross-sectional flow velocities, sediment concentrations, and measured scouring and deposition results for that section. The sediment-carrying capacity proportional coefficient, flow velocity index coefficient, and hydraulic radius index coefficient are not arbitrarily set values, but are obtained by inversion from historical measured samples for each section category. Specifically, the acquisition method is as follows: a sample set is constructed using the average cross-sectional flow velocity, hydraulic radius, and measured sediment concentration from multiple historical periods under the same section category. While keeping the form of Zhang Ruijin's formula unchanged, a set of sediment-carrying capacity proportional coefficients, flow velocity index coefficients, and hydraulic radius index coefficients that minimize the deviation between the calculated and measured sediment concentrations is obtained through least-squares fitting. When the same section category corresponds to multiple cross-sections, the arithmetic mean of the fitting results for each cross-section is taken as the final preset empirical coefficient set for that section category. When step S202 is invoked, the channel category is first determined based on the channel to which the target area belongs. Then, the sediment carrying capacity ratio coefficient, velocity index coefficient, and hydraulic radius index coefficient corresponding to that channel category are read from the hydrodynamic model base. These coefficients, along with the extracted flow velocity and hydraulic radius, are substituted into Zhang Ruijin's formula to complete the sediment carrying capacity calculation. When the target area spans multiple cross sections, the data processing server first extracts the flow velocity of the corresponding cross section according to the overlap length between the geometric evolution envelope and each cross section. and hydraulic radius Then, the input values ​​are weighted and merged according to the same overlap length into a single set for the target region to ensure that only one sediment carrying capacity corresponding to the target region is generated subsequently. .

[0045] Subsequently, the data processing server performs sediment scattering intensity analysis on the mapped sonar echoes corresponding to the target area. Specifically, it extracts the dominant frequency band energy, echo envelope attenuation rate, and backscattering peak from the mapped sonar echoes and matches them with a pre-calibrated sediment type-echo characteristic lookup table to obtain the sediment type and corresponding grain size range. When the echo characteristics fall into a single range, the median value of that grain size range is directly taken as the median grain size of the sediment. When the echo characteristics are located between two adjacent particle size intervals, the median particle size of the sediment is obtained by linear interpolation based on the characteristic distance between the two intervals. The data processing server then calculates the median particle size of the sediment. The corresponding dimensionless particle size is obtained by looking up the critical Shields parameter in the pre-stored Shields curve lookup table. When the values ​​fall between adjacent entries, linear interpolation is used, and the bed shear strength is calculated accordingly. : , Complete the sediment carrying capacity and bed surface shear strength After calculation, the data processing server will convert the two-dimensional real-time feature points... Linear normalization is performed on the minimum-maximum normalization parameters that are fixed and saved during the training phase of the radial basis function neural network (RBFNN), and then fed into the pre-trained RBFNN. The RBFNN is trained before going online using samples from historical landslides, sudden siltation disasters, and normal stable periods. Each sample consists of a two-dimensional input of sediment carrying capacity and bed shear resistance within the same time window. Samples showing drastic bed changes are marked as 1, and samples without such changes are marked as 0. The centroid of the historical topographic change features in the hidden layer is also considered. The Gaussian kernel width is obtained by performing clustering on the training samples. The average distance from the corresponding centroid to its nearest neighbor centroid is used to output the layer connectivity coefficients. and bias Determined through supervised training. During invocation, the data processing server first applies the Gaussian kernel function. , , in, For the first The centroid of a historical topographical change feature is derived from the training results of historical landslides or sudden siltation disaster samples. The width of the Gaussian kernel corresponding to the centroid; These are the output layer connection coefficients; This represents the number of hidden layer centroids. The evolution sensitivity coefficient ranges from 0 to 1. The closer the evolution sensitivity coefficient is to 1, the closer the current hydrodynamic state of the target area is to historical dangerous conditions.

[0046] The pre-stored Shields curve lookup table is a pre-defined relationship table used to map the median particle size of sediment to the critical conditions for bed initiation. Its function is to determine the critical Shields parameters based on the particle size of sediment in order to further calculate the shear resistance of the bed. The pre-stored Shields curve lookup table is not an abstract concept, but a two-dimensional reference table discretized based on the Shields curve. Its construction process includes: First, based on the well-known Shields curve in fluid mechanics, the sediment initiation conditions are divided into several continuous intervals according to the dimensionless particle size; second, for each interval, a representative dimensionless particle size value is selected, and the critical Shields parameter under that value is retrieved or fitted; third, the dimensionless particle size interval, the representative value of the interval, and the corresponding critical Shields parameter are written into the lookup table in ascending order to form a hierarchical continuous correspondence; finally, the lookup table is stored in the local parameter library of the data processing server. The pre-stored Shields curve lookup table has a structure including at least three columns: the first column is the starting value of the dimensionless particle size interval, the second column is the ending value of the dimensionless particle size interval, and the third column is the critical Shields parameter corresponding to that interval. To improve lookup accuracy, a column representing interval values ​​can be added to the same table. When called in step S202, the corresponding dimensionless particle size is first calculated based on the median particle size of the sediment, and then the interval into which the dimensionless particle size falls is retrieved from the pre-stored Shields curve lookup table. When the dimensionless particle size falls exactly within an interval, the critical Shields parameter corresponding to that interval is directly read. When the dimensionless particle size is near the boundary between two adjacent intervals, linear interpolation is performed based on the critical Shields parameters of the two adjacent intervals to obtain the critical Shields parameter corresponding to the current median sediment particle size. Thus, the pre-stored Shields curve lookup table completes the conversion from the median sediment particle size to the input quantity for calculating the bed shear strength.

[0047] For example, a pre-trained radial basis function neural network can adopt a three-layer structure, including an input layer, a hidden layer, and an output layer. The input layer has two input neurons, which receive the normalized sediment carrying capacity and bed shear resistance, respectively. The hidden layer has 12 radial basis function neurons, each corresponding to a centroid of a historical topographic change feature, and the radial basis function uses a Gaussian kernel function. The output layer has one output neuron, and the activation function uses a Sigmoid function to output an evolution sensitivity coefficient between 0 and 1. The training dataset can consist of historical channel collapses, sudden siltation disasters, and normal stable periods. For example, monitoring records from the same river section over the past 5 years can be selected to form 12,000 samples, of which 3,500 are topographic change samples and 8,500 are stable samples. Each sample uses the sediment carrying capacity and bed shear resistance calculated within the same time window as two-dimensional input features, and uses whether a sudden change in the bed boundary occurs after the time window as a label, with a change recorded as 1 and no change recorded as 0. During training, minimum-maximum normalization is first performed on all input features. Then, K-means clustering is used to extract the centroids of 12 historical topographic change features from the training samples as hidden layer centers. The width of each center is the average of the distances from that center to its three nearest neighbors. Subsequently, the centroids and widths of the historical topographic change features are fixed, and gradient descent with batch iteration is used to train the output layer connection coefficients and biases. The loss function is the binary cross-entropy loss function, with an initial learning rate of 0.01, a batch size of 128, and 200 training epochs. After pre-training, the saved model parameters include at least the upper and lower bounds of the input feature normalization, the coordinates of the centroids of the 12 historical topographic change features, their corresponding widths, the output layer connection coefficients, and the bias values. For example, the upper bound of normalization can be determined by the maximum sediment carrying capacity of 120 and the maximum bed shear resistance of 95 in the training set, and the lower bound can be determined by the minimum sediment carrying capacity of 5 and the minimum bed shear resistance of 8.

[0048] In an embodiment of channel edge scouring, the data processing server extracts a larger water flow velocity and a narrower hydraulic radius from the cross-section of the target area, resulting in a higher sediment carrying capacity. Simultaneously, it resolves the median sediment particle size corresponding to fine-grained bottom sediment from the mapping sonar echoes, thus obtaining a lower bed shear resistance. After the two-dimensional real-time feature points are input into the radial basis function neural network, they become closer to the centroid of historical topographical changes formed by historical bank collapse samples, thereby increasing the evolution sensitivity coefficient. After completing this sub-step, the data processing server obtains the evolution sensitivity coefficient required for subsequent diffusion truncation.

[0049] S203: Edge weight-driven traversal diffusion, dynamic truncation, and target node subgraph output.

[0050] After the evolution sensitivity coefficients are generated, the data processing server sends the seed node set in S201 and the evolution sensitivity coefficients in S202 into the graph network edge weight-driven traversal algorithm to begin extracting the target node subgraph. At this point, the diffusion no longer terminates with a fixed radius, but dynamically adjusts the stopping boundary according to the physical risk level of the target region.

[0051] The data processing server initiates message propagation from each seed node in the seed node set, traversing outwards along adjacent edges according to their weights. To ensure the truncation condition is computable, the data processing server accumulates the graph network edge weights of all traversed edges along a given propagation path as the path energy dissipation. : , in, The current diffusion path contains the edge sequence. Let represent the edge weights of the graph network corresponding to each edge in the edge sequence. The data processing server pre-extracts the 90th percentile from the cumulative edge weight sequence of historical stable local update samples to form the baseline stopping threshold for the traversal algorithm. Then use the evolution sensitivity coefficient Generate dynamic stopping threshold : , in, The smaller the value, the less energy dissipation is allowed along the outward diffusion path under high-risk conditions. When a diffusion path satisfies... When this happens, the data processing server includes neighboring nodes in the current diffusion range and continues to propagate messages outward along their adjacent edges; when a certain diffusion path satisfies At that time, the data processing server forcibly truncates the path diffusion at the corresponding boundary and no longer incorporates subsequent nodes into the current diffusion range.

[0052] Once all diffusion paths initiated by seed nodes have been truncated, the data processing server aggregates all nodes included in the diffusion range and their connected edges to generate a target node subgraph. The target node subgraph maintains the node connectivity and edge weight relationships in the original spatiotemporal topology graph network and is output to step S3 as a locally connected graph for applying asynchronous consistent topology locks and performing adaptive geometric thinning reconstruction. In the same bend-shoal embodiment, if the evolution sensitivity coefficient is high, the dynamic stopping threshold is reduced, and the traversal diffusion only covers navigation boundary nodes and adjacent control nodes directly related to shoal changes; if the evolution sensitivity coefficient is low, the dynamic stopping threshold is relaxed, and the traversal diffusion can cover a more complete locally connected region, thus ensuring that the target node subgraph maintains consistency with the intensity of local changes. At this point, the target node subgraph has been formed, and subsequent step S3 directly applies a navigation isolation state to the target node subgraph without performing a full graph search again.

[0053] In step S2, the data processing server has mapped the incremental feature tensor of the surveying and mapping onto the spatiotemporal topological graph network, determined the seed node set and the target region, and combined with the evolution sensitivity coefficients of the water flow velocity, hydraulic radius, sediment carrying capacity, median sediment particle size, bed shear resistance, and the radial basis function neural network output, applied a dynamic truncation judgment to the graph network edge weight-driven traversal algorithm, forming a target node subgraph with clear boundaries and complete connectivity. At this point, the influence domain of local topographic changes in the topological network has been limited to the directly processable node range.

[0054] Step S2 has already narrowed the influence range corresponding to the incremental feature tensor of the mapping into the target node subgraph, indicating that the topological nodes and connected edge relationships involved in the local changes have been accurately delineated; however, within this local range, the node boundaries are about to be redrawn and replaced. If external terminals are still allowed to read directly according to the current database content, the unclosed boundaries in the update will coexist with the online navigation elements. Therefore, it is necessary to first establish isolation control on the target node subgraph, and then perform geometric thinning reconstruction under controlled conditions, so that the local update is always within the closed processing chain.

[0055] S301: Asynchronous consistent topology lock is applied with the navigation isolation state flag.

[0056] Step S2 has already output the target node subgraph. After receiving the target node subgraph, the data processing server does not immediately rewrite the node geometry, but instead establishes isolation boundaries for the local topology range that is about to change. The following sub-steps are all executed by the data processing server, with the application publishing layer interface, memory cache, and underlying spatial database serving as the execution objects controlled by the data processing server.

[0057] The data processing server first reads the set of nodes and connected edges in the target node subgraph and registers the spatial range covered by the target node subgraph as the locked range. The asynchronous consistent topology lock is implemented using a state machine for management. The state machine maintains at least three processing stages for the locked range: pending lock, locked, and pending release. When the target node subgraph enters the geometric reconstruction process, the state machine first searches the underlying spatial database for existing locked ranges that intersect with the locked range. If an intersection exists, the current target node subgraph and the existing locked range are updated to the same locked range by node union and spatial union. Then, all nodes after the union are written into the locked stage. If no intersection exists, a new locked range is directly established based on the current target node subgraph and written into the locked stage. After this writing is completed, all nodes in the target node subgraph are marked as being in a navigationally isolated state, and subsequent read paths from external terminals will no longer directly access the current geometric data of these nodes.

[0058] Within the same processing chain, the data processing server also generates historical snapshot slice pointers for the locked area. These pointers represent mappings to complete and error-free historical snapshot slices from the previous frame in the memory cache. The mapping is based on the overlap between nodes within the locked area and snapshot slices at various scales. Specifically, the data processing server performs an overlap match between the locked area and the spatial range of historical snapshot slices in the memory cache, establishing a one-to-one correspondence between the addresses of historical snapshot slices covering the locked area and the locked area itself. This correspondence is then written into the state machine along with the navigation isolation state. At this point, the target node subgraph has been transformed from a regular update region into a locked region with historical snapshot bypass routing functionality.

[0059] For example, in a scenario of shoal uplift inside a narrow bend, the target node subgraph output in step S2 covers the main channel boundary nodes and adjacent control nodes. The data processing server first registers this local area as a locked range, and then writes the address of the historical snapshot slice that has been published on both the ship's side and the scheduling dashboard in the previous frame into this locked range. Before subsequent geometric redrawing is completed, any read operation falling into this locked range will be directed to the cached historical snapshot slice, without touching the node boundaries that are being rewritten.

[0060] S302: Historical snapshot bypass routing and cross-level read request shielding.

[0061] After the navigation isolation state marking is completed, the target node subgraph is already in a locked state. However, cross-level read requests from external terminals will continue to arrive at the application publishing layer interface. Therefore, the data processing server continues to perform traffic splitting on the read path based on the locking range and historical snapshot slice pointers generated by S301. At this time, the redraw has not yet started to be submitted externally, and read control is executed before the geometry replacement action.

[0062] After receiving a cross-level read request from the large screen or ship terminal, the application publishing layer interface first parses the spatial range, scale, and timestamp of the read request, and then performs an overlap judgment between the spatial range of the read request and the locked range. When the spatial range of the read request and the locked range do not overlap, the application publishing layer interface maintains the original read path and extracts the current node data of the unlocked area from the underlying spatial database. When the spatial range of the read request and the locked range overlap, the application publishing layer interface splits the read request into locked sub-requests and unlocked sub-requests according to the overlap result. The locked sub-request directly extracts the complete and error-free historical snapshot slice from the memory cache based on the historical snapshot slice pointer, while the unlocked sub-request still returns the current database content. The application publishing layer interface then merges the two results according to the original request scale and spatial splicing relationship and outputs them to the external terminal. Thus, although cross-level read requests continue to arrive, the query path within the locked range has been intercepted by the historical snapshot bypass route.

[0063] During this process, the data processing server also sets redirection flags for database read operations within the locked range. These redirection flags are access rules written into the underlying spatial database access layer. When the read target falls into a navigation-isolated node, the access layer does not return the current node geometry, but only a pointer to a historical snapshot slice, which is then read by the application publishing layer interface. With this processing, unclosed distorted polygons will not participate in front-end rendering and assembly, and the boundary of a node being reconstructed will not be mistakenly identified as a published result by the ship's or large-screen end.

[0064] For example, after the control panel establishes the locked area, it continuously performs scale zooming. If a read request covers the locked area and the ordinary navigation segments outside it, the application publishing layer interface will directly replace the local layer corresponding to the locked area with the previous frame's historical snapshot slice, continue reading the ordinary navigation segments outside from the underlying spatial database, and then stitch the two together in the same view and return it. When the ship's terminal drags the local view at the same time, it will still hit the historical snapshot slice. Therefore, the locked area seen by both types of terminals maintains the same complete boundary, and there will be no split situation where one end reads the intermediate state and the other end reads the old state.

[0065] S303: Adaptive geometric thinning reconstruction of waterway control lines in the navigation isolation state.

[0066] After the read path is taken over by the historical snapshot bypass route, the data processing server begins to perform geometric reconstruction on the target node subgraph in the navigation isolation state. Here, the geometric evolution envelope and spatial coordinate change sequence from step S1, and the target node subgraph from step S2 are directly referenced. The reconstruction object is limited to the node boundary sequence located within the geometric evolution envelope or sharing its boundary; nodes outside the locked range are not rewritten. The channel control line participates in the calculation for the first time in this step. The channel control line is a constraint polyline pre-stored in the spatiotemporal database to define the safe navigation boundary of the main channel.

[0067] The data processing server first extracts the boundary point sequence of the nodes to be reconstructed from the target node subgraph and organizes them into continuous boundary segments according to the node connectivity. Then, it performs overlay matching between each continuous boundary segment and the geometric evolution envelope, retaining only those continuous boundary segments that fall into or intersect with the boundary of the geometric evolution envelope as candidate boundary segments for reconstruction. Next, it projects the boundary points on each candidate boundary segment onto the corresponding channel control line, obtaining the projection order along the channel control line and the lateral distance of each boundary point relative to the channel control line. For each candidate boundary segment, the data processing server first fixes the shared edge endpoints, the intersection of the geometric evolution envelope, and the nearest projection point of the channel control line as retained points, and then performs constrained Douglas-Peucker recursive thinning between adjacent retained points. To ensure that the thinning decision has clear boundaries, the data processing server performs... Local thinning tolerance is set for each candidate boundary segment of the reconstruction. : , in, For the first The minimum lateral distance from the candidate boundary segment to the channel control line is derived from the projection results of each boundary point of the boundary segment onto the channel control line; For the first The minimum indentation distance from the candidate boundary segment to the geometric evolution envelope boundary is derived from the spatial relationship between the boundary segment and the geometric evolution envelope. The data processing server connects adjacent retained points to form candidate chords, calculates the maximum perpendicular distance from each candidate chord to each intermediate boundary point, and determines the maximum perpendicular distance if the maximum perpendicular distance is not greater than a certain value. If the candidate chord segment remains on the same side of the channel control line and is within the geometric evolution envelope, the intermediate boundary point of the interval is replaced in situ with the candidate chord segment; if any condition is not met, the boundary point corresponding to the maximum vertical distance is retained and the recursive segmentation continues until the entire reconstructed candidate boundary segment is thinned out.

[0068] After boundary thinning is completed, the data processing server writes the updated boundary point sequence back to the corresponding nodes in the target node subgraph, and synchronously corrects the boundary references of the nodes on both sides of the shared edge according to the original connected edge relationship, thus obtaining the reconstructed node spatial form. Since the reconstruction is always subject to the dual constraints of the channel control line and the geometric evolution envelope, the updated boundary will neither cross the channel control line nor exceed the local change range already defined by this incremental change. Subsequently, the data processing server retains the reconstructed node spatial form, the locked range, and the navigation isolation state in the pending release stage of the underlying spatial database, as the direct input for step S4 to perform data persistence writing, release asynchronous consistency topology lock, and generate multi-scale incremental rendering slices.

[0069] For example, when the incremental feature tensor of the survey data transmitted back from the ship shows that the inner boundary of the curve is rising towards the center of the channel, the data processing server will only extract the boundary segments that fall within the geometric evolution envelope, project them onto the channel control line of the curve, and then perform constrained Douglas-Peucker recursive thinning with the shared edge endpoints and the intersection of the geometric evolution envelope as the retention points. If a candidate chord segment will cut into the safe navigation area on one side of the channel control line, the candidate chord segment will not replace the original boundary point, but will continue to split into shorter boundary intervals; if the candidate chord segment is always located on the same side of the channel control line and within the geometric evolution envelope, the intermediate redundant boundary points are replaced, and finally the reconstructed node space shape that can be directly entered into the publishing and writing stage is obtained.

[0070] Through step S3, the data processing server has applied an asynchronous consistent topology lock to the target node subgraph, generated a locked range and written it into the navigation isolation state. At the same time, it has established a historical snapshot slice pointer and implemented bypass routing for cross-level read requests that fall into the locked range. Under this isolation condition, the reconstruction range is limited according to the geometric evolution envelope and the boundary thinning process is constrained by the channel control line. The adaptive geometric thinning reconstruction of the target node subgraph is completed, forming the reconstructed node space shape that can be directly called in step S4 and the locked range that is still in the release stage.

[0071] Step S3 has completed the adaptive geometric thinning reconstruction of the target node subgraph in the air traffic isolation state. At this time, the reconstructed node spatial morphology, locking range and historical snapshot bypass routes have clear boundaries, and the local update results have also left the intermediate state in the geometric calculation. However, these results are still in the state of waiting to be released and published. They have not yet been written into the underlying spatial database, nor have they been transformed into slice results that can be called by different display scales. Therefore, step S4 needs to complete the formal writing, publication switching and storage level organization on the current basis.

[0072] S401: Confirmation of persistent data writing and release phase.

[0073] Step S3 has updated the nodes to be reconstructed in the target node subgraph to the reconstructed node spatial form and kept the locked range in the waiting-to-release phase. Before publishing, the data processing server first writes the reconstruction results to the underlying spatial database, so that subsequent release of asynchronous consistency topology locks and slice generation are based on the node results already written to the database. The following sub-steps are all executed by the data processing server.

[0074] The data processing server first reads the target node subgraph, the locked range, the navigation isolation state, and the reconstructed node spatial form, and writes them back to the underlying spatial database one by one according to the node order in the target node subgraph. During the writing process, the data processing server synchronously updates the node boundary geometry, shared edge reference relationships, and the correspondence between nodes and connected edges for each node. Then, it performs a closure check on a unit of locked range, specifically checking the beginning and end closure relationships of each node boundary, whether the boundary references of nodes on both sides of a shared edge are consistent, and whether the connected edges inside the target node subgraph still maintain the connection direction as output in step S2. Only when all three types of results satisfy the consistency relationship will the data processing server register the current write result as a publishable version and keep the locked range in the waiting-to-release stage. If any check result does not satisfy the consistency relationship, the locked range is kept in the locked stage, the current publishable version registration is revoked, and the nodes that fail the closure check are sent back to step S303 for adaptive geometry thinning reconstruction until they pass the closure check again before entering the waiting-to-release stage.

[0075] In one embodiment, the target node subgraph corresponding to the uplift of a shallow shoal inside a bend contains multiple consecutive boundary nodes. When writing the reconstructed node spatial morphology, the data processing server first writes the boundary nodes closest to the channel control line, then synchronously corrects the adjacent nodes connected to them by shared edges, and finally checks whether the boundary loop is closed on a unit of the entire locked range. At this point, a publishable version corresponding to the target node subgraph has been formed in the underlying spatial database. Subsequent read path switching no longer depends on the geometric intermediate state, but directly targets the node results that have been persistently written.

[0076] S402: Release asynchronous consistency topology lock and generate output for multi-scale incremental rendering slices.

[0077] After the data persistence write is completed, the target node subgraph is ready to transition from the isolated computation phase to the release phase. The data processing server then processes the navigation isolation state and historical snapshot bypass routes left over from S3. At this point, the publishable version, locked range, and reconstructed node space shape confirmed by S401 are being referenced.

[0078] The data processing server first switches the locked range from the pending release phase to the switchable phase in the state machine, and releases the write lock of the underlying spatial database on the target node subgraph, so that the reconstructed node spatial form can be taken over by the publishing layer, while keeping the application publishing layer interface's historical snapshot bypass route for the locked range unchanged. Subsequently, the data processing server generates multi-scale incremental rendering slices with the locked range as the clipping boundary and the reconstructed node spatial form as the geometric basis. Specifically, it first reads the outer rectangle of the locked range and the spatial boundaries of the slice grids of each scale, and quickly filters out slice grids that are obviously separated from the locked range with the outer rectangle. Then, it performs precise overlap judgment on the slice grids that have not been filtered out. When the locked range completely falls inside a slice grid, the locked range crosses the boundary of a slice grid, the spatial range of a slice grid completely falls inside the locked range, or the boundary point or connected edge of the target node subgraph within the locked range falls within the spatial range of a slice grid, the corresponding slice is determined to enter the slice set to be updated. After determining the set of slices to be updated, the reconstructed node spatial morphology is projected onto the mesh boundaries of each slice. Segmented clipping is performed on node boundaries crossing slice boundaries, maintaining endpoint consistency for the same node at adjacent slice boundaries. New geometric content and slice index are then written to each slice. If adjacent slices simultaneously hit the same locked range boundary, these adjacent slices are retained together in the set of slices to be updated. After the new slice index and the publication address corresponding to the original historical snapshot slice pointer are replaced in situ, the data processing server switches the locked range to the released stage and clears the database redirection flags corresponding to the navigation isolation state and the locked range.

[0079] Once the new tile index is replaced, the application publishing layer interface switches its read path for the locked region from the historical snapshot bypass route to the multi-scale incremental rendering tile read path. For unupdated areas still outside the locked region, the application publishing layer interface retains the original tile index; for updated areas within the locked region, the application publishing layer interface directly outputs the new multi-scale incremental rendering tiles. This process ensures that external terminals can read the updated form consistent with the underlying spatial database on the next view refresh.

[0080] For example, after the scheduling center's large screen is updated to a medium-scale view, the application release layer interface hits the new multi-scale incremental rendering slice within the locked range, rather than the historical snapshot slice from the S3 stage. When the ship continues to zoom in to the local collision avoidance view, the fine-scale slice within the same locked range has also been replaced with the result generated based on the reconstructed node spatial morphology. Therefore, the macro view and the micro view see the boundary after the same round of persistent writing, and there is no longer a situation where the old snapshot and the new geometry are output in parallel.

[0081] S403: Query popularity distribution calculation and memory page swapping trigger.

[0082] After the target node subgraph is released and the slice output is completed, the local update has returned to the normal release chain. The data processing server further manages the memory residency range based on the query popularity distribution of the graph network nodes, ensuring that the updated graph network branches and the long-running global branches maintain the available memory boundary together. At this point, the referenced objects include the released target node subgraph, all graph network node branches, and access logs generated by the application release layer and ship activities.

[0083] The data processing server continuously collects two types of access sources: one is the spatial segment covered by each ship's AIS activity trajectory within the current statistical window, and the other is the spatial layer grid area covered by the zoom operation of the dispatch center's large screen view. Then, according to the spatial coverage relationship, each access is mapped to the corresponding graph network node branch, forming a query heat distribution. Here, a graph network node branch refers to a sequence of nodes in the spatiotemporal topology graph network that maintains a continuous access relationship through connected edges. The data processing server sorts the query heat distribution according to the number of times each graph network node branch is hit within the statistical window, the time of the most recent hit, and the duration of consecutive hits. Graph network node branches corresponding to hub segments with high rankings in the ranking results are registered as active branches, while graph network node branches corresponding to remote anchorages or seasonally interrupted navigation segments with query frequencies below a preset activity threshold are registered as inactive node branches. The preset activity threshold is determined by obtaining the lower quartile of the query frequency sequence of all graph network node branches in the most recent statistical period in ascending order, ensuring that the heat distribution directly originates from the current operating load.

[0084] For graph network node branches registered as resident branches, the data processing server maintains their node attributes, connected edge relationships, and latest slice index in a persistent state in the server's high-speed memory. For node branches registered as inactive, the data processing server initiates a background paging thread to first compress the topology node attributes of the node branch into sequential binary blocks, and then serializes and writes them to the solid-state drive in the order of the node branches. Simultaneously, only the node branch identifier, spatial range index, and disk offset address are retained in the high-speed memory. If the ship's AIS activity trajectory subsequently re-enters the coverage area of ​​the node branch, or if the dispatch center's large screen zooms in on the node branch again, the data processing server can reload the corresponding binary blocks back into the high-speed memory based on the retained disk offset address.

[0085] For example, in a scenario of continuous operation of the national water network, key sections of the Yangtze River main channel may be registered as inactive branches due to a large number of vessel AIS activity tracks and frequent large-screen scheduling queries. Conversely, if a seasonally interrupted branch is not hit within the statistical window for an extended period, it will be registered as an inactive node branch and transferred to the solid-state drive. Thus, after step S4, if the target node subgraph that was just released is in a high-frequency access segment, it will continue to reside in high-speed memory and participate in subsequent responses; if its segment becomes a low-frequency access segment in subsequent statistical windows, it will enter the serialization and transfer process according to the same paging rules.

[0086] Through step S4, the data processing server has written the reconstructed node spatial form into the underlying spatial database and registered it as a publishable version after the closure verification is passed. Then, it completes the generation of multi-scale incremental rendering slices and the replacement of slice indexes, so that the target node subgraph is removed from the temporary cache path and returns to the formal release path. At the same time, the query heat distribution formed around the graph network node branches has also been statistically analyzed and used for memory paging. The storage locations of resident branches and inactive node branches have been reorganized, and the local update results have now entered a stable release and normal storage state.

[0087] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-scale waterway surveying and mapping geographic information spatiotemporal database integrated modeling method, characterized in that, Including the following steps: S1: Collect heterogeneous observation signals from channel edge monitoring equipment within a preset time window, perform density-based spatiotemporal clustering to remove surface noise, and extract the mapping incremental feature tensor containing spatial coordinate change sequences and geometric evolution envelopes. S2: Receive the incremental feature tensor of the mapping and map it to the pre-constructed spatiotemporal topology graph network. Combine the hydrodynamic physical constraints of the target area to adjust the stopping threshold of the graph network edge weight-driven traversal algorithm and extract the target node subgraph corresponding to the incremental feature tensor of the mapping. S3: Apply an asynchronous consistent topology lock with historical snapshot bypass routing function to the target node subgraph, mark the target node subgraph as a navigation isolation state, and perform adaptive geometric thinning reconstruction of the target node subgraph in combination with the airway control line under the navigation isolation state; S4: Persistently write data to the target node subgraph after adaptive geometric thinning reconstruction, release the asynchronous consistency topology lock, generate multi-scale incremental rendering slices based on the reconstructed node spatial morphology, and trigger the memory paging mechanism based on the query heat distribution of graph network nodes.

2. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 1, characterized in that, In step S1, a unified timestamp is first added to the original point cloud of the multibeam sonar and the echo of the shore-based radar and converted to the same waterway mapping coordinate reference. The spatiotemporal observation sequence is formed in ascending order of the acquisition time. Then, density-based spatiotemporal clustering is performed to mark the observation points that do not meet the density reachability relationship as water surface noise and to merge the observation points that meet the continuous density support into a set of retained observation clusters.

3. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 2, characterized in that, In step S1, the geometric outer boundary coordinate set is extracted one by one for the retained observation cluster set, and the geometric outer boundary coordinate set confirmed in the previous preset time window is read from the local cache. The current geometric outer boundary coordinate set and the geometric outer boundary coordinate set of the previous preset time window are resampled and organized into a sequence of boundary points of the same length according to the same arc length. The coordinate difference is calculated point by point along the boundary ring direction to form a spatial coordinate change sequence and a geometric evolution envelope, and the two are encapsulated into a mapping incremental feature tensor.

4. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 3, characterized in that, In step S2, the geometric evolution envelope is projected onto the spatiotemporal topology graph network, and an overlay search is performed on the geometric evolution envelope and the spatial boundaries of each node. Nodes that overlap, cross, or cover the geometric evolution envelope are identified as seed node sets, and the spatial coordinate change sequence is written into the corresponding seed node according to the projection relationship of the nearest point of the boundary. The target region is determined by the connected region covered by the seed node set.

5. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 4, characterized in that, In step S2, the flow velocity and hydraulic radius of the corresponding section of the target area are extracted from the hydrodynamic model base to calculate the sediment carrying force. The bottom scattering intensity is analyzed on the mapping sonar echo corresponding to the target area to invert the median sediment particle size and calculate the bed shear resistance. The sediment carrying force and the bed shear resistance are then fed into the pre-trained radial basis neural network as two-dimensional real-time feature points to output the evolution sensitivity coefficient.

6. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 5, characterized in that, In step S2, the graph network edge weight-driven traversal algorithm is started with the seed node set as the starting point. The graph network edge weights that have been traversed on the diffusion path are accumulated as path energy dissipation. The stopping threshold of the traversal algorithm is corrected with the evolution sensitivity coefficient. When the path energy dissipation reaches the corrected stopping threshold, the current diffusion path is truncated. After all diffusion paths have completed the truncation judgment, the nodes included in the diffusion range and their connected edges are summarized to generate the target node subgraph.

7. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 6, characterized in that, In step S3, the set of nodes and the set of connected edges in the target node subgraph are read, the spatial range covered by the target node subgraph is registered as the locking range, and the locking range is managed through a state machine in the stages of pending locking, locked, and pending release. When there is a node intersection with the existing locked range, the locked range is updated according to the node union and spatial union. When there is no node intersection, a new locked range is established with the current target node subgraph, and all nodes in the target node subgraph are written into the navigation isolation state.

8. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 7, characterized in that, In step S3, the locking range is matched with the historical snapshot slices in the memory cache to establish a historical snapshot slice pointer. After the application publishing layer interface receives a cross-level read request, the read request that overlaps with the locking range is split into a locked sub-request and an unlocked sub-request. The historical snapshot slice corresponding to the locked sub-request is returned according to the historical snapshot slice pointer, and the current node data corresponding to the unlocked sub-request is returned according to the underlying spatial database. The data is then merged and output according to the original request scale and spatial splicing relationship.

9. The integrated modeling method for spatiotemporal database of multi-scale waterway surveying geographic information according to claim 8, characterized in that, In step S3, continuous boundary segments that fall into the geometric evolution envelope or intersect with the boundary of the geometric evolution envelope are extracted from the target node subgraph as candidate boundary segments for reconstruction. Each boundary point is projected onto the channel control line. The shared edge endpoints, the intersection points of the geometric evolution envelope, and the nearest projection point of the channel control line are fixed as retained points. Constrained Douglas-Peucker recursive thinning is performed between adjacent retained points. The updated boundary point sequence is written back to the target node subgraph to form the reconstructed node space morphology.

10. The integrated modeling method for spatiotemporal database of multi-scale waterway mapping geographic information according to claim 9, characterized in that, In step S4, the reconstructed node spatial morphology is written back to the underlying spatial database and a closure check is performed. After the closure check passes, it is registered as a publishable version. Then, the set of slices to be updated is determined based on the overlap relationship between the locked range and the slice grids of each scale. Multi-scale incremental rendering slices are generated and slice index replacement is completed. At the same time, the ship AIS activity trajectory and the scaling operation of the dispatch center large screen view are mapped to the graph network node branches to form a query heat distribution. Based on this, inactive node branches are serialized and transferred to disk media.

Images