Evolution analysis method and system for seabed erosion and deposition

Abstract

The invention relates to the technical field of evolution modeling, in particular to a seabed erosion and deposition evolution analysis method and system.The method comprises the steps that the multi-dimensional adjacency relation between nodes is recognized through a graph attention network, joint judgment is conducted on the gradient direction included angle, the elevation trend and the flow direction consistency index, and the influence degree is quantized; the expression capability of complex boundary direction influence factors is improved, direction judgment is not performed according to a fixed rule, and based on quantitative comparison of flow direction and gradient components, scouring dominant path nodes are screened and direction tracks of the scouring dominant path nodes are rearranged, so that sequential sequence expression has dynamic reconstruction capability; performing clustering screening on the angle abrupt change section by means of a greedy algorithm to form a node set representing an abrupt change trend, so as to enhance the extraction stability of the local extreme scouring behavior; and the connection logic expression of the physical state between the nodes, the abnormal extraction efficiency of local evolution change, the expression continuity of the erosion and deposition trend result and the overall identification stability are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of evolution modeling, and in particular to a method and system for analyzing the evolution of seabed scouring and silting. Background Art

[0002] The field of evolutionary modeling technology mainly focuses on the modeling of changing trends in time and space dimensions, dynamic process simulation and predictive analysis, and emphasizes the construction of mathematical models or numerical models to simulate the state variables in the actual environmental system in a time series manner, so as to grasp the development laws and evolution trends.

[0003] A seabed scouring and deposition evolution analysis method aims to achieve systematic modeling and computable prediction of the seabed morphological change process, so as to support the scientific assessment of the impact of engineering activities on the seabed environment, provide a quantifiable data basis and evolution trend judgment for marine engineering planning, channel maintenance, and ecological protection, and achieve structured, modeled, and visualized processing of complex natural scouring and deposition processes, thereby improving the ability to respond to dynamic changes.

[0004] Existing technologies have a strong temporal dependence when constructing dynamic process expressions. It is difficult to construct a directional spatial logical structure in a single time phase data through differential processing of continuous elevation snapshot data, resulting in insufficient ability to identify sudden changes in behavior and difficulty in accurately defining the boundaries of structural changes. In multi-node status judgment, a single parameter is usually used as a judgment benchmark, and a multi-dimensional joint judgment method has not been formed, which limits the identification of key trend directions in complex changing environments. Trajectory judgment is often constrained by preset evolution direction rules, and there is a lack of dynamic screening and direction rearrangement mechanisms, which can easily lead to node path reconstruction results being inconsistent with the actual erosion path. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method and system for analyzing the evolution of seabed scouring and silting.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for analyzing the evolution of seabed scouring and silting, comprising the following steps:

[0007] S1: Obtain single-phase seabed grid elevation data through sensors, calculate the elevation difference of the shoal and trough area and extract the slope vector, calculate the angle between adjacent units and screen the edge pairs that meet the direction conditions, and establish a directed connection weight network graph structure;

[0008] S2: Based on the directed connection weighted network graph structure, a graph attention network is used to extract the adjacency relationship of the bank slope boundary nodes, and to determine whether the slope direction, elevation increase and decrease trend, and flow direction consistency indicators all meet the threshold. By superimposing the influence of the direction clusters, a direction-dominant influence indicator group is generated;

[0009] S3: Based on the direction-dominant influencing index group, the nodes in the control area grid where the flow direction component is greater than the slope component are screened, the evolution path trajectory is extracted and the direction sequence is rearranged to establish a time series scouring and deposition change trend array;

[0010] S4: Based on the time series scouring and silting change trend array, the continuous angle change values ​​of the notch inflection points are called and the adjacent corner rate ratios are calculated. A greedy algorithm is used to screen the mutation segments and identify the scouring starting points. The extreme value sequence is clustered through the trajectory chain to generate the scouring and silting dominant path node set.

[0011] S5: Based on the set of nodes of the dominant scouring and deposition paths, determine the changing trend of the angle of the continuous path, intercept the monotonic segment and calculate the sequence of angle difference and length value, reorganize the path structure by segmented aggregation, and establish a partition map of the characteristic structure of scouring and deposition evolution.

[0012] As a further solution of the present invention, the specific steps of generating the directed connection weight network graph structure are:

[0013] The single-phase seabed grid elevation data is acquired through sensors. The elevation values ​​of each grid cell in the shoal and trough area are extracted point by point and the indexes of the four neighboring cells are extracted. The slope direction vector is generated by calculating the elevation difference between the current node and the four neighboring nodes and dividing it by the horizontal spacing, thus generating a local slope direction vector set.

[0014] Based on the local slope direction vector set, the cosine angle of the direction vectors between each pair of connected units is calculated and whether it is less than a predetermined angle threshold is determined. Grid unit pairs that meet the criteria for similar directions are selected and valid adjacency relationships are established to obtain a set of grid unit pairs with consistent directions.

[0015] Based on the set of grid unit pairs with consistent directions, connecting edges are constructed for each pair of nodes and direction attributes are added. Edge weights are generated by the inverse of the cosine angle and a connection matrix is ​​uniformly constructed to establish a directed connection weight network graph structure.

[0016] As a further solution of the present invention, the specific steps of generating the direction-dominant influence indicator group are:

[0017] Based on the directed connection weighted network graph structure, a graph attention network is used to screen the bank slope boundary nodes and extract all adjacent node indexes. The adjacent slope direction change values, the elevation increase and decrease differences between nodes, and the flow direction angle values ​​are extracted respectively. By constructing a set of three parameter sets, a bank slope node association index set is obtained.

[0018] Based on the slope node association indicator set, the three indicators of each group of nodes are judged item by item and the node pairs that meet all the threshold conditions are marked. The node pairs that meet the requirements are numbered and classified in sequence, and the corresponding connection path number set is extracted to obtain a high-consistency direction node set;

[0019] Based on the highly consistent directional node set, the inverse of the directional angle of each group of nodes is taken, the mean of the flow distance is taken, and the normalized value of the slope difference is taken and normalized and superimposed. Continuous path segments are aggregated and extracted through sequential numbering to generate a directional dominant influence indicator group.

[0020] As a further solution of the present invention, the graph attention network is based on the formula:

[0021]

[0022] Where: α′ ij Represents node v i For adjacent node v j The improved attention weights, represents the transpose of the weight vector in the attention mechanism, represents the feature transformation weight matrix, h i 、h j Represents node v i 、v j The characteristic vector of ij , elevation difference Δh ij , flow angle φ ij and slope roughness coefficient r j , || represents the vector concatenation operation, γ1 and γ2 represent the boundary stability factor η ij Difference factor λ from historical erosion trend ij The weight balance coefficient η ij represents the boundary stability factor between nodes, λ ij represents the historical erosion trend difference factor between nodes, Represents node v i The set of adjacent nodes of node v, k represents the number of nodes that are adjacent to node v during the normalization process. i The index of other adjacent nodes, v i 、v j Represents a node in the bank slope graph structure.

[0023] As a further solution of the present invention, the specific steps of generating the time series scouring and silting change trend array are:

[0024] Based on the directional dominant influencing index group, the modulus length of the flow direction value and the slope direction value of each node is extracted, and a comparison is performed through a two-value difference operation and a judgment condition is set that the flow direction component is greater than the slope component. Nodes that meet the conditions are screened to obtain a set of dominant flow direction nodes;

[0025] Based on the dominant flow direction node set, the nodes are connected in time series and the spatial distance and direction change angle of adjacent nodes are calculated according to the node index. The node index sequence is generated by rearranging the connection direction sequence to generate the scouring and silting trajectory direction sequence set;

[0026] Based on the scouring and silting trajectory direction sequence set, the time difference sequence of each node and the direction vector change value are merged, the direction change angle sequence of each node in the continuous path is sorted and cumulatively rearranged, and a time series scouring and silting change trend array is established.

[0027] As a further solution of the present invention, the specific steps of generating the scour and sedimentation dominant path node set are:

[0028] Based on the time series scouring and silting change trend array, continuous node direction vector groups are extracted and angle change values ​​are calculated in sequence. The angular rate sequence is generated by dividing the difference by the time step and the amplitude mutation judgment rule is set to obtain the trajectory direction mutation ratio sequence;

[0029] Based on the trajectory direction mutation ratio sequence, a greedy algorithm is used to perform mutation segment division operation, perform mutation point index positioning and extract the preceding and following adjacent point sequences, accumulate and sum continuous angle difference segments, select segments with mutation amplitude greater than the set value and record the path position index to obtain the scour mutation path segment set;

[0030] Based on the erosion mutation path segment set, the maximum value of the node flow direction angle set within the segment is located and it is determined whether there is an angle difference reversal point on both sides. The nodes where the reversal point is located are spatially reorganized and numbered to generate the erosion and sedimentation dominant path node set.

[0031] As a further solution of the present invention, the greedy algorithm is based on the formula:

[0032]

[0033] Where: T s represents the comprehensive mutation intensity value of the mutation fragment, ψ i represents the tangential direction angle of the i-th trajectory point, |ψ i+1 -ψ i | represents the direction angle difference between adjacent trajectory points, reflecting the degree of path turning, e i represents the seabed elevation value of the i-th trajectory point, d i,i+1 Represents the horizontal distance between trajectory points i and i+1 It represents the rate of change of elevation per unit horizontal distance between adjacent trajectory points, reflecting the sudden change of local slope. i represents the local curvature value at the i-th trajectory point, |κ i+1 -κ i | represents the curvature variation between adjacent points, si represents the unit path flow velocity at the i-th trajectory point, Represents the average flow velocity of all trajectory points in the current mutation segment, represents the point velocity offset value, μ1, μ2, μ3, and μ4 represent the weighted coefficients of directional mutation, elevation gradient, curvature fluctuation, and velocity offset factor, respectively.

[0034] As a further solution of the present invention, the specific steps of generating the scouring and silting evolution characteristic structure partition map are as follows:

[0035] Based on the erosion and sedimentation dominant path node set, the direction vectors of continuous nodes are extracted and the angle calculation between adjacent nodes is performed. The direction change pattern is identified through the positive and negative sign sequence of the angle difference, and the start and end positions of the monotonic continuous sequence are marked to generate a monotonic direction path segment set.

[0036] Based on the monotonic directional path segment set, the direction angle difference of each node pair and the distance value of the adjacent nodes are extracted in the order of the path segments, and an angle sequence and a length sequence are constructed respectively. The node numbers are grouped and combined with the codes to obtain a path angle length parameter sequence set;

[0037] Based on the path angle length parameter sequence set, the angle sequence fluctuation values ​​and length spacing change values ​​are extracted according to the node numbers within the segment. The structural segment numbers are rearranged after classification based on the value difference range within continuous segments, and the paths of the same type are merged and collected to establish a characteristic structural zoning map of scouring and deposition evolution.

[0038] A seabed scouring and silting evolution analysis system is provided, wherein the seabed scouring and silting evolution analysis system is used to execute the above-mentioned seabed scouring and silting evolution analysis method, and the system comprises:

[0039] Elevation analysis module: Based on single-phase seabed grid elevation data, the elevation values ​​of adjacent grid cells in the shoal and channel area are extracted, the elevation difference and slope vector direction are calculated pair by pair, the slope vector angle is obtained and whether it falls within the set angle range is determined, the unidirectional edge pairs are selected, and the elevation difference and slope direction are used as edge weight values ​​to establish a directional edge weight graph structure;

[0040] Edge Mapping Module: Based on the directional edge weight graph structure, a graph attention network is used to extract directional edge relationships from the adjacent node set of each node. The slope direction angle value, elevation increase and decrease continuity mark, and flow direction consistency ratio of each connecting edge are judged to see whether they meet the corresponding thresholds. All directional edge groups that meet the conditions are selected, and the cumulative connection frequency of directional clusters in the graph is counted. An indicator set is generated based on the frequency to form a directional dominant influence indicator group.

[0041] Dominant direction module: Based on the direction-dominant influence index group, the flow direction component value and slope component value of each grid node in the control area are extracted, and the node numbers of all nodes whose flow direction components are greater than the slope components are screened. The flow direction paths of each numbered node in multiple time periods are read, and the flow direction sequences are reorganized after sorting. The structural change trends of all sequences in the time dimension are summarized to establish a direction evolution path sequence group;

[0042] Path evolution module: Based on the directional evolution path sequence group, the turning point numbers of the slots in each path are extracted, the angle difference between each pair of turning points and the ratio of the continuous turning angle change rate are calculated, and whether the change rate ratio exceeds a threshold is determined. The first node of the path mutation section is marked as the scour starting point, all extreme points are read to form a path chain, and a greedy algorithm is used to cluster the path extreme nodes. The cluster center position is output to form a scour and silt distribution node set;

[0043] Structural division module: Based on the scour and sedimentation distribution node set, determine whether the change direction of the angle between adjacent nodes in the path segment is consistent, screen out all monotonous angle path segments, read the angle difference sequence and path length numerical sequence in each path segment respectively, remap the path structure order according to the combined value of the angle and length, aggregate all segment paths to form a path segment set, and establish a scour and sedimentation evolution partition map.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are:

[0045] In this paper, the multi-dimensional adjacency relationship between nodes is identified through the graph attention network, and the three parameters of slope direction angle, elevation trend, and flow consistency index are jointly judged and the degree of influence is quantified, thereby improving the ability to express the influencing factors of complex boundary directions;

[0046] In the present invention, instead of using fixed rules to judge the direction, the scour dominant path nodes are screened and their direction trajectories are rearranged based on the quantitative comparison of flow direction and slope components, so that the time series expression has the ability of dynamic reconstruction;

[0047] In the present invention, the calculation of the notch angle rate ratio is combined with the greedy algorithm to cluster and screen the angle mutation segments to form a node set representing the mutation trend, thereby enhancing the stability of the extraction of local extreme scour behavior.

[0048] In the present invention, path reconstruction is achieved by combining the angle sequence with the path length value, and a structural map that can identify evolution laws is constructed, which effectively improves the connection logic expression of the physical state between nodes, the efficiency of extracting anomalies of local evolution changes, and the expression continuity and overall recognition stability of the scouring and silting trend results. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0050] Figure 2 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0052] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0053] Example 1

[0054] See also Figure 1 The present invention provides a technical solution: a method for analyzing the evolution of seabed scouring and silting, comprising the following steps:

[0055] S1: Obtain single-phase seabed grid elevation data through sensors, calculate the elevation difference of the shoal and trough area and extract the slope vector, calculate the angle between adjacent units and screen the edge pairs that meet the direction conditions, and establish a directed connection weight network graph structure;

[0056] S2: Based on the directed connection weighted network graph structure, a graph attention network is used to extract the adjacency relationship of the bank slope boundary nodes, determine whether the slope direction, elevation increase and decrease trend, and flow direction consistency indicators all meet the threshold, and generate a direction-dominant influence indicator group by superimposing the influence of direction clusters;

[0057] S3: Based on the direction-dominant influencing index group, the nodes in the control area where the flow direction component is greater than the slope component are screened, the evolution path trajectory is extracted and the direction sequence is rearranged to establish a temporal scouring and deposition change trend array;

[0058] S4: Based on the time series scouring and deposition change trend array, the continuous angle change values ​​of the slot inflection points are called and the adjacent corner rate ratios are calculated. A greedy algorithm is used to screen the mutation segments and identify the scouring starting points. The extreme value sequence is clustered through the trajectory chain to generate the scouring and deposition dominant path node set.

[0059] S5: Based on the set of nodes of the dominant paths of erosion and deposition, the changing trend of the angle of the continuous path is determined, the monotonic segments are intercepted and the angle difference and length value sequence are calculated, the path structure is reorganized by segmented aggregation, and a partition map of the characteristic structure of erosion and deposition evolution is established.

[0060] The specific steps to generate the directed connection weight network graph structure are:

[0061] The single-phase seabed grid elevation data is acquired through sensors. The elevation values ​​of each grid cell in the shoal and trough area are extracted point by point and the indexes of the four neighboring cells are extracted. The slope direction vector is generated by calculating the elevation difference between the current node and the four neighboring nodes and dividing it by the horizontal spacing, thus generating a local slope direction vector set.

[0062] Based on the local slope direction vector set, the cosine angle between each pair of connected cells is calculated and judged to be less than the specified angle threshold. By screening out pairs of grid cells that meet the criteria of similar directions and establishing effective adjacency relationships, a set of grid cell pairs with consistent directions is obtained.

[0063] Based on a set of grid cell pairs with consistent directions, a connection edge is constructed for each pair of nodes and a direction attribute is added. The edge weight is generated by the inverse of the cosine angle and a connection matrix is ​​uniformly constructed to establish a directed connection weight network graph structure.

[0064] Based on the single-phase seabed grid elevation data acquired by sensors, a two-dimensional index mapping method is used to assign row and column numbers to each grid cell in the shoal channel area. After extracting the number value of each cell, the index of the four neighboring cells is constructed. The elevation values ​​of the central cell and the four neighboring cells are read, and the elevation numerical difference operation is performed in sequence. The obtained difference is divided by the fixed horizontal spacing of 2 meters item by item to obtain the slope component value set. The four directional unit vectors are extracted according to the azimuth setting. Each directional unit vector is multiplied by the corresponding component value dimensionally and then summed. The coordinate components of the directional vector are obtained and normalized to generate a local slope direction vector set.

[0065] Based on the local slope direction vector set, the vector cosine calculation method is used to calculate the directional relationship between adjacent grid cells. The three-dimensional coordinate components of each pair of direction vectors are read, and the multiplication and summation operations of the dimensional components are performed in sequence. The square sums of the dimensional components of the two direction vectors are respectively calculated and then square rooted. The obtained values ​​are combined, ratioed, and inverse trigonometric functions are performed to obtain the angle values. The obtained angle values ​​are uniformly judged to be less than the set threshold of 20 degrees. All number combinations that meet the conditions are extracted and written into the valid unit pair record table to obtain a set of grid unit pairs with consistent directions.

[0066] Based on a set of grid cell pairs with consistent directions, connecting edges are constructed for each pair of nodes and directional attributes are assigned. The inverse weighted function method is used to read the directional angle value corresponding to each group of nodes. The inverse processing of the angle value is performed in sequence to form an array of initial weight values. All edge weight values ​​are normalized and standardized with a unified length unit. The direction is set from the node with a smaller index value to the node with a larger index value. When establishing the edge connection relationship, a direction label is attached and the original node number is retained. After constructing a record table of all directed edges, the node set, edge set and weight set contents are merged and written into the connection matrix array to establish a directed connection weight network graph structure.

[0067] The specific steps for generating the direction-dominant impact indicator group are as follows:

[0068] Based on the directed connection weighted network graph structure, a graph attention network is used to screen the bank slope boundary nodes and extract the indexes of all adjacent nodes. The adjacent slope direction change values, the elevation increase and decrease differences between nodes, and the flow direction angle values ​​are extracted respectively. By constructing a set of three parameter sets, a set of bank slope node association indicators is obtained.

[0069] Based on the slope node correlation index set, the three indicators of each node group are judged item by item and the node pairs that meet all the threshold conditions are marked. The node pairs that meet the requirements are numbered and classified in sequence, and the corresponding connection path number set is extracted to obtain the high-consistency direction node set;

[0070] Based on the set of highly consistent directional nodes, the inverse of the direction angle, the mean of the flow distance, and the normalized value of the slope difference of each group of nodes are taken and normalized and superimposed. Continuous path segments are aggregated and extracted through sequential numbering to generate a group of directional dominant influence indicators.

[0071] Based on the directed connection weight network graph structure, a graph attention network is used to screen boundary nodes. Specifically, the multi-head attention mechanism is initialized for all nodes in the graph, with the number of heads set to 4, the input dimension to 64, and the output dimension to 16. The eigenvalue of each node is linearly transformed with the eigenvalue corresponding to the index of its adjacent node. The LeakyReLU activation function is called to activate the transformation result, and the negative slope parameter is set to 0.2. The softmax normalization operation is performed on the activation result to obtain the attention coefficient matrix. The matrix is ​​used to perform item-by-item weighted processing on the slope direction change value, elevation difference, and flow direction angle value corresponding to the connecting edges between nodes, and the values ​​are written into the standardized parameter table. Three structured parameter groups are constructed to generate a set of bank slope node association indicators.

[0072] Based on the slope node association index set, threshold judgment operations are performed on the three parameters in each group of nodes. The slope direction change threshold is set to 15 degrees, the elevation difference threshold is set to 0.5 meters, and the flow angle threshold is set to 30 degrees. Each value in the parameter set is read group by group and compared one by one with the above three thresholds. If all three parameters meet the conditions less than or equal to the threshold, the corresponding node pair is marked as 1, otherwise it is marked as 0. A number mapping operation is performed on all node pairs marked as 1 and the results are written to the path mark set. The output list is written in the order of the node pair combination path number to obtain a high-consistency direction node set.

[0073] Based on the highly consistent directional node set, the corresponding directional angle value is read for each pair of node numbers and the inverse is taken. The flow distance value between the nodes is read and two average operations are performed. The node slope value is read and the two-value difference calculation is performed, and then normalized using the MinMaxScaler method. The three values ​​are directly numerically superimposed to obtain the combined impact value. For all node combinations, the continuous path number set is extracted in the order of numbering. The number intervals are merged in the paragraph aggregation method and written into the node sequence output table to generate a directional dominant impact indicator group.

[0074] Graph attention network, according to the formula:

[0075]

[0076] Where: α′ ij Represents node v i For adjacent node v j The improved attention weights, represents the transpose of the weight vector in the attention mechanism, represents the feature transformation weight matrix, h i 、h j Represents node v i 、v j The characteristic vector of ij , elevation difference Δh ij , flow angle φ ij and slope roughness coefficient r j , || represents the vector concatenation operation, γ1 and γ2 represent the boundary stability factor η ij Difference factor λ from historical erosion trend ij The weight balance coefficient η ij represents the boundary stability factor between nodes, λ ij represents the historical erosion trend difference factor between nodes, Represents node v i The set of adjacent nodes of node v, k represents the number of nodes that are adjacent to node v during the normalization process. i The index of other adjacent nodes, vi 、v j Represents a node in the bank slope graph structure;

[0077] Execution process: First, based on the screening results of the slope boundary nodes, extract each target node v i and all adjacent nodes v j The basic eigenvector h i 、h j , the feature content includes the direction change value θ of the node connection edge ij , the elevation difference Δh is obtained by calculating the angle between the line connecting the nodes and the reference coastline ij , obtained by calculating the difference in seabed elevation points, the flow angle φ ij , based on the tidal simulation data, the flow vector angle and the slope roughness coefficient r are calculated. j , calculated based on the measured terrain surface roughness or multi-source remote sensing inversion parameters, and then for each pair of nodes v i 、v j Two innovative structural parameters are introduced to improve the structural expressiveness of erosion and deposition analysis, namely the boundary stability factor η ij and the historical erosion trend difference factor λ ij , where η ij Through the block geometric stability analysis model calculation, the slope of the bank block, hydrological permeability coefficient, rock shear strength and other indicators are comprehensively considered for weighted scoring, and after normalization, it is used as an additional factor to participate in the attention score, λ ij Based on the historical three-year seabed elevation change data, a sliding time window is used to extract the regional elevation mean time series, and the mean square error between the adjacent node regional series is further calculated as the indicator of scouring and deposition trend difference. Finally, all feature vectors are spliced ​​in sequence and input into the attention calculation formula. During the splicing process, η ij and λ ij Multiply them by the corresponding balance coefficients γ1 and γ2 respectively. The two coefficients are optimized by cross-validation on the training dataset with the goal of minimizing the attention output error. After completing the linear transformation and activation of the splicing vector, the final attention weight value α′ is normalized. ij , which is used to express the coupling strength between nodes in the seabed erosion and deposition evolution graph model, thereby guiding subsequent analysis processes such as node embedding, evolution pattern clustering and slope stability prediction.

[0078] The specific steps for generating a time series erosion and deposition change trend array are as follows:

[0079] Based on the directional dominant influencing index group, the modulus length of the flow direction value and the slope direction value of each node is extracted. The two-value difference operation is used to compare and set the judgment condition that the flow component is greater than the slope component. The nodes that meet the conditions are screened to obtain the dominant flow direction node set.

[0080] Based on the set of dominant flow direction nodes, the nodes are connected in time series and the spatial distance and direction change angle of adjacent nodes are calculated according to the node index. The node index sequence is generated by rearranging the connection direction sequence, and the scouring and deposition trajectory direction sequence set is generated.

[0081] Based on the scouring and silting trajectory direction sequence set, the time difference sequence of each node and the direction vector change value are merged, and the direction change angle sequence of each node in the continuous path is sorted and cumulatively rearranged to establish a time series scouring and silting change trend array;

[0082] Based on the directional dominant influencing index group, the vector modulus difference method is used to extract the flow direction value and slope direction value of each node. Specifically, the three-dimensional flow direction vector and the three-dimensional slope direction vector are read from each node, the square sum of the three-dimensional components is calculated and the square root is taken as the modulus value. The flow direction modulus and the slope direction modulus are subtracted and stored in the difference array. A logical judgment is performed on all values ​​in the difference array that are greater than zero and a Boolean index is returned. The record numbers of the nodes with Boolean values ​​of True are extracted in the original index order to generate the dominant flow direction node set.

[0083] Based on the set of dominant flow direction nodes, the node index and timestamp number are read using the sequence interpolation connection method. All nodes are sorted in ascending chronological order. The two-dimensional coordinate information of adjacent nodes is extracted and Euclidean distance calculation operations are performed in sequence. The cosine value of the vector direction of two adjacent nodes is processed and the inverse trigonometric function is performed to obtain the angle value. The time number and angle sequence array is established. The original node index table is then rebuilt through the ascending sequence and combined to generate a direction chain. All chains are aggregated and merged into a set of segment paths according to the node number to generate a sequence set of scouring and deposition trajectory directions.

[0084] Based on the direction sequence set of scouring and silting trajectories, the path sequence stacking method is used to extract the time numbers of adjacent nodes in each group of paths and perform pairwise subtraction operations to write them into the time difference array. The direction vector coordinate value set between the corresponding nodes is read and differential processing is performed to obtain the direction change vector. The square root operation of each dimension of the change vector is then performed to generate a direction change value sequence. The time difference array and the direction change value sequence are stacked row-wise in the path dimension. The direction change angle of each group of paths is extracted and sorted in ascending order according to the path sequence. The sorted sequence is read and an item-by-item addition operation is performed from the beginning to the end to generate the cumulative change value. Finally, all path numbers and change sequences are combined to establish a time series scouring and silting change trend array.

[0085] The specific steps for generating the erosion and deposition dominant path node set are as follows:

[0086] Based on the time series scouring and silting change trend array, the continuous node direction vector group is extracted and the angle change value is calculated in sequence. The angular rate sequence is generated by dividing the difference by the time step, and the amplitude mutation judgment rule is set to obtain the trajectory direction mutation ratio sequence.

[0087] Based on the trajectory direction mutation ratio sequence, a greedy algorithm is used to perform mutation segment division, locate the mutation point index and extract the adjacent point sequence. By accumulating and summing the continuous angle difference segments, the segments with mutation amplitude greater than the set value are selected and the path position index is recorded to obtain the scour mutation path segment set.

[0088] Based on the erosion mutation path segment set, the maximum value of the node flow angle set within the segment is located and whether there is an angle difference reversal point on both sides is determined. The nodes where the reversal point is located are spatially reorganized and numbered to generate the erosion and sedimentation dominant path node set.

[0089] Based on the time series scouring and silting change trend array, the direction vector group of continuous nodes in the path is extracted using the direction vector angle difference calculation method. The direction vector values ​​of each pair of adjacent nodes are read in turn, and the three-dimensional coordinate components are used to perform a dimension-by-dimension product operation and then accumulated. The modulus value of each vector is read and divided to obtain the cosine value of the angle. The inverse trigonometric function is then performed to convert it into an angle value to generate a sequence of angle change values. Subsequently, the corresponding time number of the node is read and the difference between adjacent time numbers is calculated as the time step. Each group of angle change values ​​is divided by the corresponding time step and written into the angular rate sequence array. The angular rate judgment threshold is set to 10. It is judged whether the adjacent rate difference exceeds the threshold, and the node number that meets the conditions is output to obtain the trajectory direction mutation ratio sequence.

[0090] Based on the trajectory direction mutation ratio sequence, a greedy algorithm is used to perform the mutation segment division operation. Each mutation ratio in the sequence is read and the corresponding node index position is recorded in the order of the path. The initial cluster starting point index is set as the first mutation node number of the path. From the current position, one node index is moved backward each time and the current cluster cumulative angle difference is recorded. If the cumulative value exceeds the threshold of 30, the process ends and the start and end node indexes are saved. The last node is then used as the new starting point to continue the above process. The mutation segment division and index extraction of all path segments are performed in sequence. The set of all mutation segment numbers and the node combination path index table are recorded to obtain the scour mutation path fragment set.

[0091] Based on the erosion mutation path segment set, the directional extreme value judgment method is used to read all node index values ​​in each segment, extract the flow direction angle set of the corresponding node, perform a maximum value search on each set of angles, extract the node number where the maximum angle value is located, and then read the angle values ​​of the three adjacent nodes on the left and right of the maximum corner point in turn and construct a difference sequence. It is determined whether the adjacent difference signs have opposite signs. If there is an alternating positive and negative change, it is marked as a reversal point. All reversal points are numbered and uniformly coded, and the relative path segment position is recorded. The two-dimensional coordinate positions of adjacent reversal points are extracted and rearranged according to the numbering order, written into the node sequence table, and the erosion and sedimentation dominant path node set is generated.

[0092] Greedy algorithm, according to the formula:

[0093]

[0094] Where: T s represents the comprehensive mutation intensity value of the mutation fragment, ψ i represents the tangential direction angle of the i-th trajectory point, |ψ i+1 -ψ i | represents the direction angle difference between adjacent trajectory points, reflecting the degree of path turning, e i represents the seabed elevation value of the i-th trajectory point, d i,i+1 Represents the horizontal distance between trajectory points i and i+1 It represents the rate of change of elevation per unit horizontal distance between adjacent trajectory points, reflecting the sudden change of local slope. i represents the local curvature value at the i-th trajectory point, |κ i+1 -κ i | represents the curvature variation between adjacent points, s i represents the unit path flow velocity at the i-th trajectory point, Represents the average flow velocity of all trajectory points in the current mutation segment, represents the point velocity offset value, μ1, μ2, μ3, and μ4 represent the weighting coefficients of directional mutation, elevation gradient, curvature fluctuation, and velocity offset factor, respectively;

[0095] Execution process: First, the potential mutation area is preliminarily identified from the trajectory direction mutation ratio sequence, and the start and end indexes p and q of each mutation segment are determined. Then, for each segment, the required parameter information is extracted point by point, and the tangential direction angle ψ of the adjacent trajectory points is calculated. i , and then get the directional mutation value |ψ i+1 -ψ i |, used to characterize the degree of path turning mutation and extract the seabed elevation value e of the trajectory point i , calculate the elevation difference and the corresponding horizontal distance d i,i+1 Ratio As the local slope change rate of the segment, the fitting circle formed by the local three points is extracted to calculate the local curvature value κ of the trajectory point i , obtain the curvature mutation |κ i+1 -κ i |, used to measure the severity of path bending and read the unit path flow velocity s in hydrodynamic simulation or observation results i , calculate the average flow velocity of the current segment Offset Reflecting the severe scour area that may be caused by local velocity anomalies, the above four features are finally multiplied by the corresponding weighting coefficients μ1, μ2, μ3, and μ4, and the improved mutation intensity value T is obtained by accumulating the entire segment. s , each weight coefficient is determined by the parameter adjustment method based on the best accuracy of the validation set. Through grid search combined with cross-validation strategy, the coefficient combination that performs best in the mutation fragment identification task is selected to enhance the model's comprehensive response ability to direction, slope, structure and dynamic anomalies, thereby accurately extracting the scour mutation path fragment set.

[0096] The specific steps for generating the characteristic structural zoning map of erosion and deposition evolution are as follows:

[0097] Based on the erosion and deposition dominant path node set, the direction vectors of continuous nodes are extracted and the angle between adjacent nodes is calculated. The direction change pattern is identified by the positive and negative sign sequence of the angle difference, and the start and end positions of the monotonic continuous sequence are marked to generate a monotonic direction path segment set.

[0098] Based on the monotonic directional path segment set, the direction angle difference of each node pair and the distance value of adjacent nodes are extracted in the order of the path segments, and the angle sequence and length sequence are constructed respectively. The codes are merged and processed by grouping the node numbers to obtain the path angle length parameter sequence set;

[0099] Based on the path angle length parameter sequence set, the angle sequence fluctuation value and length interval change value are extracted according to the node number within the segment. The structural segment numbers are rearranged after the value difference range within the continuous segment is classified. The paths of the same type are merged and pooled to establish the characteristic structural zoning map of scouring and deposition evolution.

[0100] Based on the erosion and deposition-dominated path node set, a symbol sequence recognition method is used to extract the direction vectors of consecutive nodes in the path. The three-dimensional direction vector coordinates of each node and the next node are read separately, and the angle between adjacent direction vectors is calculated. The angle extraction operation is completed by combining the dot product of the direction vectors with the modulus value. All angle differences are recorded in the order of path segments. The sign of each angle difference is read and converted to a ±1 mark. A direction change symbol sequence is constructed. A sliding window is used to traverse the sequence item by item and find the continuous symbol consistent interval. The starting node number and ending node number of each consistent interval are marked. All marked segment number lists are merged to generate a monotonic direction path segment set.

[0101] Based on the monotone directional path segment set, the path parameter decomposition method is used to read the node number pair set for each path segment. The direction vector is extracted for each pair of nodes and the angle difference is calculated. At the same time, the spatial coordinates of the corresponding nodes are read and the Euclidean distance calculation operation is performed to generate an angle value sequence and a distance value sequence. Each item in the two sequences is bound to the corresponding node number pair to form a path parameter item. A path segment dictionary structure is established based on the path segment number as the grouping basis. A key-value mapping relationship between angle and distance is constructed for each group of path segments and encoded into a unified structured table. The encoding result is output and written into a number item list to obtain the path angle length parameter sequence set.

[0102] Based on the path angle length parameter sequence set, the interval classification aggregation method is used to extract all angle change values ​​and corresponding length values ​​for each path segment in the order of the node numbers within the segment. The range of each group of angle change values ​​is counted and the range level is recorded. Then, the length difference calculation results between all adjacent nodes are read and classified into five-segment grade intervals. A unique category identifier is constructed according to the combination of the angle range level and the length interval level. The category grouping operation is performed on all path segment numbers and the numbers are sorted from small to large. The segment numbers of the classified path set are rearranged and the partition index is completed. The data are written into the structure map index table to establish the structural partition map of the scouring and deposition evolution characteristics.

[0103] See also Figure 2 A seabed scouring and silting evolution analysis system is provided. The seabed scouring and silting evolution analysis system is used to execute the above-mentioned seabed scouring and silting evolution analysis method. The system comprises:

[0104] Elevation analysis module: Based on single-phase seabed grid elevation data, the elevation values ​​of adjacent grid cells in the shoal and channel area are extracted, the elevation difference and slope vector direction are calculated pair by pair, the slope vector angle is obtained and whether it falls within the set angle range is determined, the unidirectional edge pairs are selected, and the elevation difference and slope direction are used as edge weight values ​​to establish a directional edge weight graph structure;

[0105] Edge Mapping Module: Based on the directional edge weight graph structure, a graph attention network is used to extract directional edge relationships from the adjacent node set of each node. The slope direction angle value, elevation increase and decrease continuity mark, and flow direction consistency ratio of each connecting edge are judged to see if they meet the corresponding thresholds. All directional edge groups that meet the conditions are selected, and the cumulative connection frequency of directional clusters in the graph is counted. An indicator set is generated based on the frequency to form a directional dominant influence indicator group.

[0106] Dominant direction module: Based on the direction-dominant influence index group, the flow direction component value and slope component value of each grid node in the control area are extracted, and the node numbers with flow direction components greater than slope components are screened. The flow path of each numbered node in multiple time periods is read, and the flow direction sequence is reorganized after sorting. The structural change trends of all sequences in the time dimension are summarized to establish a direction evolution path sequence group;

[0107] Path evolution module: Based on the directional evolution path sequence group, the turning point numbers of the slots in each path are extracted, the angle difference between each pair of turning points and the ratio of the continuous turning angle change rate are calculated, and the ratio of the change rate is determined to determine whether it exceeds the threshold. The first node of the path mutation section is marked as the scour starting point, all extreme points are read to form a path chain, and a greedy algorithm is used to cluster the path extreme nodes. The cluster center position is output to form a node set for the scour and silt distribution.

[0108] Structural division module: Based on the node set of scour and sedimentation distribution, determine whether the direction of change of the angle between adjacent nodes in the path segment is consistent, screen out all monotonous angle path segments, read the angle difference sequence and path length numerical sequence in each path segment respectively, remap the path structure order according to the combined value of the angle and length, aggregate all segment paths to form a path segment set, and establish a scour and sedimentation evolution partition map.

[0109] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for analyzing the evolution of seabed erosion and deposition, characterized in that: The following steps are involved: S1: Obtain single-phase seabed grid elevation data through sensors, calculate the elevation difference of the shoal and trough area and extract the slope vector, calculate the angle between adjacent units and screen the edge pairs that meet the direction conditions, and establish a directed connection weight network graph structure; S2: Based on the directed connection weighted network graph structure, a graph attention network is used to extract the adjacency relationship of the bank slope boundary nodes, and to determine whether the slope direction, elevation increase and decrease trend, and flow direction consistency indicators all meet the threshold. By superimposing the influence of the direction clusters, a direction-dominant influence indicator group is generated; S3: Based on the direction-dominant influencing index group, the nodes in the control area grid where the flow direction component is greater than the slope component are screened, the evolution path trajectory is extracted and the direction sequence is rearranged to establish a time series scouring and deposition change trend array; S4: Based on the time series scouring and silting change trend array, the continuous angle change values ​​of the notch inflection points are called and the adjacent corner rate ratios are calculated. A greedy algorithm is used to screen the mutation segments and identify the scouring starting points. The extreme value sequence is clustered through the trajectory chain to generate the scouring and silting dominant path node set. S5: Based on the set of nodes of the dominant scouring and deposition paths, determine the changing trend of the angle of the continuous path, intercept the monotonic segment and calculate the sequence of angle difference and length value, reorganize the path structure by segmented aggregation, and establish a partition map of the characteristic structure of scouring and deposition evolution.

2. The method for analyzing the evolution of seabed scouring and silting according to claim 1, characterized in that: The specific steps of generating the directed connection weight network graph structure are: The single-phase seabed grid elevation data is acquired through sensors. The elevation values ​​of each grid cell in the shoal and trough area are extracted point by point and the indexes of the four neighboring cells are extracted. The slope direction vector is generated by calculating the elevation difference between the current node and the four neighboring nodes and dividing it by the horizontal spacing, thus generating a local slope direction vector set. Based on the local slope direction vector set, the cosine angle of the direction vectors between each pair of connected units is calculated and whether it is less than a predetermined angle threshold is determined. Grid unit pairs that meet the criteria for similar directions are selected and valid adjacency relationships are established to obtain a set of grid unit pairs with consistent directions. Based on the set of grid unit pairs with consistent directions, connecting edges are constructed for each pair of nodes and direction attributes are added. Edge weights are generated by the inverse of the cosine angle and a connection matrix is ​​uniformly constructed to establish a directed connection weight network graph structure.

3. The method for analyzing the evolution of seabed scouring and silting according to claim 1, characterized in that: The specific steps for generating the direction-dominant impact indicator group are as follows: Based on the directed connection weighted network graph structure, a graph attention network is used to screen the bank slope boundary nodes and extract all adjacent node indexes. The adjacent slope direction change values, the elevation increase and decrease differences between nodes, and the flow direction angle values ​​are extracted respectively. By constructing a set of three parameter sets, a bank slope node association index set is obtained. Based on the slope node association indicator set, the three indicators of each group of nodes are judged item by item and the node pairs that meet all the threshold conditions are marked. The node pairs that meet the requirements are numbered and classified in sequence, and the corresponding connection path number set is extracted to obtain a high-consistency direction node set; Based on the highly consistent directional node set, the inverse of the directional angle of each group of nodes is taken, the mean of the flow distance is taken, and the normalized value of the slope difference is taken and normalized and superimposed. Continuous path segments are aggregated and extracted through sequential numbering to generate a directional dominant influence indicator group.

4. The method for analyzing the evolution of seabed scouring and silting according to claim 3, characterized in that: The graph attention network, according to the formula: Where: α′ ij Represents node v i For adjacent node v j The improved attention weights, represents the transpose of the weight vector in the attention mechanism, represents the feature transformation weight matrix, h i 、h j Represents node v i 、v j The characteristic vector of ij , elevation difference Δh ij , flow angle φ ij and slope roughness coefficient r j , || represents the vector concatenation operation, γ1 and γ2 represent the boundary stability factor η ij Difference factor λ from historical erosion trend ij The weight balance coefficient η ij represents the boundary stability factor between nodes, λ ij represents the historical erosion trend difference factor between nodes, Represents node v i The set of adjacent nodes of node v, k represents the number of nodes that are adjacent to node v during the normalization process. i The index of other adjacent nodes, v i 、v j Represents a node in the bank slope graph structure.

5. The method for analyzing the evolution of seabed scouring and silting according to claim 1, characterized in that: The specific steps of generating the time series scouring and silting change trend array are as follows: Based on the directional dominant influencing index group, the modulus length of the flow direction value and the slope direction value of each node is extracted, and a comparison is performed through a two-value difference operation and a judgment condition is set that the flow direction component is greater than the slope component. Nodes that meet the conditions are screened to obtain a set of dominant flow direction nodes; Based on the dominant flow direction node set, the nodes are connected in time series and the spatial distance and direction change angle of adjacent nodes are calculated according to the node index. The node index sequence is generated by rearranging the connection direction sequence to generate the scouring and silting trajectory direction sequence set; Based on the scouring and silting trajectory direction sequence set, the time difference sequence of each node and the direction vector change value are merged, the direction change angle sequence of each node in the continuous path is sorted and cumulatively rearranged, and a time series scouring and silting change trend array is established.

6. The method for analyzing the evolution of seabed scouring and silting according to claim 1, characterized in that: The specific steps for generating the erosion and deposition dominant path node set are as follows: Based on the time series scouring and silting change trend array, continuous node direction vector groups are extracted and angle change values ​​are calculated in sequence. The angular rate sequence is generated by dividing the difference by the time step and the amplitude mutation judgment rule is set to obtain the trajectory direction mutation ratio sequence; Based on the trajectory direction mutation ratio sequence, a greedy algorithm is used to perform mutation segment division operation, perform mutation point index positioning and extract the adjacent point sequence before and after, accumulate and sum the continuous angle difference segments, select the segments with mutation amplitude greater than the set value and record the path position index to obtain the scour mutation path segment set; Based on the erosion mutation path segment set, the maximum value of the node flow direction angle set within the segment is located and it is determined whether there is an angle difference reversal point on both sides. The nodes where the reversal point is located are spatially reorganized and numbered to generate the erosion and sedimentation dominant path node set.

7. The method for analyzing the evolution of seabed scouring and silting according to claim 6, characterized in that: The greedy algorithm is based on the formula: Where: T s represents the comprehensive mutation intensity value of the mutation fragment, ψ i represents the tangential direction angle of the i-th trajectory point, |ψ i+1 -ψ i | represents the direction angle difference between adjacent trajectory points, reflecting the degree of path turning, e i represents the seabed elevation value of the i-th trajectory point, d i,i+1 Represents the horizontal distance between trajectory points i and i+1 It represents the rate of change of elevation per unit horizontal distance between adjacent trajectory points, reflecting the sudden change of local slope. i represents the local curvature value at the i-th trajectory point, |κ i+1 -κ i | represents the curvature variation between adjacent points, s i represents the unit path flow velocity at the i-th trajectory point, Represents the average flow velocity of all trajectory points in the current mutation segment, represents the point velocity offset value, μ1, μ2, μ3, and μ4 represent the weighted coefficients of directional mutation, elevation gradient, curvature fluctuation, and velocity offset factor, respectively.

8. The method for analyzing the evolution of seabed scouring and silting according to claim 1, characterized in that: The specific steps of generating the scouring and silting evolution characteristic structure partition map are as follows: Based on the erosion and sedimentation dominant path node set, the direction vectors of continuous nodes are extracted and the angle calculation between adjacent nodes is performed. The direction change pattern is identified through the positive and negative sign sequence of the angle difference, and the start and end positions of the monotonic continuous sequence are marked to generate a monotonic direction path segment set. Based on the monotonic directional path segment set, the direction angle difference of each node pair and the distance value of the adjacent nodes are extracted in the order of the path segments, and an angle sequence and a length sequence are constructed respectively. The node numbers are grouped and combined with the codes to obtain a path angle length parameter sequence set; Based on the path angle length parameter sequence set, the angle sequence fluctuation values ​​and length spacing change values ​​are extracted according to the node numbers within the segment. The structural segment numbers are rearranged after classification based on the value difference range within continuous segments, and the paths of the same type are merged and collected to establish a characteristic structural zoning map of scouring and deposition evolution.

9. A seabed scouring and silting evolution analysis system, characterized in that: The method for analyzing the evolution of seabed scouring and silting according to any one of claims 1 to 8, wherein the system comprises: Elevation analysis module: Based on single-phase seabed grid elevation data, the elevation values ​​of adjacent grid cells in the shoal and channel area are extracted, the elevation difference and slope vector direction are calculated pair by pair, the slope vector angle is obtained and whether it falls within the set angle range is determined, the unidirectional edge pairs are selected, and the elevation difference and slope direction are used as edge weight values ​​to establish a directional edge weight graph structure; Edge Mapping Module: Based on the directional edge weight graph structure, a graph attention network is used to extract directional edge relationships from the adjacent node set of each node. The slope direction angle value, elevation increase and decrease continuity mark, and flow direction consistency ratio of each connecting edge are judged to see whether they meet the corresponding thresholds. All directional edge groups that meet the conditions are selected, and the cumulative connection frequency of directional clusters in the graph is counted. An indicator set is generated based on the frequency to form a directional dominant influence indicator group. Dominant direction module: Based on the direction-dominant influence index group, the flow direction component value and slope component value of each grid node in the control area are extracted, and the node numbers of all nodes whose flow direction components are greater than the slope components are screened. The flow direction paths of each numbered node in multiple time periods are read, and the flow direction sequences are reorganized after sorting. The structural change trends of all sequences in the time dimension are summarized to establish a direction evolution path sequence group; Path evolution module: Based on the directional evolution path sequence group, the turning point numbers of the slots in each path are extracted, the angle difference between each pair of turning points and the ratio of the continuous turning angle change rate are calculated, and whether the change rate ratio exceeds a threshold is determined. The first node of the path mutation section is marked as the scour starting point, all extreme points are read to form a path chain, and a greedy algorithm is used to cluster the path extreme nodes. The cluster center position is output to form a scour and silt distribution node set; Structural division module: Based on the scour and sedimentation distribution node set, determine whether the change direction of the angle between adjacent nodes in the path segment is consistent, screen out all monotonous angle path segments, read the angle difference sequence and path length numerical sequence in each path segment respectively, remap the path structure order according to the combined value of the angle and length, aggregate all segment paths to form a path segment set, and establish a scour and sedimentation evolution partition map.

Images