CN-121982259-A - Reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data

CN121982259ACN 121982259 ACN121982259 ACN 121982259ACN-121982259-A

Abstract

The invention provides a reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data, which relates to the field of reservoir sediment monitoring and modeling, and comprises the steps of constructing a reservoir space grid framework model, and collecting multi-time-sequence multi-beam sounding point cloud data and corresponding hydrologic auxiliary data; the method comprises the steps of carrying out space division by adopting a self-adaptive tile subdivision method, generating multi-level point cloud tiles by adopting a characteristic point downsampling method, establishing a three-dimensional composite index, processing an anisotropic ICP registration method for indexed multi-level point cloud tile sets to obtain multi-time-sequence registered point cloud tile sets, establishing an initial irregular triangular net sediment model, carrying out differential partitioning by utilizing historical hydrologic auxiliary data, carrying out dynamic updating based on partition states to obtain a dynamic reservoir sediment model, calculating to obtain sedimentation parameters, and mapping sedimentation to obtain an interactive three-dimensional dynamic sedimentation situation map by adopting a three-dimensional engine. The method realizes high-precision dynamic visualization of the reservoir sediment accumulation process.

Inventors

ZHAO WENJIE
ZHANG QIANJUN
ZHAO YANG
JIAN XINGXIANG
WANG XIANGPENG
WANG LUAN
LIU ZHONGQUAN
WU WENHAO

Assignees

四川省紫坪铺开发有限责任公司

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (10)

1. A reservoir sediment modeling method based on multi-time sequence multi-beam sounding point cloud data is characterized by comprising the following steps: s1, constructing a reservoir space grid skeleton model based on reservoir mapping data, and collecting multi-time-sequence multi-beam point cloud data and corresponding hydrologic auxiliary data for a reservoir according to a preset period; s2, preprocessing multi-time sequence multi-beam point cloud data to obtain processed point cloud data, performing space division on the processed point cloud data by adopting a self-adaptive tile subdivision method, generating multi-level point cloud tiles by adopting a characteristic point downsampling method of fusing hydrologic flushing differential strength indexes, and establishing a three-dimensional composite index fusing space coding, time sequence labels and detail levels to obtain an indexed multi-level point cloud tile set; S3, processing the indexed multi-level point cloud tile set by using an anisotropic ICP registration method by taking the tiles as units, and restraining according to hydrologic auxiliary data to obtain a multi-time sequence registration point cloud tile set; S4, establishing an initial irregular triangular mesh sediment model by adopting a Delaunay triangulation method based on the point cloud tile set with multi-time sequence registration, and carrying out differential partitioning on the reservoir space grid framework model according to the hydrologic state distance by utilizing historical hydrologic auxiliary data to obtain a partitioning state; and S5, calculating the sedimentation parameters based on the peak heights of adjacent time sequence triangular meshes in the dynamic reservoir sediment model, and mapping the sedimentation parameters to the dynamic reservoir sediment model in a color gradient mode by adopting a three-dimensional engine to obtain an interactable three-dimensional dynamic sedimentation situation map.
2. The reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data as set forth in claim 1, wherein the step S1 specifically includes: Acquiring reservoir mapping data and historical hydrologic auxiliary data, and determining a reservoir static constraint boundary by taking the highest historical water level of the reservoir as a reference standard to obtain a reservoir space range, wherein the hydrologic auxiliary data comprises water level, flow and sand content; Performing space subdivision according to a preset size by adopting a regular grid subdivision method in a space range of a reservoir area to obtain a plurality of grid cells, and storing all the grid cells according to a quadtree coding rule to obtain a reservoir area space grid skeleton model; Carrying out multi-beam underwater sounding on the reservoir universe according to a preset period to obtain sounding system parameters, and storing the sounding system parameters in grid units corresponding to a reservoir space skeleton model in an associated mode to obtain multi-time sequence multi-beam point cloud data and corresponding hydrologic auxiliary data, wherein the sounding system parameters comprise beam emergence angles corresponding to sounding points Measuring depth of water Sound velocity profile data.
3. The reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data as set forth in claim 2, wherein the method for performing spatial division on the processed point cloud data by adopting the adaptive tile subdivision method comprises the following steps: Taking the reservoir space grid skeleton model as an initial tile division basis, and counting the point cloud density of the point cloud data processed in each grid unit to obtain the point cloud density distribution of each grid unit; according to the density of the point cloud of each grid unit, a preset high density threshold value is set And preset low density threshold Hierarchical subdivision is performed in which : The point cloud density of the grid unit exceeds a preset high density threshold value Performing quadtree recursion subdivision on the grid cells of the tree to obtain a first detail level tile; the point cloud density of the grid unit is lower than a preset low density threshold value Performing quadtree reverse merging on the grid units of the tree, merging the quadtree reverse merging into a parent node tile level to obtain a third detail level tile; Placing the point cloud density of the grid cells at The grid units of the intervals keep the node scale of the current level to obtain a second detail level tile; forming a set of adaptively sized tiles based on the first level of detail tile, the second level of detail tile, and the third level of detail tile; for each tile in the adaptively sized set of tiles, each of its four sides is outwardly expanded in width And is arranged with a width inside the static constraint boundary of the reservoir region And obtaining a multi-level tile structure.
4. The reservoir sediment modeling method based on multi-time sequence multi-beam sounding point cloud data as set forth in claim 3, wherein the generating the multi-level point cloud tile by adopting the characteristic point downsampling method of fusing hydrologic impulse sediment differential intensity indexes specifically comprises the following steps: The method comprises the steps of searching k nearest neighbor points in a spherical neighborhood of a radius r by taking each point p as a core for the point cloud preprocessed in each tile in a multi-layer tile structure, and constructing a k neighborhood point set of each point; Respectively calculating curvature mean value of each point based on k neighborhood point set And normal change rate ; High program column in the first M historical time sequences of time sequence dimension query point p through three-dimensional compound index Calculating the differential strength of time sequence flushing ; Respectively normalizing the curvature mean value, the normal change rate and the time sequence erosion differential strength of each point in the tiles to which each point belongs, and calculating the comprehensive feature importance score of each point by adopting a weighted summation method Obtaining importance scoring results of each point; according to the cloud density of target points of each detail level tile, sorting each point in each tile according to the importance scores of the comprehensive features from high to low, sequentially screening and reserving the points, wherein the importance scores of the comprehensive features are lower than a preset minimum importance threshold value Is set at a preset space sampling interval And performing uniform rasterization, and clipping each tile point cloud according to the upper limit of the point cloud density of each detail level to obtain a multi-level point cloud tile.
5. The reservoir sediment modeling method based on multi-time-series multi-beam sounding point cloud data as set forth in claim 4, wherein the calculation formula of the comprehensive feature importance score W (p) is as follows: ; ; Wherein, the The differential strength of the time sequence flushing is represented, Is taken as a point In the first place The elevation values in the historical timing sequence, For the total number of historical timings existing before the current timing, Is the hydrologic similarity distance decay coefficient, For the current time sequence Is used to determine the hydrological state vector of (c), Is the first The hydrologic state vector of each historical sequence, Is the hydrologic state similarity distance; Represents the normalized curvature average value, Indicating the normalized normal change rate of the line, Represents the normalized time sequence impulse differential intensity, A weight coefficient representing the normalized curvature mean value, A weight coefficient representing the normalized normal change rate, And a weight coefficient representing normalized time sequence impulse differential intensity.
6. The reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data as set forth in claim 2, wherein the step S3 specifically includes: Using the depth sounding system parameters in the corresponding grid cells of the reservoir space grid skeleton model, and using the multi-beam acoustic depth sounding error propagation relationship as each point in each tile in the multi-level point cloud tile set Construction Diagonal anisotropic error covariance matrix Forming an anisotropic error covariance matrix set; ; Wherein, the Is the standard deviation of the horizontal position error in the vertical line direction, To be the standard deviation of the horizontal position error along the direction of the line, Is the standard deviation of the direction error of the elevation, In each time sequence point cloud tile, a tile corresponding to the first acquisition time sequence is used as a reference tile, when the mean value of anisotropic error covariance matrix determinant of tiles corresponding to other time sequences is smaller than that of the tiles corresponding to the first acquisition time sequence, the tile corresponding to the time sequence is replaced by the reference tile, and for target tiles corresponding to other time sequences, the reference tile is used as registration reference, and a rotation matrix is solved based on the anisotropic error covariance matrix and an objective function And translation vector Until the change quantity of the objective function of two adjacent iterations is lower than a preset convergence threshold value, outputting a converged rotation matrix And translation vector : ; Wherein, the A three-dimensional coordinate vector representing the i-th point in the target tile, Representing the sum in a reference tile The three-dimensional coordinate vector of the corresponding nearest matching point, Representing the ith point in the target tile Is a matrix of anisotropic error covariances of (c), Representing the i-th corresponding point in the reference tile Is a anisotropic error covariance matrix; According to the converged rotation matrix And translation vector Performing coordinate transformation on the target time sequence point cloud tiles to obtain time sequence tiles after spatial alignment; and carrying out weighted average fusion on the time sequence point cloud tiles subjected to spatial alignment, and reserving the spatial difference characteristics of the time sequence point clouds according to the time sequence labels in the three-dimensional composite index to obtain a multi-time sequence registration point cloud tile set.
7. The reservoir sediment modeling method based on multi-time sequence multi-beam sounding point cloud data as set forth in claim 1, wherein the utilizing the historical hydrologic auxiliary data to differentially partition the reservoir space grid skeleton model according to hydrologic state distances specifically comprises: Acquiring hydrologic auxiliary data of a current time sequence, including flow rate of the current time sequence Sand content Water level and water level And construct the current time sequence hydrologic state vector Calculating the current time sequence hydrologic state vector according to the historical time sequence hydrologic auxiliary data in each grid unit of the grid framework model of the reservoir space With each time sequence hydrologic state vector of history Hydrologic state distance between : ; Wherein, the 、、 Respectively history of The flow rate, the sand content and the water level of each time sequence, 、、 Respectively counting standard deviations of flow, sand content and water level in all historical time sequences; Will be Below a preset similarity threshold The historical time sequence of the (a) is judged to be the historical similar time sequence with similar hydrologic conditions as the current time sequence, and a historical similar time sequence set is obtained ; Counting each tile in the spatial grid skeleton model of the reservoir area Mean value of the point cloud elevation change amplitude of each history time And is matched with a preset elevation change amplitude threshold value And (3) performing comparison: When (when) Exceeding a preset elevation change amplitude threshold The corresponding tile is determined to be a significant tile of the fouling change, namely the high active region, and the elevation change confirmation threshold is The area formed by outwards expanding a circle of adjacent tiles in the spatial distribution range of the last time sequence of the high active area is defined as a boundary expansion area, and the elevation change of the boundary expansion area confirms a threshold value Wherein ; When (when) Below a preset elevation change amplitude threshold Determining the corresponding tile as a very micro tile with elevation change, namely a low active region, and determining the elevation change as a threshold value ; Writing the partition state of each tile and the corresponding elevation change confirmation threshold value into the tile state zone bit of the three-dimensional composite index to obtain the partition state comprising the high active region, the low active region and the boundary expansion region.
8. The reservoir sediment modeling method based on multi-time sequence multi-beam sounding point cloud data of claim 7, wherein the method is characterized by dynamically updating an initial irregular triangular network sediment model based on a partition state and specifically comprises the following steps: matching plane coordinates of newly added point cloud data of the current time sequence with a quadtree space index in a three-dimensional composite index, searching a tile space code in which the newly added point cloud data falls, determining a tile set affected by the newly added data, reading tile state flag bits in the three-dimensional composite index of each tile, and acquiring partition states of each tile and corresponding elevation change confirmation thresholds to obtain a tile set to be updated of the current time sequence and elevation change confirmation thresholds of each tile; Comparing the elevation change quantity of each point in the newly added point cloud data with an elevation change confirmation threshold corresponding to the tile to be updated for each tile in the tile set, reserving a point cloud data subset with the elevation change quantity exceeding the corresponding elevation change confirmation threshold as a current time sequence effective change point cloud subset, and judging the point cloud data with the elevation change quantity not exceeding the corresponding elevation change confirmation threshold as an ineffective change subset; merging the effective change point cloud subsets in each tile into the existing point cloud of the corresponding tile, recursively splitting the existing point cloud into left and right subsets again until the number of the subset point clouds does not exceed a preset splitting threshold value, independently executing Delaunay triangulation on each subset, and reconstructing a local irregular triangular network of the tile; Multiplexing an initial irregular triangular network for the invalid change subsets in each tile; And merging the local irregular triangular net after the reconstruction of each tile with the multiplexed initial irregular triangular net, and writing the current time sequence label into the three-dimensional composite index time sequence dimension of the corresponding tile to obtain the dynamic reservoir sediment model.
9. The reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data as set forth in claim 1, wherein the step S5 specifically includes: based on each vertex of the triangular net corresponding to the adjacent time sequence in the dynamic reservoir sediment model, respectively calculating the sedimentation parameters at each vertex, including sedimentation thickness Rate of fouling Amount of siltation in the area of each triangular patch : ; ; ; Wherein, the For the thickness of the deposit between adjacent time sequences at the apex of the triangle mesh, For the current time sequence Corresponding to the elevation value of the vertex, For the elevation value of the vertex corresponding to the immediately preceding time sequence, For the rate of fouling at this vertex, For adjacent time sequences And (3) with The time interval between the two times of the two, As the amount of fouling of the target area, Is the first The area of the triangular face piece is equal to the area of the triangular face piece, Is the first Deposition thickness at three vertices of each triangular patch Is the average value of (2); and loading irregular triangular mesh model tiles corresponding to each time sequence and each detail level of the dynamic reservoir sediment model by using a three-dimensional engine, and mapping each sedimentation parameter to the surface of the dynamic reservoir sediment model in a color gradient mode to obtain an interactable three-dimensional dynamic sedimentation situation map.
10. The method for modeling reservoir sediment based on multi-time-series multi-beam sounding point cloud data as set forth in claim 9, wherein the method for modeling reservoir sediment further comprises: Constructing a time sequence prediction model by adopting a long-short-term memory neural network, and training the time sequence prediction model by taking each time sequence siltation parameter sequence and a corresponding hydrologic auxiliary data sequence as input to obtain a trained time sequence prediction model; Inputting preset future working condition scene parameters into the trained time sequence prediction model to obtain a prediction result of sediment accumulation amount, accumulation range and elevation change in a specified period under the working condition scene, wherein the preset future working condition scene parameters are water level preset values in the future specified period set according to engineering requirements Preset flow value Preset value of sand content 。

Description

Reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data Technical Field The invention relates to the technical field of reservoir sediment monitoring and modeling, in particular to a reservoir sediment modeling method based on multi-time sequence multi-beam sounding point cloud data. Background Reservoir sediment accumulation is one of the core problems affecting the long-term operation safety and regulation capacity of reservoirs. The water flow in the reservoir area carries a large amount of suspended solids and bed load sediment, and is continuously settled and piled up in the reservoir area after the flow speed is reduced, so that the reservoir flood control standard is reduced and the water supply energy-regulating capacity is weakened. Along with the increase of the operation years of reservoirs, sediment accumulation problems become more and more serious, and if the accumulation space distribution and dynamic evolution rule cannot be mastered in time, the safe operation and dredging decision of the water supply reservoir bring great hidden trouble. Therefore, an effective reservoir sediment accumulation monitoring and modeling system is established, and is an important technical requirement for reservoir operation management. The multi-beam sounding technology gradually becomes a main stream means for measuring underwater topography of a reservoir by virtue of the full-coverage and high-density underwater topography data acquisition capability. Compared with the traditional single-beam sounding and section method measurement mode, the multi-beam sounding system can acquire banded high-density three-dimensional point cloud data in a single operation, and the topography restoration precision of a reservoir area is remarkably improved. However, the multi-beam sounding point cloud data volume is huge, measurement errors are accumulated among the multi-time sequence data, and in addition, the reservoir region topography is obvious along with the dynamic change of the hydrologic process, so that how to efficiently organize, accurately register and dynamically model the multi-time sequence multi-beam point cloud data is still a technical problem to be solved in the current research and engineering application. The Chinese patent application number 202411178622.3 discloses a short-term prediction method and a system for the generating power of a small hydropower station, the method carries out full coverage scanning on a reservoir basin through multi-beam sonar, and acquiring sediment deposition thickness distribution, deposition volume distribution, deposition morphological parameter distribution and other regional sediment deposition distribution characteristics, reconstructing the three-dimensional surface of the sediment accumulation body by adopting a Delaunay triangular grid algorithm, and carrying out short-term prediction on the power generation power of the small hydropower station by combining a hybrid neural network model. However, the method cannot dynamically update according to the underwater topography of the reservoir area in time, so that the data measurement accuracy is reduced, and the engineering requirements of fine modeling and dynamic evolution quantitative analysis of the sediment accumulation distribution of the reservoir are difficult to meet. Disclosure of Invention In view of the above, the invention provides a reservoir sediment modeling method based on multi-time-sequence multi-beam sounding point cloud data, which aims to solve the problem of low dynamic update efficiency of sediment models caused by irregular terrains in reservoir areas in the prior art and realize high-precision dynamic visualization of reservoir sediment accumulation processes. The technical scheme of the invention is realized as follows: In one aspect, the invention provides a reservoir sediment modeling method based on multi-time sequence multi-beam sounding point cloud data, which comprises the following steps: s1, constructing a reservoir space grid skeleton model based on reservoir mapping data, and collecting multi-time-sequence multi-beam point cloud data and corresponding hydrologic auxiliary data for a reservoir according to a preset period; s2, preprocessing multi-time sequence multi-beam point cloud data to obtain processed point cloud data, performing space division on the processed point cloud data by adopting a self-adaptive tile subdivision method, generating multi-level point cloud tiles by adopting a characteristic point downsampling method of fusing hydrologic flushing differential strength indexes, and establishing a three-dimensional composite index fusing space coding, time sequence labels and detail levels to obtain an indexed multi-level point cloud tile set; S3, processing the indexed multi-level point cloud tile set by using an anisotropic ICP registration method by taking the tiles as units, and restraining according to hydrologic auxilia