Sediment flow state path determination method suitable for high-sediment-content river channel

Abstract

The invention discloses a sediment flow state path determination method suitable for a high-sediment-content river channel. The method comprises the steps that S1, the riverbed terrain, flow velocity and flow direction data and water surface suspended sediment concentration distribution of the high-sediment-content river channel are obtained; s2, a three-dimensional fluid dynamic model is imported, a high-sediment-content river sediment transportation numerical model is constructed, and the sediment transportation process is simulated; s3, dividing different diffusion areas of sediment according to the sediment transportation numerical model; s4, according to the divided areas, obtaining a sediment flow main path, and establishing a function relationship among the main path swing amplitude, the flow, the water level and the riverbed roughness; and S5, predicting the main path of the sediment flow under the flood of different frequencies, changing the local terrain of the river channel according to a prediction result, and regulating and controlling the sediment flow state of the high-sediment-content river channel. According to the method, breakthrough is realized in concentration description, region division and prediction precision, and a more efficient and reliable technical means is provided for regulation and control of river sediment transportation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of channel silt monitoring, and particularly relates to a method for determining the path of a sediment flow state suitable for a high-sand-content river channel. BACKGROUND

[0002] The sediment transport process of a high-sand-content river channel is complex, and the instability of the sediment flow state can easily lead to river channel silting, dramatic changes in the riverbed morphology, and further exacerbate flood risks and reduce navigation capacity. Traditional river channel management methods cannot accurately grasp the dynamic evolution of the sediment flow state, resulting in a lack of scientificity and timeliness of the control measures.

[0003] Traditional high-sand-content river channel sediment management mainly relies on a method combining physical model tests and empirical formulas. The physical model simulates the river channel topography and flow conditions through scaling, and qualitatively analyzes the sediment deposition law, but has defects such as distortion of boundary conditions due to scale limitations and inability to simulate complex hydrological processes. The empirical formula is based on historical data to fit the relationship between sediment transport volume and hydraulic parameters, which is simple to calculate, but lacks dynamic tracking of the sediment particle trajectory, making it difficult to reflect the spatial differentiation characteristics of the sediment flow state under different flow and water level conditions. For example, the traditional method usually regards the river channel as a homogeneous medium, ignoring the local impact of the riverbed microtopography on sediment transport, resulting in low prediction accuracy of the model, especially in areas with dramatic changes in water flow (such as bends and notches). In addition, the traditional technology relies on manual experience to set control parameters, and cannot achieve adaptive optimization of the management scheme, resulting in low efficiency and economy.

[0004] Existing technologies gradually introduce numerical simulation methods, such as two-dimensional sediment transport models based on the finite difference method. By discretizing the river channel space, the water flow continuity equation and the sediment conservation equation are solved to achieve preliminary simulation of the sediment concentration distribution. However, such models generally use structured grids, which are difficult to adapt to complex riverbed topography, and can only simulate sediment transport under a single working condition, and cannot couple the dynamic boundary conditions of multi-frequency flood processes. For example, existing models do not consider the individual motion characteristics of sediment particles, and only describe the behavior of the sediment population through statistical averaging methods, resulting in insufficient identification accuracy of the high-concentration sediment flow core area. In terms of control, existing technologies mostly use fixed-form sediment barriers or dredging projects, lack prediction of the swing law of the main path of the sediment flow, and are difficult to achieve precise control in line with the situation, often causing new silting or scouring problems. SUMMARY

[0005] Based on the above technical problems, the present application discloses a method for determining the path of a sediment flow state suitable for a high-sand-content river channel, comprising:

[0006] S1, obtaining the riverbed topography, flow velocity and flow direction data, and the suspended sediment concentration distribution of the water surface of the high-sand-content river channel;

[0007] S2, import the collected riverbed topographic data, flow velocity and direction data, and suspended sediment concentration distribution data into a three-dimensional fluid dynamics model, discretize the river channel by unstructured mesh, construct a high-sediment-concentration river channel sediment transport numerical model, and simulate the sediment transport process under different flow rates and water levels;

[0008] S3, according to the sediment transport numerical model, mark a large number of virtual sediment particles in the model, track their trajectories under the action of water flow, divide the high-concentration sediment flow core area, the transition area and the low-concentration diffusion area according to the sediment concentration distribution data, and determine the boundaries of each area;

[0009] S4, according to the divided sediment flow area, extract the center line of the high-concentration sediment flow core area as the main path of the sediment flow, analyze the swing law of the main path in different hydrological periods, and establish the functional relationship between the swing amplitude of the main path and the flow rate, water level and riverbed roughness;

[0010] S5, combining historical flood data and future hydrological prediction information, using the function relationship, the main path of sediment flow under different frequency floods is predicted, according to the prediction result, by changing the local topography of the river channel, guiding the sediment flow to transport along the predetermined path, realizing the regulation and control of the high-sediment-concentration river channel sediment flow state.

[0011] Preferably, the water surface suspended sediment concentration distribution in S1 is obtained by: using an unmanned ship equipped with a hyperspectral imager and a laser particle size analyzer to periodically cruise along the river channel section, the hyperspectral imager collects water surface spectral data, the laser particle size analyzer measures the sediment particle size data in real time, the sediment particle size data is coded into a multi-scale feature vector through a spectral inversion model, the spatial-spectral features of the spectral data are interacted through a feature fusion module, and the particle size-spectral coupling feature matrix is used to realize layered inversion of the water surface suspended sediment concentration at different positions. The inversion result of the water surface suspended sediment concentration at different positions is obtained.

[0012] Preferably, the spectral inversion model structure is a three-dimensional convolutional neural network, the input layer receives continuous waveband spectral data collected by the hyperspectral imager, the spatial local features of the spectral data are extracted through the two-dimensional convolutional layer, and the spatial-spectral joint features of the spectral data are captured through the three-dimensional convolutional layer; a double attention module is introduced, wherein a channel attention submodule obtains spectral channel features through global average pooling and global maximum pooling, calculates channel weights through a multilayer perceptron, and performs weighted fusion, and a spatial attention submodule generates a spatial attention map based on the channel features and element-wise multiplies the spectral data feature map; the sediment particle size data measured by the laser particle size instrument are encoded into a feature vector through a fully connected layer, cross-modal interaction is performed between the spectral features through a gating fusion unit, a multi-scale deconvolution layer is arranged at the end of the model to restore the spatial resolution, a residual connection structure is used to fuse different levels of features, and the output layer outputs the inversion results of the suspended sediment concentration at different positions on the water surface.

[0013] Preferably, in S2, when the sediment transport numerical model of the high-sediment-concentration river channel is constructed, an unstructured grid discretization method based on the finite volume method is used to divide the spatial region Ω of the river channel into N control bodies, for each control body, a continuity equation of sediment transport is established: wherein α is the porosity, C is the sediment concentration, t is the time, is a flow velocity vector, D is a diffusion coefficient, S is a sediment source-sink term, a variation multiscale finite element technique is used to couple large-scale grids with small-scale river micro-topographic features, adaptive grid division processing is performed on the riverbed topographic data, a local grid refinement strategy is used in areas where the flow changes dramatically and the riverbed morphology is complex, and high-precision numerical simulation of sediment transport under different flow and water level conditions is realized.

[0014] Preferably, in S2, when the sediment transport process under different flow and water level conditions is simulated, a dynamic boundary condition coupling technique is used to embed the flow Q and the water level H as time-varying parameters into the model, and a flow-water level-sediment transport joint control equation is constructed: wherein is a velocity vector, ρ is the fluid density, p is the pressure, v is the kinematic viscosity coefficient, is an external force term, is a momentum source term caused by sediment transport, a Monte Carlo random sampling method is used to model the probability distribution of the flow and water level fluctuations under different working conditions, synchronous simulation of the sediment transport process under multiple working conditions is realized, and the sediment transport process under different flow and water level conditions is obtained by capturing the convection-diffusion trajectories of the sediment particles.

[0015] Preferably, in S3, a density peak clustering algorithm is used to divide the region in combination with the space-time trajectory features, a three-dimensional space trajectory matrix T is constructed according to the time sequence of the labeled virtual sediment particle motion trajectories, n×mwhere n is the number of particles, m is the time step; the local density of each particle trajectory point is calculated where χ is the indicator function, d ij is the Euclidean distance between particles i and j, d c is the truncation distance, the particles with high local density and far distance from higher density points are selected as cluster centers, combined with the sediment concentration distribution data, the concentration threshold C th1 and C th2 , C th1 > C th2 , the area with local density greater than p high and concentration greater than C th1 is divided into the high-concentration sediment flow core area, the area with local density between p mid and p high and concentration between C th2 and C th1 is divided into the transition area, and the remaining area is divided into the low-concentration diffusion area, connecting the boundary points of each area to determine the boundary of each area.

[0016] Preferably, in S4, when extracting the center line of the high-concentration sediment flow core area, a combination algorithm based on topological skeleton extraction and dynamic programming is adopted, first, the two-dimensional plane projection of the high-concentration sediment flow core area is converted into a binary image, and morphological erosion operation is used to peel off the area boundary layer by layer until a single-pixel-wide topological skeleton S is obtained, and a cost function J = is constructed where (x i , y i ) is the skeleton node coordinates, d j is the shortest distance from the node to the core area boundary, w s is the weight coefficient, n s is the total number of nodes, and m s is the total number of paths, and the path with the minimum cost function J is searched on the skeleton S by the dynamic programming algorithm, which is taken as the center line of the high-concentration sediment flow core area, i.e. the main path of the sediment flow.

[0017] Preferably, in S4, when establishing the functional relationship between the swing amplitude of the main path and the flow, water level, and river bed roughness, a space-time joint feature extraction and multivariate nonlinear regression model is adopted, first, the swing trajectory of the main path in different hydrological periods is subjected to space-time discretization processing, and the time series feature vector and the spatial distribution feature matrix M of the swing amplitude are extracted, the river bed roughness coefficient n r , the flow Q, and the water level H are introduced as independent variables, and a multivariate nonlinear function is constructed where a i , b i , g i , d iFor the corresponding parameter coefficients, the mean square error between the observed swing amplitude and the model prediction is minimized by optimizing the model parameters. Establish a high-precision functional relationship between the main path swing amplitude and flow rate, water level, and riverbed roughness.

[0018] Preferably, in S5, the flow rate, water level, riverbed roughness, and corresponding main path swing amplitude from historical flood data are used as training samples to construct a fusion model of long short-term memory network and functional relationship, which is used to predict the flow rate Q in future hydrological information. pred Water level H pred Riverbed roughness By establishing the functional relationship A=f(Q,H,n) r Calculate the predicted initial swing amplitude A init Then combine it with A four-dimensional input vector is formed and fed into an LSTM model for temporal feature mining. An attention mechanism is introduced into the model to calculate the weight ω of each input feature. i By weighted summation Where X i Using the input vector elements, the final predicted swing amplitude is obtained. Combined with the main path swing centerline offset algorithm, the predicted swing amplitude is converted into the spatial coordinates of the main path of sediment flow under different frequency floods, and the prediction results are obtained.

[0019] Preferably, in step S5, when adjusting the sediment flow regime based on the prediction results, a river topography adjustment cost function is constructed by predicting the main sediment flow path offset trend. in L represents the target length and width of the predetermined path. j E j λ represents the actual length and width of the current path. j Using weighted coefficients, dynamic reshaping of local river topography is performed to achieve efficient transport of sediment along a predetermined path.

[0020] Compared with the prior art, the technical solution of this application has the following technical effects:

[0021] This invention acquires the concentration distribution of suspended sediment on the water surface using an unmanned vessel equipped with a hyperspectral imager and a laser particle size analyzer. By combining a three-dimensional fluid dynamics model with unstructured mesh technology, it achieves high-precision simulation of sediment transport processes in high-sediment-content rivers. The spectral inversion model adopts a three-dimensional convolutional neural network structure. It extracts the spatial local features of spectral data through two-dimensional convolutional layers and captures the spatial-spectral joint features through three-dimensional convolutional layers. It also introduces a dual attention module to enhance the interaction between channels and spatial features. At the same time, it integrates sediment particle size data measured by the laser particle size analyzer to achieve hierarchical inversion of sediment concentration at different locations.

[0022] The application realizes dynamic division of high-concentration sediment flow core area, transition area and low-concentration diffusion area by marking virtual sediment particles and tracking their movement trajectories, combining a density peak clustering algorithm with space-time trajectory features. The specific technical path includes: constructing particle movement trajectories into a three-dimensional space trajectory matrix, determining clustering centers by calculating local density, and combining sediment concentration threshold to divide area boundaries. Compared with the existing technology relying on static concentration threshold partition method, this scheme can dynamically capture the convection-diffusion trajectory of sediment particles, accurately identify the spatial distribution and boundary change of the high-concentration core area.

[0023] The application realizes accurate prediction of the main path of sediment flow under different frequency floods by establishing a multivariate nonlinear function relationship between the swing amplitude of the main path, flow, water level and river bed roughness, and fusing a long short-term memory network (LSTM) for time series feature mining. The space-time characteristics of the swing amplitude are extracted through space-time discretization processing, a multivariate nonlinear regression model is constructed, the LSTM model is used for time series analysis of future hydrological parameters, the feature weight is optimized by introducing an attention mechanism, and finally the prediction result is converted into spatial position coordinates by combining the center line offset algorithm. This breaks through the limitations of existing fixed form hydraulic structures, realizes the technical leap from "passive response" to "active regulation", and significantly improves the efficiency and economy of high-sediment-content river regulation.

[0024] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the contents of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following will be described in detail with the preferred embodiments of the present application and with the help of the accompanying drawings.

[0025] According to the detailed description of the specific embodiments of the present application below in combination with the drawings, those skilled in the art will more clearly understand the above and other purposes, advantages and features of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creating any creative labor. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual proportion.

[0027] Figure 1 A flow chart of a sediment flow state path determination method suitable for a high-sediment-content river channel of the present application;

[0028] Figure 2 The schematic diagram of the spectrum inversion model structure of the application (three-dimensional convolutional neural network architecture);

[0029] Figure 3 The comparative diagram of the spatial distribution of sediment concentration of the application (a. Vertical distribution comparison; b. Lateral distribution and core area comparison);

[0030] Figure 4 The comparative diagram of the concentration characteristics of the water flow severe change area of the application;

[0031] Figure 5 The comparative diagram of the main path swing prediction error time series of the application;

[0032] Figure 6 The comparative diagram of the main path swing amplitude prediction of the application under the 5% frequency flood working condition;

[0033] Figure 7 The comparative diagram of the main path deflection rate of the application in the curved section. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments of the present application. In the following description, specific details such as specific configurations and components are provided only to help a comprehensive understanding of the embodiments of the present application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, in order to be clear and concise, the description of known functions and structures is omitted in the embodiments.

[0035] It should be understood that the "one embodiment" or "the embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "one embodiment" or "the embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner.

[0036] In addition, reference numerals and / or letters can be repeated in different examples of the present application. Such repetition is for the purpose of simplification and clarity, and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0037] The term "and / or", used in the present document, only describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent three cases of A alone, B alone, and A and B together. The term " / " in the present document describes another association relationship of the associated objects, which means that there can be two relationships, for example, A / and B can represent two cases of A alone and A and B together. In addition, the character " / " in the present document generally represents an "or" relationship between the associated objects before and after it.

[0038] The term "at least one" in the present document only describes the association relationship of the associated objects, which means that there can be three relationships, for example, at least one of A and B can represent three cases of A alone, A and B together, and B alone.

[0039] It should also be noted that, in the present document, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion.

[0040] Embodiment 1

[0041] The present embodiment mainly describes a method for determining the path of sediment flow in a high-sediment-content river channel, as shown in Figure 1 The specific steps include:

[0042] S1, obtaining the riverbed topography, flow velocity and flow direction data, and water surface suspended sediment concentration distribution of the high-sediment-content river channel;

[0043] S2, importing the collected riverbed topography data, flow velocity and flow direction data, and suspended sediment concentration distribution data into a three-dimensional fluid dynamics model, discretizing the river channel by using an unstructured grid, constructing a high-sediment-content river channel sediment transport numerical model, and simulating the sediment transport process under different flow rates and water levels;

[0044] S3, according to the sediment transport numerical model, marking a large number of virtual sediment particles in the model, tracking the motion trajectory under the action of water flow, combining the sediment concentration distribution data, dividing the high-concentration sediment flow core area, the transition area and the low-concentration diffusion area, and determining the boundaries of each area;

[0045] S4, according to the divided sediment flow area, extracting the center line of the high-concentration sediment flow core area as the main path of the sediment flow, analyzing the swing law of the main path in different hydrological periods, and establishing the functional relationship between the swing amplitude of the main path and the flow rate, water level, and riverbed roughness;

[0046] S5, in combination with historical flood data and future hydrological prediction information, the main path of sediment flow under different frequency floods is predicted by using a functional relationship, according to the prediction result, the local topography of the river is changed to guide the sediment flow to move along the predetermined path, and the regulation of the sediment flow state in the high sediment concentration river is realized.

[0047] Further, the water surface suspended sediment concentration distribution in S1 is obtained, specifically: an unmanned ship equipped with a hyperspectral imager and a laser particle size analyzer is used to periodically cruise along the river section, the hyperspectral imager collects water surface spectral data, the spectral inversion model is constructed, the sediment particle size data measured by the laser particle size analyzer in real time is coded into a multi-scale feature vector, the spatial-spectral features of the spectral data are interacted through the feature fusion module, the particle size-spectral coupling feature matrix is combined to realize the layered inversion of the water surface suspended sediment concentration at different positions, and the inversion result of the water surface suspended sediment concentration at different positions is obtained.

[0048] Further, as shown in Figure 2 The spectral inversion model structure is a three-dimensional convolutional neural network, the input layer receives the continuous waveband spectral data collected by the hyperspectral imager, the spatial local features of the spectral data are extracted through the two-dimensional convolutional layer, and then the spatial-spectral joint features of the spectral data are captured through the three-dimensional convolutional layer; a double attention module is introduced, wherein the channel attention submodule obtains the spectral channel features through global average pooling and global maximum pooling, calculates the channel weights through a multilayer perceptron and performs weighted fusion, and the spatial attention submodule generates a spatial attention map based on the channel features and element-wise multiplies the spectral data feature map; the sediment particle size data measured by the laser particle size analyzer is coded into a feature vector through a fully connected layer, and the spectral features are interacted through a gated fusion unit, a multi-scale deconvolutional layer is set at the end of the model to restore the spatial resolution, a residual connection structure is used to fuse the features at different levels, and the inversion result of the water surface suspended sediment concentration at different positions is output by the output layer.

[0049] Further, in S2, when the numerical model of sediment transport in high sediment concentration river is constructed, an unstructured grid discretization method based on finite volume method is used to divide the river space region Ω into N control bodies, for each control body, the continuity equation of sediment transport is established: Wherein α is the porosity, C is the sediment concentration, t is the time, is the flow velocity vector, D is the diffusion coefficient, S is the sediment source and sink term, through the variational multiscale finite element technology, the large-scale grid is coupled with the small-scale river micro-topographic features, the riverbed topographic data is processed by adaptive grid division, the local grid refinement strategy is used in the complex area of water flow and riverbed morphology, and high-precision numerical simulation of sediment transport under different flow and water level conditions is realized.

[0050] Further, in S2, when simulating the sediment transport process under different flow and water level conditions, the dynamic boundary condition coupling technology is adopted to embed the flow Q and water level H as time-varying parameters into the model, and the flow-water level-sediment transport joint control equation is constructed: wherein is the velocity vector, ρ is the fluid density, p is the pressure, v is the kinematic viscosity coefficient, is the external force term, is the momentum source term caused by sediment transport, the probability distribution modeling of flow and water level fluctuation under different working conditions is realized through the Monte Carlo random sampling method, the synchronous simulation of sediment transport process under multiple working conditions is realized, and the sediment transport process under different flow and water level conditions is obtained by capturing the convection-diffusion trajectory of sediment particles.

[0051] Further, in S3, the density peak clustering algorithm is adopted to divide the region combined with the space-time trajectory characteristics, the labeled virtual sediment particle motion trajectory is constructed into a three-dimensional space trajectory matrix T n×m wherein n is the particle number, and m is the time step; the local density of each particle trajectory point is calculated wherein χ is the indicator function, d ij is the Euclidean distance between particles i and j, d c is the cut-off distance, the particles with high local density and far distance from higher density points are selected as the clustering centers, combined with the sediment concentration distribution data, the concentration threshold C th1 and C th2 , C th1 >C th2 , the region with local density greater than ρ high and concentration greater than C th1 is divided into the high-concentration sediment flow core area, the region with local density between ρ mid and ρ high and concentration between C th2 and C th1 is divided into the transition area, and the remaining region is divided into the low-concentration diffusion area, the boundary points of each region are connected to determine the boundary of each region.

[0052] Further, in S4, when extracting the center line of the high-concentration sediment flow core area, the combination algorithm based on topological skeleton extraction and dynamic programming is adopted, first, the two-dimensional plane projection of the high-concentration sediment flow core area is converted into a binary image, the morphological erosion operation is used to peel off the region boundary layer by layer until the topological skeleton S with a single-pixel width is obtained, and the cost function J= is constructed wherein (x i , y i ) is the skeleton node coordinates, d j is the shortest distance from the node to the core area boundary, ω s is the weight coefficient, n sN is the total number of nodes, m s N is the total number of paths, and the path is taken as the center line of the core area of the high-concentration sediment flow, that is, the main path of the sediment flow, by searching for the path that minimizes the cost function J on the skeleton S through the dynamic programming algorithm.

[0053] Further, in S4, when establishing the functional relationship between the swing amplitude of the main path and the flow, water level, and river bed roughness, the space-time joint feature extraction and multivariate nonlinear regression model is adopted. First, the swing trajectory of the main path in different hydrological periods is discretized in space and time to extract the time sequence feature vector of the swing amplitude and the spatial distribution feature matrix M, the river bed roughness coefficient n r , the flow Q, and the water level H are introduced as independent variables to construct a multivariate nonlinear function where α i , β i , γ i , and δ i are the corresponding parameter coefficients, and the model parameters are optimized to minimize the mean square error between the observed value and the predicted value of the swing amplitude A high-precision functional relationship between the swing amplitude of the main path and the flow, water level, and river bed roughness is established.

[0054] Further, in S5, the flow, water level, river bed roughness, and corresponding main path swing amplitude in the historical flood data are taken as training samples to construct a long short-term memory network and function relationship fusion model. The flow Q pred , water level H pred , and river bed roughness are calculated through the established function relationship A = f(Q, H, n r ) to obtain the initial swing amplitude prediction value A init , which is then combined with to form a four-dimensional input vector, which is input into the LSTM model for time series feature mining. The attention mechanism is introduced in the model to calculate the weight ω i of each input feature, and the final swing amplitude prediction value is obtained by weighted summation where X i is the input vector element, and the final swing amplitude prediction value is obtained. Combined with the main path swing center line offset algorithm, the predicted swing amplitude is converted into the spatial position coordinates of the main path of the sediment flow under different frequency floods to obtain the prediction result.

[0055] Further, in S5, according to the prediction result, the sediment flow state is regulated by the predicted sediment flow main path offset trend to construct a river channel topography regulation cost function where L is the target length and width of the predetermined path, L j , Wj is the actual length and width of the current path, λ j is a weight coefficient, which dynamically reshapes the local terrain of the river channel to achieve efficient transport of sediment flow along the predetermined path.

[0056] This embodiment describes in detail that through multi-source data fusion and intelligent algorithm, the precise simulation and regulation of high-sediment-content river channel sediment flow regime are realized, dynamic sediment concentration data are obtained by using an unmanned ship equipped with a hyperspectral imager and a laser particle size analyzer, a three-dimensional fluid dynamics model and an unstructured grid technology are combined to simulate the sediment transport process with high precision; by marking the virtual sediment particle trajectory, the density peak clustering algorithm is used to dynamically divide the flow regime area, the center line of the high-concentration core area is extracted and a swing law function is established, the LSTM model and the terrain regulation cost function are fused to realize the prediction and dynamic reshaping of the main path of sediment flow under different frequency floods, and the problems of low simulation accuracy and regulation lag of traditional technologies are solved, thereby providing an efficient and precise technical solution for river flood control, navigation and ecological management.

[0057] Based on embodiment 1, this embodiment describes in detail the specific implementation effects of the present application, specifically:

[0058] In the spatial distribution of sediment concentration, the present application technology and the existing conventional numerical simulation technology (such as the method based on the MIKE series or EFDC model) show significant differences. The present application periodically cruises by using an unmanned ship equipped with a hyperspectral imager and a laser particle size analyzer, and uses a three-dimensional convolutional neural network spectral inversion model to realize the layered inversion of the suspended sediment concentration on the water surface, which can accurately capture the subtle changes of the sediment concentration in the vertical and horizontal directions. Taking a typical bend channel area as an example, as shown in Figure 3 , Figure 3 a, the present application technology can clearly show that the sediment concentration at 0.5m from the riverbed bottom is 32.45kg / m 3 , and the concentration at 0.1m below the surface of the water surface is 18.72kg / m 3 , Figure 3 The concentration gradient changes from the concave bank to the convex bank in b, which is 28.65kg / m 3 (concave bank)→22.34kg / m 3 (main flow area)→15.17kg / m 3 (convex bank), and under the action of the bend circulation, a local high-concentration core area (35.28kg / m 3 ) is formed near the concave bank deep groove area, and the root mean square error with the measured data is only 1.23kg / m 3 .

[0059] The existing conventional numerical simulation technology has poor boundary condition processing for complex terrain (such as curved shoal, river bed protrusion) due to the use of fixed grid resolution and empirical parameter setting, resulting in large deviation of simulated sediment concentration value from actual distribution, such as Figure 4 As shown, the simulation results of the same curved area show that the concentration at 0.5 m from the river bed bottom is 27.10 kg / m 3 , the surface layer concentration is 21.45 kg / m 3 , the lateral concentration gradient is only 25.30 kg / m 3 → 23.80 kg / m 3 → 20.50 kg / m 3 , and the high concentration core area of the concave bank deep groove cannot be captured (the simulation value is only 29.50 kg / m 3 ), and the root mean square error with the measured data reaches 5.87 kg / m3. In addition, in the area where the flow changes sharply (such as the downstream of the confluence point of the tributary within 100 m), the present application can identify the sediment concentration pulse zone (concentration fluctuation range 22.54-28.97 kg / m 3 ) formed due to the sudden change of flow velocity, while the existing technology only presents a smooth concentration transition zone (simulation value 24.20-26.10 kg / m 3 ), which cannot reflect the concentration gradient mutation under the real flow state. The comparison results show that the present application is significantly superior to the existing technology in terms of the accuracy of the description of the spatial distribution of sediment concentration, the ability to capture subtle features, and dynamic adaptability through multi-source data fusion and adaptive grid refinement.

[0060] In the sediment flow area, the present application and the traditional fixed threshold clustering method (such as K-Means) show significant differences. The present application adopts the method of density peak clustering combined with the space-time trajectory characteristics, which can dynamically integrate the space-time continuity of the motion trajectory of the sediment particles and the hydrodynamic parameters, realize the accurate division of the high concentration core area, the transition area and the low concentration diffusion area, and take a typical mixed beach river area as an example. Based on the three-dimensional space trajectory matrix T 200×50 (200 particles, 50 time steps), the local density p i is calculated, the cut-off distance d c = 15, 23 m, the concentration threshold C th1 = 28.54 kg / m 3 and C th2 = 15.67 kg / m 3 , the area with local density p high > 38.65 and concentration higher than C th1 is divided into high concentration core area (accounting for 22.4%), the boundary of which accurately fits the main flow axis, and the coincidence degree with the water flow area with flow velocity greater than 1.85 m / s reaches 92.3%; the transition area (local density pmid = 19.32 ~ 38.65, concentration 15.67 ~ 28.54 kg / m 3 ) then clearly defines the mainstream and the interaction zone of the beach, the area ratio is 37.8%; the remaining area is a low concentration diffusion zone, which is consistent with the space of the slow flow area of the river beach, and the spatial coincidence degree is 89.7%.

[0061] The traditional fixed threshold clustering method only depends on a single index of concentration (such as a fixed threshold C fix = 20 kg / m 3 ) division, in the contrast results of the same area, the high concentration area is mistakenly expanded to the beach area (area ratio 35.1%), and the coincidence degree with the actual mainstream area is only 65.4%, and the transition flow area with a flow rate of 0.5-1.2 m / s is all classified into the low concentration area, resulting in a transition area ratio of only 18.9%. In the confluence scene of the tributary (confluence angle 32.5°, main tributary flow ratio 1:0.6), the tongue-shaped high concentration core area (length 128.4 m, width 25.7 m) formed by the density current can be identified by the technology, and the boundary angle with the flow line is 18.3°; while the traditional method misjudges this area as a transition area, and the boundary angle is 45.2°, and it fails to capture the local low concentration vortex (concentration 12.4 kg / m 3 , area 328.5 m 2 ) formed by the circular flow. In addition, through dynamic updating of the concentration threshold and trajectory features, the technology can achieve dynamic and accurate division of the sediment flow area in flood period (flow 850.3 m 3 / s) and dry period (flow 120.7 m 3 / s), and the core area position deviation errors are 1.87 m and 0.92 m respectively, while the traditional method has a deviation error of 8.54 m and 4.31 m due to the fixed threshold. The comparison shows that the technology solves the problems of "fuzzy boundary, flow state misjudgment, and poor adaptability of working conditions" of the traditional method through spatiotemporal feature coupling and dynamic threshold optimization, and realizes dynamic and accurate division of the sediment flow area.

[0062] In the comparison chart of the main path swing range and prediction accuracy, the technology based on topological skeleton extraction and multivariate nonlinear regression model, combined with long short-term memory network (LSTM) and attention mechanism, realizes dynamic and accurate prediction of the main path swing, as shown in Figure 5 , taking a typical wandering river as an example, in the hydrological cycle of 2023 (flow fluctuation range 82.4-785.6 m3 / s, water level 3.25-8.74 m, river bed roughness 0.021-0.038), the root mean square error (RMSE) of the main path swing amplitude predicted by the technology and the measured value is 1.87 m, while the RMSE of the traditional empirical formula method is 8.54 m. In the flood period (flow 720.3 m 3 / s, water level 8.12m), the maximum offset of the main path predicted by this application is 15.23m, the actual measured value is 16.01m, the error is only 4.87%; the traditional method predicts a value of 24.50m, the error is 53.02%.

[0063] From long-term series comparisons, the prediction bias of the proposed technology shows a stable convergence trend across different hydrological cycles. Taking monitoring data from 2018 to 2023 (60 months) as an example, the average spatial deviation between the predicted main path centerline position and the measured trajectory is 2.34m, and this deviation is even lower during the dry season (flow ≤ 150m³ / h). 3 / s) and during the middle water period (flow rate 150-400 m³ / s) 3 The deviations of the flow rate ( / s) were 1.12m and 1.98m respectively, demonstrating high adaptability to low flow conditions; the average deviation of the traditional empirical formula method reached 9.76m, and during the middle water period, the deviation increased sharply to 12.45m due to neglecting the dynamic changes in riverbed roughness. Figure 6 As shown, under the design flood condition with a frequency of 5% (predicted flow rate 850.3 m³ / s), 3 / s, water level 9.05m), the main path swing amplitude predicted by the fusion model of functional relationship and LSTM in this application is 22.48m, which is 36.85% lower than the traditional method (35.60m), and the agreement with the physical model test results (23.12m) is 97.23%.

[0064] The comparison chart also shows that the technology of this application can accurately capture the spatiotemporal characteristics of the main path oscillation, such as... Figure 7 As shown, in the curved section (radius of curvature R = 300.5 m), the rate at which the main path shifts towards the concave bank with increasing flow is 0.028 m / (m²). 3 The predicted velocity is 0.011 m / (m² / s), while the traditional method, which does not consider the influence of bend circulation, predicts a velocity of only 0.011 m / (m² / s). 3 The current method, which uses a constant flow rate ( / s), results in a significant underestimation of the offset during high flow rates. Furthermore, this application employs an attention mechanism, assigning weights of 0.45, 0.32, and 0.23 to flow rate, water level, and roughness, respectively, dynamically adjusting the contribution of each factor to the oscillation. In contrast, the traditional formula uses fixed weights (0.3:0.3:0.4), failing to reflect the changes in dominant factors under different operating conditions. In summary, this application's technology, through multi-model fusion and spatiotemporal feature decoupling, solves the problems of poor adaptability to operating conditions, lagging dynamic response, and insufficient multi-factor coupling analysis in traditional methods, improving prediction accuracy by over 70% and providing more reliable technical support for river sediment regulation.

[0065] The embodiment details that the application can capture vertical and lateral concentration gradient and concentration characteristics of special areas more accurately by multi-source data fusion and adaptive model, and is closer to real distribution than existing numerical simulation technology; the contrastive diagram of sediment flow state region division shows that the application can accurately define the boundaries of each flow state region and reflect the evolution of the boundaries with hydrodynamic conditions by combining space-time trajectory and dynamic clustering method, and overcomes the defects of traditional fixed threshold method such as fuzzy boundary and flow state misjudgment; the contrastive diagram of main path swing prediction proves that the application significantly improves the prediction accuracy of the swing amplitude and position of the main path by multi-model fusion and space-time feature analysis, and the adaptability to different hydrological periods and extreme working conditions is far superior to that of the traditional empirical formula method, and the application technology realizes breakthroughs in concentration description, region division and prediction accuracy, and provides more efficient and reliable technical means for river sediment transport regulation.

[0066] The above is only the preferred embodiment of the application, and does not limit the protection scope of the application. The application can have various changes and variations for those skilled in the art; any change, modification, replacement, integration and parameter change of the embodiments within the spirit and principle of the application, which realizes the same function without departing from the principle and spirit of the application, falls within the protection scope of the application.

Claims

1. A method for determining the path of sediment flow in a high sediment concentration river, characterized by, The method comprises the following steps: S1, acquiring the riverbed topography, flow velocity and flow direction data, and water surface suspended sediment concentration distribution of the high-sediment river channel; S2, importing the collected riverbed topography data, flow velocity and flow direction data, and suspended sediment concentration distribution data into a three-dimensional fluid dynamics model, discretely processing the river channel by using an unstructured grid, constructing a high-sediment river channel sediment transport numerical model, and simulating the sediment transport process under different flow rates and water levels; S3, according to the sediment transport numerical model, marking a large number of virtual sediment particles in the model, tracking the motion trajectory under the action of water flow, dividing the high-concentration sediment flow core area, the transition area and the low-concentration diffusion area according to the sediment concentration distribution data, and determining the boundaries of each area; S4, according to the divided sediment flow area, extracting the center line of the high-concentration sediment flow core area as the main path of the sediment flow, analyzing the swing law of the main path in different hydrological periods, and establishing the functional relationship between the swing amplitude of the main path and the flow rate, water level and riverbed roughness; S5, combining historical flood data and future hydrological prediction information, using the functional relationship to predict the main path of the sediment flow under different frequency floods, and guiding the sediment flow to transport along the predetermined path by changing the local topography of the river channel according to the prediction result, so as to realize the regulation and control of the sediment flow state of the high-sediment river channel.

2. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 1, wherein, In the S1, the water surface suspended sediment concentration distribution is obtained by using an unmanned ship equipped with a hyperspectral imager and a laser particle size analyzer to periodically cruise along the river section, the hyperspectral imager collects water surface spectral data, the spectral inversion model is constructed, the sediment particle size data measured by the laser particle size analyzer in real time is coded into a multi-scale feature vector, the spatial-spectral features of the spectral data are interacted through the feature fusion module, the particle size-spectral coupling feature matrix is combined to realize the hierarchical inversion of the water surface suspended sediment concentration at different positions, and the inversion results of the water surface suspended sediment concentration at different positions are obtained.

3. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 2, wherein, The spectral inversion model structure is a three-dimensional convolutional neural network, the input layer receives the continuous waveband spectral data collected by the hyperspectral imager, the spatial local features of the spectral data are extracted through a two-dimensional convolutional layer, and then the spatial-spectral joint features of the spectral data are captured through a three-dimensional convolutional layer; A double attention module is introduced, wherein a channel attention submodule obtains spectral channel features through global average pooling and global maximum pooling, calculates channel weights through a multilayer perceptron, and performs weighted fusion, a spatial attention submodule generates a spatial attention map based on the channel features, and element-wise multiplication is performed between the spatial attention map and the spectral data feature map; the sediment particle size data measured by the laser particle size analyzer are coded into a feature vector through a fully connected layer, cross-modal interaction is performed between the spectral features and the feature vector through a gating fusion unit, a multi-scale deconvolutional layer is arranged at the end of the model to restore the spatial resolution, a residual connection structure is used to fuse the features at different levels, and the output layer outputs the inversion results of the water surface suspended sediment concentration at different positions.

4. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 1, wherein, In the S2, when constructing the numerical model of sediment transport in high sediment concentration river, the unstructured grid discretization method based on finite volume method is adopted to divide the river space region Ω into N control bodies. For each control body, the continuity equation of sediment transport is established: Where α is the porosity, C is the sediment concentration, t is the time, is the flow velocity vector, D is the diffusion coefficient, S is the sediment source and sink term. Through the variational multiscale finite element technology, the large-scale grid is coupled with the small-scale river micro-topographic features. The adaptive grid division processing is performed on the riverbed topographic data. The local grid refinement strategy is adopted in the areas with severe flow changes and complex riverbed morphology to realize high-precision numerical simulation of sediment transport under different flow and water level conditions.

5. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 1 or 4, wherein, When simulating the sediment transport process under different flow and water level conditions in S2, the dynamic boundary condition coupling technology is adopted to embed the flow Q and water level H as time-varying parameters into the model, and the flow-water level-sediment transport joint control equation is constructed: where is the velocity vector, ρ is the fluid density, p is the pressure, v is the kinematic viscosity coefficient, is the external force term, is the momentum source term caused by sediment transport, the probability distribution modeling of flow and water level fluctuation under different working conditions is realized through the Monte Carlo random sampling method, the synchronous simulation of multi-working condition sediment transport process is realized, and the sediment transport process under different flow and water level conditions is obtained by capturing the convection-diffusion trajectory of sediment particles.

6. The method for determining the path of sediment flow regime in a high sediment concentration river channel as claimed in claim 1, wherein, The density peak clustering algorithm is combined with the space-time trajectory characteristics to divide the region in the S3, and the marked virtual sediment particle motion trajectory is constructed into a three-dimensional space trajectory matrix T according to the time sequence n×m , wherein n is the particle number, and m is the time step; the local density of each particle trajectory point is calculated , wherein χ is an indicator function, d ij is the Euclidean distance between particles i and j, d c is the truncation distance, the particles with high local density and far distance from higher density points are selected as the clustering centers, the concentration threshold C th1 and C th2 are set according to the concentration distribution data of the sediment, C th1 >C th2 , the region with local density greater than ρ high and concentration greater than C th1 is divided into the high-concentration sediment flow core area, the region with local density between ρ mid and ρ high and concentration between C th2 and C th1 is divided into the transition area, and the remaining region is divided into the low-concentration diffusion area, the boundary points of the regions are connected, and the boundaries of the regions are determined.

7. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 1, wherein, In the S4, when extracting the center line of the high-concentration sediment flow core area, a combined algorithm based on topological skeleton extraction and dynamic programming is adopted. First, the two-dimensional plane projection of the high-concentration sediment flow core area is converted into a binary image, and a morphological erosion operation is used to peel off the regional boundary layer by layer until a single-pixel-wide topological skeleton S is obtained. A cost function is constructed where (x i ,y i ) are the skeleton node coordinates, d j is the shortest distance from the node to the core area boundary, ω s is the weight coefficient, n s is the total number of nodes, and m s is the total number of paths. The path that minimizes the cost function j is searched on the skeleton S by a dynamic programming algorithm, and the path is taken as the center line of the high-concentration sediment flow core area, i.e., the main path of the sediment flow.

8. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 7, wherein, In the function relationship between the main path swing amplitude and the flow, water level and river bed roughness in S4, the space-time combined feature extraction and multivariate nonlinear regression model is adopted. First, the swing trajectory of the main path in different hydrological periods is discretized in space-time to extract the time series feature vector of the swing amplitude and the spatial distribution feature matrix M. The river bed roughness coefficient n r , the flow Q and the water level H are introduced as the independent variables to construct a multivariate nonlinear function where α i , β i , γ i and δ i are the corresponding parameter coefficients. The model parameters are optimized to minimize the mean square error between the observed swing amplitude and the model predicted value The function relationship between the high-precision main path swing amplitude and the flow, water level and river bed roughness is established.

9. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 8, wherein, In S5, historical flood data, including flow rate, water level, riverbed roughness, and corresponding main path swing amplitude, are used as training samples to construct a fusion model of a long short-term memory network and functional relationships. This model is then used to predict the flow rate Q in future hydrological information. pred Water level H pred Riverbed roughness By establishing the functional relationship A = f(Q, H, n) r Calculate the predicted initial swing amplitude A init Then combine it with Q pred H pred , A four-dimensional input vector is formed and fed into an LSTM model for temporal feature mining. An attention mechanism is introduced into the model to calculate the weight ω of each input feature. i By weighted summation Where X i Using the input vector elements, the final predicted swing amplitude is obtained. Combined with the main path swing centerline offset algorithm, the predicted swing amplitude is converted into the spatial coordinates of the main path of sediment flow under different frequency floods, and the prediction results are obtained.

10. The method for determining the path of sediment flow regime in a high sediment concentration river channel according to claim 1 or 9, wherein, In the step S5, the river terrain regulation cost function is constructed according to the predicted sediment flow main path deviation trend when the sediment flow state is regulated according to the prediction result Wherein L and W are the target length and width of the predetermined path j , W j L and W are the actual length and width of the current path j λ is a weight coefficient, and the dynamic reshaping of the local terrain of the river is performed to realize the efficient transport of the sediment flow along the predetermined path.

Images