A method for automatically sampling and monitoring the content and particle size of seawater drifting quicksand

Abstract

The present application provides a method for automatically sampling and monitoring the content and particle size of running sand in seawater, comprising: using high-resolution seabed topography mapping and flow velocity measurement equipment to collect topographic data of the target sea area and flow velocity data of seawater; identifying the flow pattern of seawater according to the collected flow velocity data, and predicting the behavior of sand particles according to the flow pattern, wherein the flow pattern includes information on the high and low flow velocity of seawater and information on direction changes; obtaining the movement pattern of sand particles under different flow velocity conditions according to the flow pattern, wherein the movement pattern includes sand particle suspension data, sand particle sedimentation data, and sand particle migration data; obtaining the accumulation and dispersion law of sand particles according to the movement behavior of different types of sand particles under changing flow velocity conditions; formulating a sensor layout strategy according to the simulation results of the CFD simulation method, and designing a sensor network according to the sensor layout strategy.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a method for automatically sampling and monitoring the content and particle size of seawater quicksand. Background Art

[0002] When deploying sensor networks in marine environments to monitor seawater velocity and sand movement, there are technical challenges brought about by the complex interaction between sensor density and velocity variables. First, the marine environment is extremely complex and involves a variety of physical processes. The accurate monitoring of these processes requires highly complex sensor networks and data processing technologies. Second, the choice of sensor density directly affects the accuracy and reliability of data acquisition. Too low density may lead to insufficient spatial resolution of data and fail to accurately capture local changes in velocity; while too high density may lead to data redundancy and energy consumption problems, and fail to provide a more scientific and systematic technical support for seabed resource management and environmental protection. Finally, changes in seawater velocity have an important impact on the performance of sensor networks. Under high velocity conditions, the sampling frequency of sensors needs to be increased accordingly to ensure the temporal resolution of the data, but it also faces limitations in energy consumption and data transmission bandwidth. In addition, the motion characteristics of sand particles will also change under different velocity conditions, further increasing the difficulty of data interpretation and modeling. Summary of the invention

[0003] The present invention provides a method for automatically sampling and monitoring the content and particle size of seawater drifting quicksand, which mainly includes:

[0004] Use high-resolution seabed topography mapping and current velocity measurement equipment to collect topographic data of the target sea area and seawater flow velocity data;

[0005] Identifying the flow pattern of seawater based on the collected flow velocity data, and predicting the behavior of sand particles based on the flow pattern, wherein the flow pattern includes information on the speed and direction of seawater flow;

[0006] According to the flow pattern, a movement pattern of sand particles under different flow velocity conditions is obtained, wherein the movement pattern includes sand particle suspension data, sand particle settling data and sand particle migration data;

[0007] Collecting sand samples and classifying the sand samples to obtain physical properties of the sand, wherein the physical properties include particle size, density and shape, and obtaining behavioral characteristics of the sand samples according to the different physical properties of the sand;

[0008] According to the movement behavior of different types of sand particles under changing flow velocity conditions, the accumulation and dispersion rules of sand particles are obtained;

[0009] According to the movement pattern data and the accumulation and dispersion law data of the sand particles, the sensor deployment density is adjusted to obtain a sensor deployment density scheme, wherein the sensor deployment density includes the layout, spacing and number of sensors;

[0010] Formulate a sensor deployment strategy based on the simulation results of the CFD simulation method, and design a sensor network based on the sensor deployment strategy;

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for automatically sampling and monitoring the content and particle size of seawater drifting quicksand. This method uses high-resolution seabed topography mapping and flow velocity measurement equipment to improve the accuracy and efficiency of seawater drifting quicksand data, so that marine environmental monitoring and marine engineering projects can more accurately predict sand behavior and conduct environmental assessments. The physical properties of sand particles are combined with the flow pattern to describe in detail the behavior characteristics of sand particles under specific seawater flow conditions. The behavior analysis helps determine the accumulation and dispersion laws of sand particles, thereby effectively monitoring the dynamic changes of seabed sand particles. The sensor deployment density is adjusted according to the sand particle movement pattern data and the accumulation and dispersion law data. This strategy is based on the results of CFD simulation methods, so that the design of the sensor network can be more accurate and efficient.

[0013] The understanding of the movement patterns of ocean sand particles has also been deepened, and reliable and efficient ocean dynamic environment monitoring solutions have been obtained, which in turn helps the management and protection of marine resources. By optimizing the deployment of sensors, the present invention significantly improves the efficiency of resource utilization and the response speed of the monitoring system, thereby providing more scientific and technological support for seabed resource management and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flow chart of a method for automatically sampling and monitoring the content and particle size of seawater quicksand.

[0015] Figure 2 It is a schematic diagram of an automatic sampling and monitoring method for the content and particle size of seawater quicksand according to the present invention.

[0016] Figure 3 It is another schematic diagram of the automatic sampling and monitoring method of the content and particle size of seawater quicksand according to the present invention. DETAILED DESCRIPTION

[0017] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0018] like Figure 1-3 In this embodiment, a method for automatically sampling and monitoring the content and particle size of seawater quicksand may specifically include:

[0019] Step S101, using high-resolution seabed topography mapping and current velocity measurement equipment to collect topographic data of the target sea area and seawater flow velocity data.

[0020] Utilize the high-resolution seabed topography mapping equipment and the multi-beam sounding method to obtain high-precision seabed topography data of the target sea area, the mapping accuracy reaches the centimeter level, and the mapping range covers the entire target sea area; collect seabed topography digital elevation model data by optimizing the mapping path and sampling frequency; use high-precision flow velocity measurement equipment and the Doppler flow profiler method to obtain flow velocity data at different depths of the target sea area, the accuracy of the flow velocity data reaches the millimeter / second level, and the measurement range covers the entire water body; obtain three-dimensional flow field information, and establish a hydrodynamic model based on the three-dimensional flow field information, the three-dimensional flow field information includes flow velocity size, flow velocity direction and flow velocity change; perform data fusion processing on the acquired seabed topography data and flow velocity data, and generate numerical simulation terrain data and flow field data through spatial interpolation and gridding methods; adjust the grid resolution according to simulation requirements optimization; based on the fused terrain data and the flow rate data, the hydrodynamic model is used to simulate the water flow movement in the target sea area; the hydrodynamic model is verified with the measured data in combination with the seabed terrain, tides and wind stress, and the parameters of the hydrodynamic model are continuously optimized; according to the hydrodynamic simulation results, the water flow characteristics of the target sea area are analyzed to obtain the flow rate distribution, flow direction change and vortex structure; through data visualization technology, a water flow movement image is generated; the hydrodynamic simulation results are coupled and analyzed with the seabed terrain data to display the influencing factors of the seabed terrain on the water flow movement; through numerical experiments, the water flow changes under different terrain conditions are simulated to display the interaction mechanism between terrain and hydrodynamics; based on the seabed terrain data and the hydrodynamic simulation results, a marine engineering design optimization platform is set up to integrate terrain surveying and mapping data, flow rate measurement data and numerical simulation data.

[0021] Specifically, high-resolution seabed topography mapping equipment is used to obtain high-precision seabed topography data of the target sea area through multi-beam bathymetry technology. The mapping accuracy reaches the centimeter level, and the mapping range covers the entire target sea area. By optimizing the mapping path and sampling frequency, the mapping efficiency is improved, and high-quality seabed topography digital elevation model data is obtained. On this basis, high-precision flow velocity measurement equipment, such as acoustic Doppler flow profilers, is used to reasonably arrange the measurement position and range according to the characteristics of the seabed topography to obtain three-dimensional flow field information at different depths in the target sea area, including flow velocity size, flow direction, flow velocity change, etc. The flow velocity accuracy reaches the millimeter / second level, providing accurate boundary conditions for subsequent hydrodynamic models. The acquired seabed topography and flow velocity data are fused and processed, and high-resolution structured grid data suitable for numerical simulation is generated through methods such as Kriging interpolation. The grid size is adaptively adjusted according to the complexity of the seabed topography and the simulation accuracy requirements, while ensuring the calculation efficiency and taking into account the stability and accuracy of the model. Based on the fused terrain and velocity data, the three-dimensional non-constant Reynolds average Navier-Stokes equation model is used to simulate the water flow movement in the target sea area with high precision. The model fully considers the undulation of the seabed terrain, tides, wind stress and other influencing factors. Through comparison and verification with the measured data, the model parameters are continuously optimized to improve the reliability and accuracy of the simulation. The simulation results include detailed hydrodynamic characteristic information such as velocity distribution, flow direction change, vortex structure, etc. at different times and depths in the target sea area. Using high-performance visualization tools such as ParaView, the hydrodynamic simulation results are rendered and animated in three dimensions. Through streamlines, isosurfaces, slices and other visualization methods, intuitive, multi-angle water flow images and videos are generated to assist researchers in in-depth analysis of the hydrodynamic characteristics of the sea area and understand the interaction mechanism between the seabed terrain and water flow. The hydrodynamic simulation results are further coupled with high-resolution seabed topography data for analysis, and the influence of the slope, undulation and other characteristics of the seabed topography on the velocity and direction of the near-bottom water flow is studied. At the same time, the scouring and shaping of the seabed topography by the water flow is analyzed. Through the combination of numerical simulation and experimental verification, the complex feedback mechanism between the seabed topography and hydrodynamics is quantitatively revealed, providing more scientific and detailed theoretical guidance for marine engineering design. Based on the above seabed topography and hydrodynamic data, a design optimization platform for different types of marine engineering is developed. Taking offshore wind power engineering as an example, the platform integrates multi-source data such as topography, water depth, and flow rate required for wind turbine foundation site selection and underwater cable laying. Through intelligent optimization methods such as genetic algorithms and particle swarm algorithms, it automatically generates wind turbine layout plans and construction plans that meet engineering constraints, and conducts multi-objective evaluation and optimization. At the same time, the platform can also provide customized terrain and hydrodynamic data processing, solution optimization and other functions for other marine projects such as cross-sea bridges and submarine tunnels, greatly improving the intelligence level and comprehensive benefits of engineering design.In actual seabed topography mapping, the SeaBat7125 multi-beam bathymetric system can be used. The system has a transmission frequency of 200-400kHz, the number of beams can reach 512, and the bathymetric accuracy is better than 5cm. By rationally planning the ship speed and route, millions of high-precision water depth point data can be obtained per square kilometer. For velocity measurement, RDI's 300kHz WorkHorse monitoring ADCP can be selected, with a velocity measurement range of ±5m / s and an accuracy of ±0.5%. By deploying multiple ADCPs at different water depths, three-dimensional flow field data with high temporal and spatial resolution can be obtained. In data fusion processing, the improved Kriging interpolation algorithm is used, anisotropic variogram is introduced, and the major and minor axis ratio of the search ellipsoid is adaptively adjusted according to the density of data point distribution, generating an unstructured triangular grid with a grid resolution of 1-10m. In the hydrodynamic numerical simulation, the open source FVCOM model is used. It solves the three-dimensional non-steady Reynolds averaged Navier-Stokes equations based on the finite volume method. The semi-implicit and modesplit algorithms are used to improve the computational efficiency. At the same time, the Mellor-Yamada 2.5 level turbulence closure scheme is introduced to improve the applicability of the model. The model verification uses the measured tide level and flow field data from July 2019 to June 2020 for one year. The results show that the root mean square errors of tide level and flow velocity are 15cm and 10cm / s, respectively. ParaView5.8 version is used to visualize the simulation results, and the LIC texture lighting rendering based on velocity gradient is realized through Python scripts, and high-resolution animations are generated using ImageMagick software. In the coupled analysis of terrain and hydrodynamics, the areas with significant changes in seabed scouring and silting are extracted, and the terrain slope, curvature and other parameters are calculated. Pearson correlation analysis is performed with the vertical shear stress and turbulent viscosity of the bottom water flow. It is found that in areas with a slope greater than 5°, the shear stress of the water flow can reach 10N / m. 2 , which is 2-3 times that of flat areas. Finally, in the offshore wind power design optimization platform, an improved multi-island genetic algorithm is used, with wind turbine efficiency, construction cost, cable length, etc. as optimization targets, and multiple wind farm layout schemes are optimized in parallel through crossover mutation and immigration operations. After 500 generations of evolution, the wind turbine layout with the best comprehensive benefits is obtained, with a 3% increase in unit power generation and a reduction in construction costs.

[0022] Step S102, identifying the flow pattern of seawater according to the collected flow velocity data, and predicting the behavior of sand particles according to the flow pattern, wherein the flow pattern includes information on the velocity of seawater and information on changes in direction.

[0023] According to the collected flow velocity data, the moving average filtering method and the outlier elimination method are used to preprocess the flow velocity data to obtain the preprocessed flow velocity data; the preprocessed flow velocity data are unified to regular grid points through the interpolation algorithm to obtain regularized flow velocity data; based on the regularized flow velocity data, the empirical orthogonal decomposition method is used to identify the flow field mode to obtain the time coefficient and spatial distribution of each mode; by analyzing the time coefficient and spatial distribution, the dominant mode of flow velocity and flow direction change is determined, and the importance of each mode is measured by energy percentage; the flow field at different times is visualized in three dimensions using the streamline tracing method and the flow tube visualization method, and the spatiotemporal evolution process of the flow velocity size and direction is presented through animation to show the dynamic characteristics of the flow pattern; the time coefficient obtained by the empirical orthogonal decomposition method is used as Based on the sample characteristics, the K-means clustering method is used to classify the flow patterns, and the silhouette coefficient is used to evaluate the clustering effect. The time proportion and conversion law of various flow patterns are analyzed to show the dynamic change trend of the flow field. The numerical simulation is analyzed for the identified flow pattern, and the velocity and pressure information of the flow field are obtained by solving the Reynolds-averaged Navier-Stokes equation. According to the sand entrainment process of water flow and the sedimentation process of sand particles, the transport behavior of sand particles under different flow patterns is simulated to obtain the suspended sediment concentration and scouring and silting distribution, and the influence mechanism of the flow pattern on the sediment dynamic process is determined. The quantitative relationship model between the flow pattern and the sand behavior is constructed by combining the flow pattern identification and numerical simulation results. The quantitative relationship model is trained and verified by the machine learning algorithm, and the sand behavior is predicted according to the flow velocity data observation data.

[0024] Specifically, the collected seawater flow field data are preprocessed, and moving average filtering is used to eliminate high-frequency noise. The Laida criterion is used to eliminate outliers exceeding 3 times the standard deviation to improve data quality. The irregular data are mapped to regular grid points through Kriging interpolation to facilitate subsequent analysis. Based on the preprocessed flow field data, the empirical orthogonal decomposition (EOF) is used to extract the main modes of the flow field. EOF decomposes the flow field into a linear combination of spatial functions and time functions to identify several modes with the largest contribution. By analyzing the spatial distribution characteristics of each mode and the corresponding time coefficient changes, the dominant mode of flow velocity and flow direction changes is determined. The flow field at different times is visualized in three dimensions using streamline tracing, flow tube visualization and other technologies, and the spatiotemporal evolution process of flow velocity and direction is presented through animation, which intuitively reveals the dynamic characteristics of the flow pattern. The time coefficient of each mode obtained by EOF is used as the sample feature vector, and the K-means clustering algorithm is used to classify the flow pattern. By calculating the distance from each sample to the cluster center, the cluster center is iteratively updated until the convergence condition is met. The clustering effect is evaluated by using the silhouette coefficient, and the time proportion and conversion law of various flow modes are analyzed to characterize the dynamic change trend of the flow field. Numerical simulation studies are carried out for the main flow modes identified. By solving the three-dimensional incompressible Reynolds-averaged Navier-Stokes equations, combined with boundary conditions and initial conditions, the finite volume method is used for discretization to obtain detailed information such as velocity and pressure of the flow field. The numerical simulation results are compared and verified with the measured data. By adjusting the turbulence model parameters and grid density, the model settings are continuously optimized to improve the simulation accuracy. According to the process of water flow entrainment and sand sedimentation, the transport behavior of sand particles under different flow modes is simulated to obtain the suspended sediment concentration and scouring and silting distribution, and the influence mechanism of the flow mode on the sediment dynamic process is determined; the Lagrangian particle tracking method is used to calculate the motion trajectory of a single sand particle in the turbulent flow field, while considering the collision between particles and the influence of the entrainment force. By tracking and simulating a large number of sand particles, key information such as suspended sediment concentration and scouring and silting distribution is obtained. The control effect of flow velocity, turbulence intensity and other parameters on the sand transport rate was analyzed, and the influence mechanism of flow pattern on sediment dynamic process was quantitatively revealed. Based on the analysis results of flow pattern recognition and numerical simulation, a machine learning algorithm was used to establish a mapping relationship model between flow pattern and sand behavior. The training samples were constructed with parameters such as flow velocity and flow direction as input features and suspended sediment concentration and erosion and siltation as output targets. The model was trained and verified using algorithms such as support vector machines and neural networks, and the model structure and parameters were continuously optimized to improve the prediction accuracy. The trained model was applied to engineering practice. According to the real-time flow velocity observation data, the possible sediment deposition problems were predicted, dredging plans were formulated in advance, and coastal protection, port maintenance and other work were guided. By establishing a quantitative relationship model between flow pattern and sand behavior, real-time early warning and dynamic control of sediment dynamic process were achieved.When preprocessing the seawater flow field data, the Savitzky-Golay filter with a 5-point cubic polynomial fitting can be used to eliminate high-frequency noise while retaining the trend characteristics of the original signal to the maximum extent. For outliers with obvious physical significance, such as extremely high flow velocity during typhoons, the threshold method can be used to distinguish and retain them to avoid the loss of important information. In EOF analysis, SVD singular value decomposition can be used for fast calculation, and the first few modes with a cumulative variance contribution rate of 90% are selected as the main features. The power spectrum analysis of the EOF time coefficient is performed, and the periodic characteristics of the flow pattern are judged by extracting the dominant frequency, such as the semi-diurnal cycle caused by tides and the annual cycle changes caused by monsoons. In numerical simulation, the SIMPLE algorithm can be used to solve the velocity-pressure coupling problem, and the algebraic multigrid method (AMG) can be used to accelerate convergence. For high Reynolds number flows, the LES large eddy simulation method can be used, and the sub-grid stress is closed using the Smagorinsky-Lilly model. Through parameter sensitivity analysis, the turbulence model constants and grid density are optimized to improve the computational efficiency while ensuring accuracy. In sediment dynamics simulation, the multi-particle size effect can be considered, and the NTLM non-equilibrium transport model can be used to describe the migration and diffusion process of suspended sediment concentration. By analyzing the spatial distribution of bed shear stress, the location and intensity of scouring and silting can be determined. The relationship between flow pattern and sand behavior is modeled using ensemble learning algorithms such as random forests, and the model hyperparameters are optimized through grid search to improve generalization ability. The established mapping relationship is applied to new monitoring data to predict the spatiotemporal distribution of sediment deposition, and compared with the measured terrain changes to further correct the model coefficients and achieve accurate engineering applications.

[0025] Step S103, obtaining the movement pattern of sand particles under different flow velocity conditions according to the flow pattern, wherein the movement pattern includes sand particle suspension data, sand particle settling data and sand particle migration data.

[0026] According to the identified flow pattern, the preset flow velocity conditions in the motion mode under different flow velocity conditions are extracted, and the preset flow velocity conditions include the average flow velocity and the turbulence intensity. The preset flow velocity conditions are used as the boundary conditions for analyzing the motion of sand particles, and the flow velocity conditions are mapped to the location of the sand particles through the interpolation method; the DSMC direct simulation Monte Carlo method is used to perform random tracking simulation on the motion trajectory of a single sand particle under different flow velocity conditions, and the statistical characteristics of the sand particles in the three motion modes of suspension, sedimentation and migration are obtained through simulation statistics of a large number of sand particles, and the statistical characteristics include suspension time, sedimentation velocity and migration distance; in the DSMC simulation, the influence of the particle characteristics of the sand particles on the motion of the sand particles is combined, and the response of the sand particle volume to the hydrodynamic force is corrected by introducing the equivalent particle size, and the sphericity and roundness parameters are used to characterize the influence of the sand particle shape on the vertical and lateral motion, and the quantitative relationship between the particle characteristics and the motion mode is established; the regulating effect of fluid viscosity and turbulence intensity on the motion of sand particles is analyzed, and the Reynolds number criterion is used to calculate the relationship between the particle characteristics and the motion mode. The flow state is judged, different mechanical models are used for the movement of sand particles under laminar and turbulent conditions, and the velocity pulsation under turbulent conditions is statistically analyzed to reveal the relationship between turbulence intensity and sand suspension probability; in the simulation of sand particle movement, the setting of boundary conditions and initial conditions is strictly controlled, and the logarithmic velocity distribution is used to describe the velocity profile in the near-bottom area, and the spectral method is used to generate a random velocity field that conforms to the energy distribution for the velocity pulsation under turbulent conditions; high-speed photography and particle image velocimetry are used to track and measure the movement of sand particles under the preset flow conditions to obtain the real movement data of sand particle suspension, sedimentation and migration, and the accuracy of the model is verified by comparative analysis with the DSMC simulation results, and the model parameters are corrected; the verified sand particle movement model is integrated into the two-dimensional water-sand coupling numerical simulation program, and the dynamic influence of the flow field on the sand particle movement is realized by the dynamic grid technology, and the scouring and silting distribution law under different flow conditions is obtained by collective statistical analysis of the migration trajectories of a large number of sand particles.

[0027] Specifically, according to the main flow modes identified above, the preset flow velocity conditions under each mode are extracted, including parameters such as the average flow velocity, flow direction change characteristics and turbulence intensity. The flow field information is mapped to the location of the sand particles through the nearest neighbor interpolation method to obtain the local flow field conditions that drive the movement of the sand particles. Experimental techniques such as high-speed photography and particle image velocimetry are used to track and record the movement trajectory of a single sand particle in a typical flow field, obtain parameters such as the instantaneous position, velocity and acceleration of the sand particles, analyze the influence of flow velocity and flow direction changes on the suspension height and migration speed of sand particles, and establish a quantitative relationship between water flow characteristics and sand particle movement parameters. In the analysis of sand particle movement characteristics, the particle characteristic parameters of sand particles, such as equivalent particle size, sphericity and roundness, are comprehensively considered, where sphericity is defined as the ratio of the particle volume to the volume of the circumscribed sphere, and roundness is defined as the ratio of the particle projection area to the area of ​​the circumscribed circle. Three-dimensional laser scanning technology is used to obtain sand particle shape parameters, and the mechanism of the influence of particle characteristics on sand particle suspension, sedimentation and migration is revealed through numerical simulation and experimental research. In the flow field condition analysis, the Reynolds number criterion is used to judge the transition from laminar flow to turbulent flow. The Reynolds number of 500 is used as the critical value. Different sand force models are used for laminar and turbulent states. The turbulent pulsation is analyzed by spectrum analysis and correlation length calculation to quantitatively describe the relationship between turbulence intensity and sand suspension probability. The DSMC direct simulation Monte Carlo method is used to randomly track and simulate the motion trajectories of a large number of sand particles under different flow field conditions. The velocity boundary conditions and the initial sand distribution are strictly controlled. The logarithmic distribution is used for the near-bottom velocity. The turbulent pulsation velocity is generated by the spectrum method. The sand particle start-up and retention caused by bed scouring are considered. Through the statistical analysis of a large number of sand particle motion trajectories, the conditions and laws of the transition of sand particle suspension, sedimentation and migration motion modes are obtained. The DSMC simulation results are verified by experimental data. The model accuracy is evaluated by indicators such as root mean square error and correlation coefficient. The mechanical model coefficients and calculation parameters are adjusted for key parameters such as sand particle position, velocity and motion trajectory to make the simulation results consistent with the experimental data. The verified sand movement model is coupled to the hydrodynamic simulation, and the dynamic grid technology is used to realize the bidirectional coupling effect of the flow field on the sand movement. The fluid-solid coupling equation is solved by the finite element method. When the sand migrates significantly, the grid is deformed and reconstructed to obtain the influence of sand migration on the flow field. Through the statistical analysis of a large number of sand movement trajectories, the distribution of suspended sediment concentration and the law of scouring and silting under different flow conditions are obtained, revealing the dynamic mechanism of coastal landform evolution under the coupling of water and sand. When extracting the preset flow conditions, the spectral clustering algorithm can be used to partition the flow field, and the number of cluster centers can be determined by the maximum modal curvature criterion to obtain the preset flow field under different flow conditions. When mapping the flow field information, the inverse distance weighted interpolation method is used, and the inverse of the distance from the location of the sand to each flow field node is used as the weight to obtain the average flow velocity and turbulent pulsation velocity of the local flow field.The motion process of a single sand grain was continuously filmed at a frequency of 1000 frames per second using a high-speed camera. The contour of the sand grain was extracted by the Canny edge detection algorithm. The velocity of the sand grain was calculated by the Lucas-Kanade optical flow method. Combined with the measurement data of the acceleration sensor, the motion parameters of the sand grain were fused and estimated by the Kalman filter algorithm. The three-dimensional shape of the sand grain was obtained based on the principle of laser triangulation. The sphericity and roundness were calculated by extracting 5000 sampling points. The correlation between the particle parameters and the force of the sand grain was analyzed by the mutual information method. A BP neural network model was constructed from the mapping of particle characteristics to the force coefficient. The network adopted a three-layer structure, and the hidden layer contained 20 neurons. The training was carried out by the gradient descent method, and the relative error was controlled within 5%. The large eddy simulation method was used to numerically calculate the turbulent field. The sub-grid stress was closed by the Smagorinsky model. The velocity pulsation that met the Karman spectrum distribution was generated by the spectral method. The flows with Reynolds numbers of 200, 500 and 1000 were simulated respectively, and the influence of turbulence intensity on the average suspension height of sand grains and the turbulent throwing frequency was analyzed. In the DSMC method, a hash table structure is used to store flow field information and sand positions, and the octree algorithm is used to achieve fast neighborhood retrieval. The motion trajectories of 1 million sand particles are calculated in parallel, with each time step of 0.1 seconds. The boundary condition adopts the mirror reflection model, and the distribution characteristics of sand trajectories are analyzed by statistical methods to extract the critical velocity conditions for the conversion of sand motion modes. Based on the DSMC simulation results, the principal component analysis method is used to screen the key factors affecting sand motion, and the model parameters are optimized by the experimental design method, so that the root mean square error is reduced by 20%, and the displacement correlation coefficient reaches above 0.9. In the fluid-solid coupling simulation, the ALE arbitrary Lagrangian-Euler method is used to describe the fluid-particle interaction, and the mesh deformation and reconstruction caused by particle motion are realized by the dynamic mesh technology. When the deformation exceeds 20% of the initial mesh size, the mesh reconstruction is triggered. The parallel repartitioning method is used to achieve load balancing, and the computational efficiency is improved by 30%. The probability distribution characteristics of scouring and silting changes are analyzed by the cloud model method, the fuzzy boundaries of the silting area are extracted, and the dredging scheme is dynamically optimized.

[0028] Step S104, collecting sand samples, and classifying the sand samples to obtain physical properties of the sand, wherein the physical properties include particle size, density and shape, and obtaining behavioral characteristics of the sand samples according to the different physical properties of the sand.

[0029] The sand samples are collected by adopting the method of equidistant grid distribution, and the grid spacing is determined according to the average particle size and distribution uniformity of the sand samples. At the same time, the sampling points are encrypted in the area where the scouring and silting changes are significant to obtain a statistically representative sample data set; the collected sand samples are pretreated, and the pretreatment includes removing impurities and organic matter, obtaining sand particle samples of different particle sizes by vibration screening, and numbering and marking the samples of each particle size; the particle size distribution of the sand is measured by a laser diffraction particle size analyzer, and the particle size frequency distribution curve and cumulative distribution curve are plotted to quantitatively describe the particle size composition characteristics of the sand; the density of the sand is determined by the pycnometer method, deionized water is added to the scale, the total mass in the bottle is weighed, and the density of the sand is calculated according to the total mass difference in the bottle and the density of water; the shape of the sand is quantitatively described, the two-dimensional contour of the sand is extracted by image processing technology, and the shape is characterized by calculating Dimensionless parameters, the dimensionless parameters include aspect ratio, roundness and sphericity, and the shapes of sand particles are classified based on the cluster analysis method, and a mapping relationship between shape parameters and categories is established; in combination with the particle size, density and shape of the sand particles, a decision tree is used to comprehensively classify the sand samples, and classification rules are constructed by a recursive partitioning method. The depth of the tree and the number of leaf nodes are optimized by a cross-validation method, and the classification accuracy is greater than a preset accuracy value, so that different types of sand particle samples are obtained; according to the different types of sand particles, the instantaneous motion of sand particles in water flow is measured by a planar laser induced fluorescence method to obtain the preset motion mode of the sand particles, analyze the critical flow velocity conditions for the start-up and sedimentation of the sand particles, and construct a quantitative relationship between the particle size, density and shape parameters and the sand particle motion parameters to display the behavioral characteristics of the sand particles under the action of water flow, wherein the preset motion mode includes suspension, saltation and rolling.

[0030] Specifically, 5-10 times the average particle size is generally taken, and sampling points are appropriately encrypted in areas with significant changes in scouring and silting to obtain a statistically representative sample data set. While sampling, a digital camera is used to image the surface morphology of the sand particles to obtain high-resolution sand image data to provide data support for subsequent shape analysis. The particle size range is from 0.075mm to 2mm, and the grading standard meets the sedimentological classification indicators. Samples of each particle size are numbered and marked to prepare for subsequent tests. The instrument measurement range is 0.02-2000μm. Repeat the measurement 3 times to take the average value, calculate the median particle size by volume percentage, and draw the particle size frequency distribution curve and cumulative distribution curve to quantitatively describe the particle size composition characteristics of the sand particles. The density of the sand particles is determined by the specific gravity bottle method. 50g of dried sand sample is weighed, deionized water is added to the scale, the total mass in the bottle is weighed, and the density of the sand particles is calculated based on the mass difference and the density of water. Repeat the measurement 3 times to take the average value, and the accuracy reaches 0.1g / cm 3. The shape of sand particles is quantitatively described. The two-dimensional contour of sand particles is extracted by an image processing algorithm based on edge detection and threshold segmentation. The dimensionless parameters characterizing the shape, including aspect ratio, roundness, sphericity, etc., are calculated. The shape of sand particles is classified based on the K-means clustering analysis method, and the mapping relationship between shape parameters and categories is established. The preset accuracy value is 90; the Latinhypercube sampling method is used to select representative particles from each type of sand sample, and the water flow movement characteristics of sand particles under different particle size, density and shape combinations are studied by orthogonal experimental design. By changing the hydraulic conditions such as water depth, flow velocity and turbulence intensity of the water tank, the quantitative relationship between the critical Shields number of sand particle start-up and sedimentation and the particle size, density and shape parameters is analyzed. The support vector machine regression method is used to establish a mapping relationship model from sand characteristic parameters to movement parameters such as start-up probability and average velocity, and the mapping relationship model is trained and verified using experimental data. The average relative error is controlled within 15%. By establishing a quantitative relationship between sand characteristics and movement behavior, the migration law of different types of sand under the action of water flow is revealed, providing a scientific basis for sediment movement simulation. 100 sampling points are arranged in the study area, with a grid spacing of 100m. The spacing is reduced to 50m in areas with drastic changes in underwater terrain. A grab bucket mud sampler is used to collect about 1kg of surface sand samples. At the same time, a Canon EOS90D camera is used to take 5x optical zoom shots of the sand surface morphology to obtain digital images with a resolution of no less than 4000×3000. After the sand samples are dried at 60℃ for 48h, they are separated by particle size using a RetschAS200 vibrating sieving instrument. The sieve specifications include 2mm, 1mm, 0.5mm, 0.25mm, 0.125mm and 0.075mm, with an amplitude of 2mm and a duration of 15min. The particle size was measured by Malvern Mastersizer 2000 laser diffraction particle size analyzer. The sample was dispersed in deionized water and tested on the machine after ultrasonic oscillation for 30 seconds. The instrument repeated the measurement 5 times and took the average result to obtain the particle size distribution curve and characteristic parameters d16, d50, d84, etc. The density was measured by a 25mL specific gravity bottle. 50.00g of sample was weighed and distilled water was added to the scale. The water temperature was controlled at 20±0.5℃, accurate to 0.01g, and weighed with a 1 / 10,000 balance. The measurement was repeated 3 times and the average value was calculated. After Gaussian filtering and denoising, the sand image was edge detected by Canny operator, and then the binary image was obtained by Otsu threshold segmentation. Freeman chain encoding was performed on the contour pixels, and shape parameters such as perimeter and area were calculated. K-means clustering was performed based on Euclidean distance metric, and the optimal number of categories was determined by the elbow rule. The clustering results were visualized in three dimensions.The CART decision tree algorithm is used to classify the sand particles based on their particle size, density, shape and other multi-dimensional parameters. The optimal partitioning attribute is selected through the Gini index, a recursive binary tree is generated, and the tree structure is optimized using the cost complexity pruning strategy to finally obtain the classification rules. 200 particles are randomly selected for each type of sand particles, and a two-dimensional orthogonal table L is used. 16 (4 5 )A test scheme with five factors and four levels was designed. The factors included particle size, density, aspect ratio, roundness and sphericity. A water tank with a length of 20 m, a width of 0.8 m and a height of 1.5 m was built to conduct sand particle motion experiments under different flow rates (0.2-1.5 m / s), water depths (0.2-1 m) and turbulence degrees (5%-20%). The instantaneous flow field was acquired by using a DANTEC Doppler particle image velocimeter. At the same time, fluorescent tracer particles were injected and pulsed laser illumination was used to record the particle motion using a Lane RP1600 digital high-speed camera with a shooting frequency of 1000 frames / s, a resolution of 1600×1200 pixels and an exposure time of 0.5 ms. After the sequence images were processed by histogram equalization, Gaussian smoothing and correlation filtering, the motion trajectory of the center of mass of the fluorescent particles was extracted, and the parameters such as the instantaneous velocity, acceleration and Shields number of the particles were calculated. A physical model of turbulent pulsation and particle suspension law was constructed. Based on the experimental data, the grid search and K-fold cross-validation methods were used to optimize the SVM model hyperparameters, and a nonlinear mapping relationship between sand characteristic parameters and start-up probability was established. The prediction accuracy of the physical model of turbulent pulsation and particle suspension law reached 85%. The physical model of turbulent pulsation and particle suspension law was used to estimate the sand transport in the study area.

[0031] Step S105, obtaining the accumulation and dispersion rules of sand particles according to the movement behaviors of different types of sand particles under the condition of changing flow velocity.

[0032] For different types of sand, the initial distribution conditions and bottom boundary conditions of the sand in the numerical simulation are determined according to the critical starting shear stress and sedimentation velocity obtained by the sand in the flume test; the format of the numerical values ​​is processed by discretizing the convection terms in an active upwind manner to simulate a series of preset flow velocity conditions, which include steady flow, tidal current and wave; based on the simulation results, the analysis results of the scouring and deposition change process of sediment under different dynamic conditions are obtained; the change curves of sediment concentration and scouring and deposition thickness over time are extracted from the analysis results, and the main modes of sand movement are identified through spectrum analysis to show the accumulation and dispersion laws of sand under the action of water flow.

[0033] Specifically, according to flume tests of different types of sand particles under different water flow conditions, the motion parameters of the sand particles are obtained, wherein the motion parameters include kinetic probability and average velocity; the motion parameters are input into a three-dimensional water-sediment coupling numerical model, and a non-steady flow calculation is used to simulate the sediment transport process under the action of continuous flow velocity, so as to obtain the finite volume method solution result of the Reynolds-averaged Navier-Stokes equation; according to the solution result, a sediment concentration convection-diffusion model considering the influence of turbulent pulsation is obtained; the velocity distribution in the near-wall area is processed by introducing a wall function, and the bidirectional coupling between the flow field and the sediment is realized by using a fractional step method, wherein when the wall function is introduced to process the velocity distribution in the near-wall area and the fractional step method is used to realize the bidirectional coupling between the flow field and the sediment, an unstructured grid with variable grid density is used to discretize the calculation domain, and the grid is encrypted near the bottom and at the water-sand interface; the model is based on the Reynolds-averaged Navier-Stokes equation and the sediment convection-diffusion model, and the VOF method is used to capture the water-sand interface, and the influence of turbulent pulsation on sediment movement is considered by the k-ε turbulence model. The model is solved by finite volume method and pressure-velocity coupling is realized by SIMPLE algorithm. In the model calculation, unstructured triangular mesh is used for the study area. According to the complexity of the terrain and the characteristics of water depth change, the minimum mesh size is set to 1m and the maximum mesh deformation angle is less than 150° to ensure that the mesh quality meets the requirements of numerical solution. The mesh is appropriately encrypted near the bottom and the water-sand interface, and the total number of meshes is controlled at about 3 million, which ensures accuracy while taking into account the calculation efficiency. For different types of sand particles, the bottom boundary conditions are determined according to the critical starting shear stress obtained in the flume test. The critical starting shear stress reflects the critical state of sand particles starting to move under given hydrodynamic conditions. It can be estimated by the relationship between the Shields curve and the dimensionless particle size, and the critical starting shear stress is converted into the thickness of the bottom erodible layer as the bottom boundary condition for numerical simulation. In addition, the source and sink terms in the sediment convection-diffusion model are determined according to the settling velocity formula of sand particles. In the numerical solution, the convection term is discretized by upwind format to ensure the stability and monotonicity of the numerical format and avoid the occurrence of negative concentration. A series of typical flow conditions, including steady flow, tidal current and waves, were simulated to analyze the scouring and silting process of sediment under different dynamic conditions. The simulated sediment concentration and scouring and silting thickness curves with time were subjected to fast Fourier transformation to obtain the amplitude spectrum and power spectrum of each frequency component, identify the dominant frequency with higher energy density, analyze its relationship with dynamic processes such as tides, wind and waves, and reveal the periodic accumulation and dispersion of sand particles under the action of water flow.For different types of sand, the equilibrium sediment transport rate under different flow conditions is calculated, that is, the volume of sediment passing through a unit section per unit time, and the relationship curve between sediment transport rate and flow velocity is plotted. The universal sediment transport rate formula considering parameters such as sand particle size, density and shape is further fitted, and the least square method is used to optimize the parameters to minimize the root mean square error between the fitting curve and the numerical simulation results. Combined with hydrodynamic conditions such as bottom shear stress and turbulence intensity, the discriminant analysis method is used to establish the discriminant function between sediment type and water-sediment conditions, and the common sediment types such as flaky bedding, cross-bedding, and massive bedding are quantitatively identified. On this basis, the established numerical model is used to simulate the sediment movement in the actual sea area under long-term tidal, wind and wave dynamic conditions. For tidal currents, the tidal level and flow velocity boundary conditions are set according to the results of tidal harmonic analysis; for wind and waves, the third-generation spectral wave model is used to simulate the irregular waves under the actual wind field, and the radiation stress caused by the waves is input into the water-sediment coupling model as the boundary condition. Through long-term continuous simulation, the spatial accumulation distribution characteristics of sand on typical geomorphic units such as submarine sand ridges, sand bars, and tidal deltas are obtained, and the influence of environmental factors such as flow velocity, water depth, bed stress, and terrain undulation on sand deposition is analyzed to reveal the mechanism of selective sedimentation of sand. The measured topography and sediment grain size data are input into the established numerical model, and the average relative error and correlation coefficient between the simulation results and the measured data are calibrated to meet the preset accuracy requirements. The optimized model is used to predict the spatial distribution of sediment grain size in the study area in the future, and the dynamic change trend on different time scales is obtained, providing a scientific basis for coastal engineering planning and dredging scheme design. When constructing a three-dimensional water-sand coupling numerical model, an unstructured triangular grid is used to divide the study area. The total number of grid units is about 2.8 million, the minimum grid size is 0.8m, and the maximum grid deformation angle is 135°. A hexahedral boundary layer grid is used to process the near-wall area. The first layer grid height is 0.001m, and the grid is encrypted to 0.05m at the water-sand interface. The grid encryption test is used to ensure that the y+ value is less than 1. For fine sand with a median particle size d50 of 0.25mm, the critical starting shear stress is determined to be 0.15Pa according to the Shields curve. Combined with the bottom bed thickness observed in the field, the bottom 1cm is set as the active layer to exchange mass with the inside of the calculation domain. The van Rijn formula is used to calculate the sedimentation velocity of sand particles. For fine sand with a d50 of 0.25mm, the sedimentation velocity is about 2.5cm / s. In the numerical simulation, the second-order central difference format is used to discretize the diffusion term, the Sweby limiter is used to process the convection term, the time step is controlled by the CFL criterion, and the Courant number is 0.8.Aiming at the tide-controlled delta of the Yellow River estuary, the water and sediment movement process under three typical tidal ranges of large, medium and small was simulated, and the velocity and sediment concentration time series of 10 representative stations in the sea area were extracted. The spectrum analysis was performed using the FFT method, and it was found that the semi-diurnal tide M2 ​​component and the quarter-diurnal tide M4 component had the strongest energy, indicating that the tide is the main driving factor controlling the periodic movement of sediment. By fitting the power function relationship between the tidal average sediment transport rate and the tidal average flow velocity at each station, a generalized sediment transport rate formula considering the sand particle size d50 and the bed surface undulation Δ was established: qs=38.2d50^1.5Δ^0.75U^3.6, where qs is the sediment transport rate (kg / m / s), U is the cross-sectional average flow velocity (m / s), and the empirical coefficient is obtained by least squares fitting, with a root mean square error of 1.2kg / m / s. Combining the simulation and spectrum analysis results, an empirical criterion for distinguishing tide-controlled coastal sediment types is proposed: when U / ωD<500, trough cross-bedding develops; when 500<U / ωD<2000, flaky bedding develops; when U / ωD>2000, massive bedding develops; where ω is the M2 frequency and D is the water depth. The underwater topographic evolution data of the Yellow River estuary from 1986 to 2005 were input into the established numerical model, and the model parameters were calibrated by trial and error method: the dry density was taken as 1450kg / m. 3 , the turbulent Schmidt number is 0.8, the suspended sediment concentration Rouse index is 0.2, the average relative error of the model verification is 12%, and the correlation coefficient is 0.87. The calibrated model is used to predict the distribution of bottom erosion and siltation in 2025. The results show that the area from Cuijiawan to Gaocun will be silted up by about 1.2m, and the erosion near Dongying Port will be about 0.6m.

[0034] Step S106, adjusting the sensor deployment density according to the motion pattern data and the accumulation and dispersion law data of the sand particles to obtain a sensor deployment density plan, wherein the sensor deployment density includes the layout, spacing and number of sensors.

[0035] The flow field data, the accumulation and dispersion law data of the sand particles and the movement mode data are statistically analyzed; the probability distribution function of flow velocity, water depth and sediment concentration is obtained; the main change modes of each parameter are extracted by combining empirical orthogonal decomposition; a data-driven model after dimensionality reduction is established to realize the mapping from flow field characteristics to sand particle movement; for different flow velocity conditions and the sand particle movement mode, a method combining uniform sampling and stratified sampling is adopted; a certain area is studied and a large number of sampling points are generated in the study area; the flow field and sand particle movement parameters of each sampling point are obtained by interpolation to form a data set for sensor layout optimization; the spatial correlation between sampling points is analyzed by using the Kriging interpolation method; the number of sensors can be reduced for areas with strong spatial correlation, while the density of sensors needs to be increased for areas with drastic spatial changes; the spacing of the sensor layout is determined by constructing a variogram and a semi-variogram; the monitored flow velocity and sediment concentration are graded according to the importance of the data and the difficulty of acquisition; for those whose importance coefficient is greater than the preset importance threshold and whose difficulty is greater than the preset degree threshold data, establish an association model with other parameters through machine learning methods; use observation data to train the association model to guide sensor layout; optimize the sensor deployment density plan under the premise of meeting the monitoring objectives; obtain the optimization result with the weighted sum of monitoring accuracy and cost as the objective function; find the optimal deployment plan through simulated annealing and genetic heuristic search methods to control the number and deployment location of sensors; for the time resolution and spatial resolution of the sensor deployment density plan, obtain the influence of the time resolution and spatial resolution on the data interpolation and inversion accuracy; for the time resolution, the sensor sampling frequency is required to comply with the Nyquist sampling theorem; for the spatial resolution, the Taylor scale is used to estimate the spatial variation scale of flow velocity and sediment concentration; after determining the sensor deployment density plan, an adaptive dynamic adjustment strategy is adopted to adjust and update the deployment plan online according to the real-time feedback of monitoring data; for areas where the deviation between the observed value and the prediction of the association model is greater than the preset threshold, the sensor deployment density is increased to form a closed loop of monitoring-analysis-optimization.

[0036] Specifically, the flow field and sediment movement data obtained from the previous numerical simulation and experimental analysis are statistically analyzed to obtain the probability distribution function of parameters such as flow velocity, water depth, and sediment concentration. The main change modes of each parameter are extracted by combining empirical orthogonal decomposition, and a data-driven model after dimensionality reduction is established to achieve rapid mapping from flow field characteristics to sand movement. The model can be constructed using common methods such as Gaussian process regression and support vector machine, and the model hyperparameters are optimized by methods such as maximum likelihood estimation, and the generalization performance of the model is evaluated by methods such as cross-validation. After determining the best data-driven model, the model is used to perform grid sampling in the study area to obtain the estimated values ​​of the flow field and sand movement parameters at each sampling point, forming a basic data set for optimizing sensor layout. The spatial correlation between sampling points is analyzed by using methods such as Kriging interpolation. By constructing a theoretical semivariogram and fitting the experimental semivariogram, the range and sill value reflecting the spatial correlation are obtained, and the optimal spacing of the sensor layout is determined. The mathematical model of Kriging interpolation can be expressed as: Z(x) = μ(x) + ε(x), where Z(x) is the observed value or predicted value at position x, μ(x) is the regionalized variable, and ε(x) is the random noise. The semivariogram model parameters are fitted by the least squares method, and the interpolation accuracy is evaluated by cross-validation. According to the importance of data and the difficulty of obtaining it, the hierarchical analysis method is used to sort the importance of monitoring indicators. For indicators with high importance but difficult to measure directly, a series of easy-to-measure parameters with the strongest correlation are selected through feature selection and data dimensionality reduction methods. The association model is established using machine learning methods such as support vector regression and random forest, and the generalization ability of the model is evaluated by cross-validation, thereby guiding the layout of sensors. Under the premise of meeting the monitoring objectives, the weighted sum of the Kriging variance and the mean square error of the association model is minimized as the optimization goal. At the same time, considering the constraints such as the number of sensors and communication distance, the simulated annealing algorithm is used to optimize the sensor layout. The simulated annealing algorithm controls the acceptance probability of the solution by introducing a temperature parameter, and approaches the global optimal solution in a gradual cooling manner. By designing a reasonable cooling schedule and perturbation function, the optimization efficiency and the quality of the solution are balanced. The Shannon sampling theorem is used to evaluate the impact of the spatiotemporal resolution of the layout scheme on the data recovery accuracy. For the time dimension, the sampling frequency is required to be at least twice the highest frequency of the signal; for the spatial dimension, the Nyquist sampling theorem is used to estimate the spatial correlation scale, and the sensor spacing must be less than half of the correlation scale. After determining the optimal sensor layout scheme, an adaptive dynamic adjustment module is developed to analyze the deviation between the monitoring data and the predicted value of the data-driven model in real time, calculate the mean and variance of the deviation and other statistics, and trigger the layout scheme adjustment when the deviation exceeds the preset threshold. By solving the optimization problem between the deviation statistics and the sensor density, the adjustment amount of the layout density is obtained, and the sensor layout is updated according to the adjusted layout density, so as to continuously improve the adaptive ability and accuracy of the monitoring system.For the sediment movement monitoring data of a certain river section, the principal component analysis method was used to extract the first three main change modes of flow velocity, water depth and sediment concentration, and the cumulative variance contribution rate reached 85%. Then, the radial basis kernel function support vector machine was used to establish the nonlinear relationship between flow field characteristics and sediment transport rate. The average relative error of 5 cross-validations was 10%. A 100×100m square grid was laid out in the plane area, and the sediment transport rate of each grid node was predicted using the trained support vector machine model. Then, the semivariogram was calculated using Kriging interpolation, and the fitting theoretical model was γ(h)=0.5+0.3×(1-exp(-h / 50)), the corresponding range was 50m, and the base value was 0.8, where γ(h) is the value of the semivariogram function, h is the lag distance between the two points, and exp(-h / 50) means that as the distance h increases, the value of the exponential function will gradually approach 0, which indicates that as the distance increases, the correlation between spatial points will gradually decrease. According to the range, the sensor layout spacing is determined to be 25m. Six monitoring indicators, such as sediment concentration and sediment content, were compared in pairs, and a judgment matrix was constructed and weights were calculated. The results showed that the importance weight of sediment concentration was as high as 0.35. However, since it is difficult to measure directly, Pearson correlation analysis was used to screen out the velocity and sediment content with the highest correlation coefficient as alternative indicators, and the random forest algorithm was used to establish the quantitative relationship between the three, with an average absolute percentage error of 8%. The objective function was the minimum weighted sum of the Kriging variance and the random forest generalization error. At the same time, the number of sensors was limited to no more than 50 and the communication distance was no more than 500m. The simulated annealing algorithm was used for layout optimization. The initial temperature was 1000, and the temperature dropped by 10% each time. After 100 iterations, the optimal layout scheme was obtained. According to Shannon's sampling theorem, the sampling frequency of the sediment concentration sensor was set to twice the highest frequency of 5Hz of the process under study, that is, 10Hz; according to the measured data, the spatial correlation scale of sediment concentration was estimated to be about 100m, so the sensor spacing was 50m. An adaptive adjustment module is developed to calculate the mean and standard deviation of the deviation between the monitoring value and the model prediction value in real time. When the mean deviation exceeds 2 standard deviations, the design scheme is adjusted. By solving the linear programming problem of the mean deviation and the sensor density, the density adjustment step is obtained, and then the sensor coordinates are updated. The simulation experiment verifies that the adaptively adjusted monitoring scheme improves the estimation accuracy of sediment transport flux by 15% and reduces the number of sensor nodes.

[0037] Based on the sand sedimentation data, the location and range of the sedimentation area are identified through GIS analysis, and the sedimentation trend is predicted using spatial analysis methods. If the predicted sedimentation trend is a high sedimentation area, the number of sensor arrangements will be increased; if the predicted sedimentation trend is a low sedimentation area, the number of sparse sensor arrangements will be reduced.

[0038] The sand sedimentation data is sorted into a standard format, which includes sampling point coordinates, sampling time, sedimentation velocity, and sedimentation amount, and imported into a GIS platform for spatial processing; a raster data layer of sedimentation velocity and sedimentation amount is generated by Kriging interpolation, and the resolution is calculated according to the density of the sampling points to perform threshold segmentation on the sedimentation velocity and sedimentation amount layers; according to sedimentation dynamics theory and empirical values, an area where the sedimentation velocity is greater than a preset velocity value and the sand sedimentation amount is greater than a preset sedimentation amount value is defined as a high sand sedimentation area, and the boundary vector of the high sand sedimentation area is extracted and area statistics and spatial distribution analysis are performed to obtain the current range of the high sand sedimentation area; the spatial autocorrelation of the high sand sedimentation area is analyzed by using the spatial statistics tool of GIS, and the Moran index and Getis-OrdG index are calculated to determine whether the high sand sedimentation area has significant spatial agglomeration characteristics. If there are hot spots or cold spots, it indicates that the sedimentation process is affected by local factors and needs to be monitored; a sedimentation monitoring sensor grid is deployed in the high sand sedimentation area, and according to The grid density is calculated based on the spatial variability of the deposition rate; sentinel sensors are deployed at the edge of the high sand deposition area to monitor the dynamic changes in the sand deposition range; time series analysis methods are used to perform trend extrapolation and periodic analysis on historical sedimentation data to identify the long-term evolution law and periodic fluctuation characteristics of the sand deposition process, and the ARIMA model and spectral analysis method are used to quantitatively predict the sand deposition trend in the future, and the time scale can be determined according to decision-making needs; based on the sand deposition trend prediction results and the spatial distribution map of the high sand deposition area in GIS, an overlay analysis is performed to identify the newly added high sand deposition areas and the disappeared original high sand deposition areas in the prediction period, adjust the sensor grid layout, conduct intensified monitoring of the newly added areas, and reduce the monitoring points of the disappeared areas; a sand deposition monitoring data management platform based on WebGIS is developed to realize the real-time reception, storage, analysis and visualization of sand deposition monitoring data, judge the sand deposition risk level according to the set sand deposition rate and range threshold, generate early warning information and push it to relevant departments.

[0039] Specifically, the sedimentation velocity and sedimentation data can be derived from field sampling monitoring or numerical simulation, with a time resolution of less than 1 hour and a spatial resolution of less than 1 square kilometer. The resolution is generally 20-50m. Kriging interpolation models the spatial autocorrelation of the sampling points to obtain the optimal linear unbiased estimate, in which the semivariogram model and interpolation parameters are optimized by cross-validation and other methods. Threshold segmentation is performed on the sedimentation velocity and sedimentation layers, and the threshold range of the high sedimentation area is adaptively determined with reference to the coastal dynamic sedimentation theory and historical monitoring data. The preset sedimentation velocity is generally 0.5-2cm / s, and the preset sedimentation is generally 2-10cm. The boundary vector of the high sedimentation area is extracted and the area statistics and spatial distribution analysis are performed to obtain the current range of the high sedimentation area. Deploy a sediment monitoring sensor grid in a high sedimentation area. The grid density is determined based on the spatial variability of sedimentation rate, terrain complexity, and diversity of sediment types. At the same time, consider the constraints such as sensor measurement accuracy and deployment cost. In areas with strong sedimentation rate variability, complex terrain, and diverse sediment types, the network density is appropriately increased. The network density is generally 10-20 per square kilometer. The prediction time scale is determined according to decision-making needs, generally 1-3 years. Identify the newly added high sedimentation areas and weakened original high sedimentation areas during the prediction period, dynamically adjust the sensor grid layout according to indicators such as prediction error threshold and monitoring cost-effectiveness ratio, conduct targeted and encrypted monitoring of the newly added high sedimentation areas, and reduce the number of monitoring points in the weakened areas as appropriate. The adjustment frequency is determined according to the dynamic change characteristics of the sedimentation process, generally half a year or one year. Develop a sedimentation monitoring data management platform based on WebGIS to realize the real-time reception, storage, analysis and visualization of sedimentation monitoring data. With reference to international coastal zone governance guidelines and engineering design specifications, the sedimentation risk level classification standards are formulated from multiple dimensions such as sedimentation rate, sedimentation thickness, and coverage. The sedimentation risk index of each monitoring area is calculated in real time. When the risk index exceeds the warning threshold, the warning information is automatically generated and pushed to the relevant departments. The advance warning time is determined according to the risk level, generally 1-3 months, and the spatial range is determined according to the risk impact domain, generally 5-10 square kilometers. Through the whole process of sedimentation monitoring, early warning, and decision support, a scientific basis is provided for the comprehensive management and disaster prevention and mitigation of the coastal zone. When conducting sedimentation monitoring in a bay, the ADCP and fixed-point OBS were used to obtain the velocity profile and suspended sediment concentration time series data on 100 monitoring sections, with a time interval of 10 minutes and a spatial interval of 500m. The sedimentation velocity was estimated by the improved Rouse formula, and the sedimentation amount was estimated by combining the velocity profile data. The geostatistical analysis module of GIS was used to perform Kriging interpolation on the sedimentation velocity and amount, the Gaussian model was used to fit the semivariogram, and the maximum likelihood method was used to estimate the nugget parameters. The optimized nugget variance was 0.15 and the sill value was 1.2.Through cross-validation, the optimal search neighborhood radius was determined to be 500m, and the root mean square error of the interpolation result was 0.2cm / s. Overlay analysis showed that the total area of ​​high sedimentation areas with sedimentation velocity greater than 1.2cm / s and sedimentation greater than 5cm was 8 square kilometers, concentrated in the coastal waters of the northern part of the bay. Using the stratified sampling method, 25 monitoring points were set up in the high sedimentation area with a monitoring frequency of once per hour; 10 monitoring points were set up outside the high sedimentation area with a monitoring frequency of once per day. Two time series prediction models, ARIMA and LSTM, were selected for training and testing using 5 years of historical monitoring data. The results showed that the peak relative error of the ARIMA model was 15%, and the root mean square error of the LSTM model was 0.8cm. Therefore, the LSTM model was selected to predict the sedimentation trend in the next year. The prediction results show that the range of the high sedimentation area will expand by 1.5 square kilometers to the north and decrease by 0.8 square kilometers to the southeast. Based on this, the monitoring network was adjusted, 5 monitoring points were set up in the newly added high sedimentation area, and 2 monitoring points were cancelled in the weakened area. The density of the adjusted monitoring network was 3 / km2. Based on the "Coastal Engineering Design Code" of the State Oceanic Administration, a 4-level sedimentation risk level standard was formulated. When the sedimentation rate is greater than 1.5cm / s, the sedimentation thickness is greater than 10cm, and the sedimentation range is greater than 5km2, it is defined as an extremely high risk level. A sedimentation warning release system based on WebGIS was developed, and the sedimentation risk index was calculated with a time step of 6 hours. When the risk index for two consecutive time steps reaches level 3 (high risk) or above, warning information is issued to the maritime department and port management department, indicating the possible sedimentation disasters and their impact range in the next week, and pushing the recommended dredging and desilting plan.

[0040] Based on the data of the migration of the sand particles, the paths and stop points of the sand particles in the water flow are calculated, and sensors are arranged according to the paths and the stop points. If the paths have intersections and turning areas, the number of sensors arranged will be increased.

[0041] Collect sand migration monitoring data under high flow rate conditions, the sand migration monitoring data includes flow rate, flow direction, sand concentration and particle size, use ADCP and OBS online monitoring equipment to obtain data, and transmit it back in real time through wireless transmission technology; pre-process the sand migration monitoring data, use wavelet analysis method to remove high-frequency noise and outliers, and select the optimal wavelet basis and decomposition scale through cross-validation; use correlation analysis method and cross-spectral analysis method to study the quantitative relationship between the flow rate and flow direction and the sand concentration and particle size, establish a multivariate regression model, and use ensemble learning method to optimize the generalization ability and prediction accuracy of the regression model; simulate the movement trajectory of the tracer sand according to the sand migration data, and then obtain the migration path of the sand; mine and analyze the big data of sand movement trajectory obtained by numerical simulation, use density clustering algorithm to extract the skeleton structure of the trajectory, and use spectral clustering algorithm to divide the branch structure of the trajectory, and then determine the intersection and turning point of the sand migration path; use trajectory complexity and coverage density to measure the importance of each key feature point, and combine measured data and mechanism analysis to finally determine the key nodes for priority deployment of sensors. , the deployment density is proportional to the importance of the node; in the non-critical area of ​​sand migration, the maximum entropy principle is used to determine the spacing of the sensor deployment, and the sensor spacing deployment plan is obtained; the functional model of the deployment spacing, monitoring error and deployment cost is established through variational derivation, and then the simulated annealing algorithm is used for global optimization to obtain the sensor deployment plan that takes into account both the key nodes and the ordinary areas; the sensor deployment plan that takes into account both the key nodes and the ordinary areas and the sensor spacing deployment plan are combined to form a composite unequal distance sensor network topology, and a multi-hop self-organizing routing coordination is used. The proposal is to realize the communication between sensor nodes and aggregation nodes; integrate terrain data and hydrological and meteorological data to establish a comprehensive integrated model of sand migration, and realize all-round simulation analysis from process mechanism to parameter identification, from static distribution to dynamic prediction; use Bayesian network method to conduct probabilistic reasoning and uncertainty analysis on the law of sand migration, combine subjective and objective weighting method to determine the key risk factors of sand migration, and use factor combination analysis and orthogonal experimental design to construct a multi-scenario and multi-condition sand migration early warning solution library, which provides decision-making basis for sensor layout optimization and early warning threshold determination.

[0042] Specifically, the movement trajectory of the tracer sand is simulated according to the data of the sand migration, and then the path of the sand migration is obtained. Specifically, based on the flow field monitoring data, the movement trajectory of the sand in the water flow is calculated by numerical simulation method, and the velocity vector and sand concentration gradient at different time and space positions are determined by solving the NS equation and the convection-diffusion equation. Then, the Lagrangian particle tracking method or the Euler-Lagrangian coupling method is used to simulate the movement trajectory of a large number of tracer sands, and then the path of the sand migration is obtained; the data sampling frequency is not less than 1Hz, and the spatial resolution is not less than 1m; the denoised data can reduce the random error by more than 20%. In addition, data synchronization, interpolation, fusion and other operations should be performed to improve the consistency and continuity of the data in time and space. On this basis, correlation analysis, cross-spectral analysis and other methods are used to study the quantitative relationship between hydrodynamic factors such as flow velocity and flow direction and sediment transport parameters such as sand concentration and particle size. At the same time, the turbulent characteristics of the water flow are considered, the k-ε turbulence model is introduced, and a water flow-sand coupling model including turbulence effects is established. Integrated learning algorithms such as AdaBoost and RandomForest are used to improve the generalization ability and prediction accuracy of the model, and the average error is controlled within 15%. In the numerical simulation, the finite volume method is used for discretization and parallel calculation, which can achieve a fully coupled solution of 1 billion grids. The key nodes where sensors are preferentially deployed are determined, and the deployment density is proportional to the importance of the nodes, generally 2-5 per 100 meters. The optimal sensor deployment plan that takes into account both important nodes and ordinary areas is obtained, with a deployment density of 0.5-2 per 100 meters. On this basis, considering the characteristics of sensor node autonomy and link dynamics, multi-hop self-organizing routing protocols such as LEACH and PEGASIS are adopted to realize reliable communication between sensor nodes and aggregation nodes, and consider fault diagnosis, network maintenance, data security and other mechanisms to improve the robustness and availability of the whole network. Machine learning methods such as Bayesian network and Markov chain are used to conduct probabilistic reasoning and uncertainty analysis on the sand migration law, Monte Carlo method is used for random sampling, interval analysis method is used for interval estimation, and the key risk factors of sand migration are determined by combining the hierarchical analysis method. Factor combination analysis and orthogonal experimental design are used to construct a multi-scenario and multi-condition sand migration early warning model, formulate graded warning thresholds based on indicators such as intensity, range, and duration, and form a sand migration monitoring and early warning business process for disaster prevention and mitigation, providing strong support for coastal erosion early warning, waterway regulation, and ecological protection. In a river estuary runoff control project, 500 sand migration profile data were obtained using ship-borne ADCP and shore-based OBS, with a time interval of 1s, a vertical resolution of 0.1m, and a horizontal resolution of 1m. The db4 wavelet basis function was used to decompose the data into 5 levels. The optimal decomposition scale was determined to be 3 through 3-fold cross validation. The mean square error of the velocity data after denoising was reduced by 25%.The Pearson correlation coefficient was used to evaluate the correlation between flow velocity and sand concentration, and the average correlation coefficient was 0.78. The cross power spectrum of the two was estimated by the Welch method, and a significant spectral peak was found near 0.05Hz. The k-ε turbulence model was used to simulate the flow field and was bidirectionally coupled with the sand concentration field. The water flow-sand relationship model was trained using the random forest algorithm, and the average absolute error was 12%. The Euler-Lagrange coupling method was used to simulate the motion trajectory of 5000 tracer particles. The particle diameter obeyed the log-normal distribution with an average value of 0.2mm. The release position was 0.1m from the riverbed surface, the time step was 0.01s, and the boundary condition adopted the rebound model. Through DBSCAN cluster analysis, three main migration paths and five key nodes were obtained, and the node density threshold was 10 / m2. The simulated annealing algorithm was used to optimize the sensor layout. The objective function was the weighted sum of monitoring error and cost. After 500 iterations, the optimal layout of 12 sensors was obtained, with an average spacing of 50m and a density of 5m near key nodes. The sensor network is built using the LEACH protocol. 500 nodes are randomly distributed. 5% of the nodes are selected as cluster heads in each round. The TDMA method is used to send data in turn. The transmission distance is 100m and the bit error rate is 0.01%. Finally, a sand migration probability model based on the Bayesian network is established. The parameters are estimated by the MCMC method. Combined with five factors such as sediment content, water level, and flow, a sand migration risk warning scheme under eight working conditions is constructed. When the sediment content exceeds 10kg / m3, the sand migration risk warning scheme is used. 3 When the flow rate exceeds 2m / s and lasts for more than 1 hour, a red warning is issued and an emergency dredging plan is initiated. By comparing with the measured data, the missed alarm rate and false alarm rate of the warning plan are 5% and 10% respectively, and the average warning lead time is 2 hours.

[0043] The sand suspension data is simulated using a CFD simulation method to obtain the suspension height, velocity and concentration distribution of the sand in the water flow.

[0044] The influence of turbulent structure on the suspension process of the sand particles is captured by the LES large eddy simulation method. Combined with the non-uniformity and non-spherical characteristics of the sand particles, the DEM discrete element method is used to simulate the collision between the sand particles, and the multi-field coupling solution results between the water flow and the sand particles and between the sand particles are obtained; the CFD simulation method is used to simulate the data of the sand suspension process, and the suspension height, velocity and concentration distribution of the sand particles under different flow conditions are obtained; the Lagrangian tracer particle method is used to track the motion trajectories of a large number of tracer particles, and the influence of the water flow velocity and turbulence intensity on the suspension characteristics of the sand particles is analyzed; according to the influence law, the quantitative relationship between the sand suspension parameters and the flow field characteristic quantities is constructed.

[0045] Specifically, the monitoring data of suspended sand under high flow conditions are collected, including multiple parameters such as suspended sand concentration, particle size distribution, and sand morphology at different water depths. The sampling frequency and vertical resolution are reasonably set according to the monitoring equipment and target requirements. For example, the vertical resolution of ADCP can reach 0.1m and the sampling frequency can reach 10Hz; the vertical resolution of laser scanner can reach 0.01m and the sampling frequency can reach 100Hz. The acquired suspended sand data are preprocessed, and signal processing methods such as wavelet transform are used to remove high-frequency noise and extract effective signal components. The main frequency components of the sand suspension process are determined by power spectrum analysis, and the suspension characteristic patterns at different time and space scales are extracted by empirical mode decomposition. Empirical mode decomposition finds the local extreme values ​​of the data sequence, constructs upper and lower envelopes, obtains several intrinsic mode functions, and realizes adaptive decomposition of the signal. At the same time, methods such as box plots are used to detect and eliminate outliers in the data, and interpolation, regression and other methods are used to fill in missing values ​​to improve the integrity and representativeness of the data. A three-dimensional water flow-sand suspension CFD coupling model is constructed. The RANS turbulence model (such as the standard k-ε model) is combined with the wall function to describe the turbulence effect. The DEM discrete element method is used to simulate the collision between sand particles. The two-way coupling between water flow and sand particles is realized through the dynamic grid technology. The model is solved using the SIMPLE algorithm, the second-order upwind format is used to discretize the convection term, the central difference format is used to discretize the diffusion term, and the algebraic multigrid method is used to accelerate convergence. The CFD model is used to carry out numerical simulation of the sand suspension process. On the basis of the sand concentration field, the Lagrangian tracer particle method is used to track the motion trajectories of a large number of tracer particles, analyze the influence of parameters such as water flow velocity and turbulence intensity on the suspension characteristics, and construct the quantitative relationship between the suspension parameters and the flow field characteristic quantities. The key areas, measurement range and sensitivity of sensor layout are further analyzed and determined. The intelligent optimization algorithm is used to determine the sensor layout scheme, with monitoring accuracy (such as mean square error) and cost (such as the number of sensors) as the optimization objective function. The sensor measurement range, resolution, durability and other factors are introduced to construct constraint conditions. The multi-objective genetic algorithm NSGA-II is used for optimization, with a crossover probability of 0.8, a mutation probability of 0.1, a population size of 100, and an evolutionary generation of 50. The individuals on the Pareto front are the candidate optimal layout schemes. According to the velocity distribution obtained by CFD simulation, a high-density, high-sensitivity sensor array is arranged in the area with the largest water velocity. The sensor spacing is less than the turbulence-related distance to meet the sampling theorem requirements. Multi-layer sensors are arranged vertically to achieve three-dimensional monitoring of the full water depth range, and the layout is appropriately intensified near the bottom and near the surface. At the same time, taking into account the differences in water and sediment conditions in different river sections, high-reliability sensors are used for river sections with turbulent water flow, and redundant designs are configured; low-power sensors are used for river sections with siltation to reduce operation and maintenance costs, and finally a comprehensive and integrated three-dimensional intelligent monitoring network is formed.The monitoring data is assimilated into the CFD model to carry out medium- and long-term forecasts of sediment movement, and is used to guide engineering practices such as river regulation and waterway maintenance, providing a scientific decision-making basis for basin sediment management and river regime control. By continuously accumulating and updating data, optimizing sensor layout, improving monitoring accuracy and efficiency, and realizing real-time perception and dynamic early warning of sediment transport in high-velocity river areas. In the monitoring of high-velocity sediment transport in the Three Gorges Reservoir, the velocity profiles of 200 vertical lines were obtained using ADCP, with a resolution of 5 cm and an average sampling frequency of 15 Hz; the suspended sediment particle size distribution data were obtained using the LISST-200X laser particle size analyzer, with a particle size range of 2-500 μm and a sampling frequency of 1 Hz. The velocity data was denoised by wavelet, and the db5 wavelet basis function was selected to decompose 5 layers to remove high-frequency noise; the particle size spectrum data was decomposed by EMD, and 5 eigenmode functions were extracted. The energy of the first three mode functions accounted for 92%, reflecting the main particle size composition of suspended sediment. Through power spectrum analysis, it is found that the suspended sediment concentration has a significant peak value near 0.05Hz, corresponding to the eddy scale of about 20m. The wild point values ​​of suspended sediment concentration were eliminated by box plot, and the wild point ratio was 2%. The missing value interpolation was realized by local weighted regression method. Based on the OpenFOAM platform, a three-dimensional turbulent-discrete particle two-phase flow model was constructed. The LES method and Smagorinsky subgrid model were used. The elastic collision and friction between particles were considered. The computational domain was 800×100×40m, the number of grids was 5 million, the time step was 0.01s, and the total simulation time was 1h. The simulation results show that in the area with a flow velocity of more than 2m / s, there is a significant gradient of sediment concentration within 1m near the bottom layer, and the maximum value is 3 times the average value. By tracking 5000 tracer particles, it was found that the average suspension height of particles with a particle size greater than 62μm was less than 0.5m, and resuspension would occur when the local turbulence intensity was greater than 0.2m / s. The NSGA-II algorithm was used to optimize the sensor layout. The objective function was the weighted sum of the monitoring error and the number of sensors. The constraint conditions were that the vertical spacing of the sensors was greater than 0.2m and the horizontal spacing was greater than 5m. After 50 generations of evolution, 10 groups of Pareto optimal solutions were obtained. Finally, it was determined that a 5-layer sensor array would be deployed in the area with a flow velocity greater than 2m / s, with a vertical spacing of 0.5m, a density of 0.2m near the bottom layer, and a horizontal spacing of 10m, meeting the sampling requirements of 1 / 4 turbulence integral scale. The sensors used industrial-grade products with a mean time between failures of 10,000h. The monitoring data was applied to the sand peak forecast model, which improved the forecast time by 6h, guided the optimization of the reservoir scheduling plan, maximized the amount of sediment discharge, and reduced the amount of siltation by 22%.

[0046] Step S107 , formulating a sensor deployment strategy according to the simulation result of the CFD simulation method, and designing a sensor network according to the sensor deployment strategy.

[0047] The optimization target is determined according to the monitoring accuracy and cost, and the measurement range, resolution and durability constraints of the sensor are obtained at the same time. The layout strategy is searched through the multi-objective genetic method and the ant colony method to obtain the sensor layout plan; according to the speed obtained by the CFD simulation method, the density of the sensors is arranged according to the speed, and the greater the speed, the greater the density of the sensors is, and the smaller the speed, the smaller the density of the sensors is, and the sensor layout strategy is generated; the sensor network is designed according to the sensor layout strategy, and in the process of designing the sensor network, the topology structure and communication protocol of the sensor network are designed by adopting a multi-level heterogeneous network architecture.

[0048] Specifically, the simulation results under different water flow conditions and sand movement conditions are statistically analyzed to extract the distribution characteristics of key parameters such as flow velocity, turbulence intensity, sand content, and particle size distribution. The principal component analysis is used to extract the main variation direction of the parameters. The number of principal components is determined by the maximum variance criterion. Generally, the first few principal components with a cumulative contribution rate greater than 80% are taken. The K-means clustering algorithm is then used to cluster the principal component scores, and the optimal number of clusters is determined by the elbow rule to obtain the main partitions for sensor deployment requirements. For different partitions, follow the spatial distribution patterns indicated by the simulation results and formulate differentiated sensor deployment strategies. When the flow velocity is greater than 1m / s and the sand content is greater than 10kg / m 3 In high sediment movement areas, the density of sensors should be increased to 2 sensors per square meter; in turbulence development areas with turbulence intensity greater than 0.2, high-frequency response sensors should be deployed first, with a sampling frequency of no less than 50Hz; in other flat areas, the density of sensors can be reduced to 1 per square meter, and the sampling frequency can be reduced to 1Hz. Intelligent sensors with adaptive range and frequency modulation functions, such as Doppler velocity profilers and laser scattering turbidity meters, are selected to dynamically adjust the working parameters of sensors according to changes in flow field and sediment conditions to expand their scope of application. At the same time, the spatial layout of sensors is optimized, with logarithmic distribution in the vertical direction, dense layout near the bottom, and nested grid layout in the horizontal direction. The grid resolution is proportional to the velocity gradient and sediment content gradient, and is appropriately dense in areas with large gradients. When selecting sensors, focus on indicators such as measurement accuracy, stability and durability. Sensors with velocity accuracy better than 1cm / s, sediment content accuracy better than 10mg / L, and mean time between failures longer than 1 year are given priority.

[0049] After determining the sensor deployment plan, the topology and communication protocol of the sensor network are further designed. Taking into account the spatial distribution of sensor nodes, data volume, transmission distance and other factors, a multi-level heterogeneous network architecture is adopted. In the short distance, short-range wireless communication protocols such as ZigBee and BlueTooth are used to achieve reliable communication between sensor nodes and aggregation nodes. The maximum communication distance can reach 100m and the data rate can reach 250kbps. In the long distance, low-power wide area network communication technologies such as LoRa and NB-IoT are used to achieve long-distance communication between aggregation nodes and monitoring centers. The transmission distance can reach 10km and the data rate can reach 100kbps. In terms of network routing, in view of the uneven distribution of network nodes and the dynamic changes in link quality, an adaptive multi-path routing strategy is designed. By regularly detecting the link quality, the optimal transmission path is dynamically selected to improve the reliability and robustness of the network. In terms of data fusion, a multi-sensor joint estimation method is adopted, and weighted average, Kalman filtering and other algorithms are used to achieve the fusion of different sensor data. Optimize fusion to improve measurement accuracy and reliability. Develop sensor data parsing and fusion algorithms to parse and calibrate asynchronous data streams of heterogeneous sensors to improve data timeliness and consistency. For abnormal situations such as network congestion and node failure, adopt fault-tolerant mechanisms such as multi-level caching and automatic retransmission to ensure data integrity and continuity. Introduce data quality assessment methods based on threshold detection and trend analysis to achieve real-time diagnosis and alarm of measurement anomalies and improve system operation and maintenance efficiency. The fused data will be used as input for subsequent model analysis to achieve seamless connection and integrated application of data in the industrial chain.

[0050] A sand grain motion feature extraction and analysis model is constructed, and the wavelet transform is used to perform multi-scale decomposition on the sand grain concentration time series to extract high-frequency and low-frequency components reflecting the sand grain suspension, saltation and other motion modes. The selection of wavelet basis function can be determined according to the spectral characteristics of sand grain motion. For example, the db4 wavelet has a strong ability to represent sharp pulse signals, and the coif2 wavelet has a good approximation ability for smooth signals. Then, the empirical mode decomposition is used to adaptively decompose the wavelet components to obtain the intrinsic mode functions on different time scales to characterize the multi-scale characteristics of sand grain motion. The empirical mode decomposition is a data-driven adaptive signal decomposition method that finds the local extreme values ​​of the signal, constructs the upper and lower envelopes, obtains the intrinsic mode functions and residual components. On the basis of extracting the sand grain motion features, a machine learning algorithm is used to establish the association model between the sand grain motion parameters and the water flow elements to achieve the classification and prediction of the sand grain motion state. Commonly used machine learning algorithms include support vector machines, random forests, BP neural networks, etc. Among them, support vector machines are suitable for small samples and nonlinear classification problems, and random forests and neural networks are suitable for modeling complex multivariate nonlinear relationships. The model is trained and tested by cross-validation and other methods to continuously optimize the generalization performance and prediction ability of the model. The model analysis results are compared with the actual sensor data. The simulation accuracy and applicability of the model are evaluated by using indicators such as root mean square error and Nash efficiency coefficient. The influence of model structure and parameter uncertainty on the simulation results is analyzed, and it is used to guide the iterative optimization of the sensor deployment scheme. On the basis of model verification, the sensor observation data is further fed back into the model by using data assimilation methods such as ensemble Kalman filtering and particle filtering to realize real-time correction of model state variables and parameters, and improve the prediction accuracy and reliability of the model. By constructing a dynamic feedback mechanism of model-sensor-model, the monitoring system can be adaptively adjusted and continuously optimized in a complex dynamic environment, and the monitoring and early warning capabilities can be continuously improved. At the same time, the applicability and limitations of the monitoring system under different water and sand conditions are fully considered. Under harsh conditions, the robustness and reliability of the system are improved by increasing sensor redundancy and optimizing measurement algorithms. Prior knowledge such as numerical simulation and expert experience is used to make up for the lack of observation data, so as to achieve stable operation of the monitoring system under all working conditions and all time periods.

[0051] The water and sediment movement monitoring data of a 100-kilometer section of a river estuary were analyzed, and five key parameters such as velocity, sediment content, and median particle size were extracted using principal component analysis, with a cumulative contribution rate of 85%. Then, the K-means algorithm was used for clustering to obtain three main partitions, namely, high sediment content area, turbulent development area, and gentle movement area, and the layout schemes of 3, 2, and 1 sensors per square meter were adopted respectively. The WorkHorse 600kHz Doppler current profiler was selected with a range of 0.1-6m / s, and the sediment content was measured using a portable OBS-3A turbidity meter with a range of 0-5000mg / L. The network topology combined with ZigBee and LoRa was adopted. The ZigBee protocol was used for short-distance transmission, with a frequency of 2.4GHz and a transmission distance of 100m, and the LoRa protocol was used for long-distance transmission, with a frequency of 433MHz and a transmission distance of 5km. The network nodes adopted the AODV routing protocol, and each node saved a complete network topology table, and the routing table update cycle was 60s. The sensor data was automatically parsed and calibrated, and the weighted recursive least squares algorithm was used for data fusion. The uncertainty of sediment measurement after fusion was better than 5%. The db4 wavelet was used to decompose the sediment time series into 5 layers to obtain the scale coefficient and wavelet coefficient, and then the scale coefficient was decomposed by EMD to obtain 5 eigenmode functions. The nonlinear regression model of sediment content and flow velocity was established by support vector machine, using radial basis kernel function, regularization parameter C=10, the ratio of training set to test set was 7:3, and the average relative error of 5 cross validations was 8%. The model prediction results were compared with the measured data, the Nash efficiency coefficient was 0.85, and the root mean square error was 2mg / L. The ensemble Kalman filter algorithm was used for data assimilation, the state variable was sediment content, the observation variable was turbidity, the covariance matrix of system error and observation error was obtained by maximum likelihood estimation, and the observation data was assimilated once in each calculation time step, which achieved a 15% improvement in model prediction accuracy. When the water flow velocity was greater than 2m / s and the sediment content was greater than 50kg / m 3 Under extreme conditions, high-precision monitoring of sediment content with an uncertainty better than 15% was achieved by increasing the measurement frequency and optimizing the sensor layout.

[0052] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for automatically sampling and monitoring the content and particle size of seawater quicksand, characterized in that: The method comprises: Use high-resolution seabed topography mapping and current velocity measurement equipment to collect topographic data of the target sea area and seawater flow velocity data; Identifying a flow pattern of seawater based on the collected flow velocity data, and predicting the behavior of sand particles based on the flow pattern, wherein the flow pattern includes information on the velocity and direction of the seawater flow; obtaining a movement pattern of sand particles under different flow conditions according to the flow pattern, wherein the movement pattern includes sand particle suspension data, sand particle settling data, and sand particle migration data; Collecting sand samples and classifying the sand samples to obtain physical properties of the sand particles, wherein the physical properties include particle size, density, and shape, and obtaining behavioral characteristics of the sand particles based on the different physical properties of the sand particles; The accumulation and dispersion laws of sand particles are obtained based on the movement behaviors of different types of sand particles under varying flow velocity conditions. Adjusting the sensor deployment density according to the movement pattern data and the sand accumulation and dispersion regularity data to obtain a sensor deployment density plan, wherein the sensor deployment density includes the layout, spacing, and number of sensors; A sensor deployment strategy is formulated according to the simulation results of the CFD simulation method, and a sensor network is designed according to the sensor deployment strategy.

2. The method according to claim 1, wherein The use of high-resolution seabed topography mapping and current velocity measurement equipment to collect topographic data of the target sea area and seawater flow velocity data includes: Using high-resolution seabed topography mapping equipment and a multi-beam bathymetry method, high-precision seabed topography data of the target sea area is obtained, wherein the mapping accuracy reaches the centimeter level and the mapping range covers the entire target sea area; Collect digital elevation model data of seabed topography by optimizing surveying paths and sampling frequency; High-precision velocity measurement equipment is used to obtain velocity data at different depths in the target sea area using the Doppler current profiler method. The velocity data has an accuracy of millimeters per second and the measurement range covers the entire water body. Acquiring three-dimensional flow field information, and establishing a hydrodynamic model based on the three-dimensional flow field information, wherein the three-dimensional flow field information includes flow velocity magnitude, flow velocity direction, and flow velocity change; Performing data fusion processing on the acquired seabed topography data and the flow velocity data, and generating numerically simulated topography data and flow field data through spatial interpolation and gridding methods; Optimize the grid resolution according to simulation requirements; Based on the fused terrain data and the flow velocity data, the water flow movement in the target sea area is simulated using the hydrodynamic model; The hydrodynamic model is verified by combining the seabed topography, tides and wind stress. The accuracy of the hydrodynamic model is verified with measured data, and the parameters of the hydrodynamic model are continuously optimized. Based on the hydrodynamic simulation results, the water flow characteristics of the target sea area are analyzed to obtain the flow velocity distribution, flow direction changes and vortex structure. Generate water flow motion images through data visualization technology; Couple the hydrodynamic simulation results with the seabed topography data to analyze the factors affecting the seabed topography on water flow. Through numerical experiments, the water flow changes under different terrain conditions are simulated to show the interaction mechanism between terrain and hydrodynamics; Based on the seabed topography data and the hydrodynamic simulation results, an ocean engineering design optimization platform is set up to integrate topographic mapping data, flow velocity measurement data and numerical simulation data.

3. The method according to claim 1, wherein The method includes identifying the flow pattern of seawater based on the collected flow velocity data and predicting the behavior of sand particles based on the flow pattern, wherein the flow pattern includes information on the height and direction of seawater flow velocity, including: Preprocessing the collected flow velocity data using a moving average filtering method and an outlier elimination method to obtain preprocessed flow velocity data; Unifying the pre-processed velocity data onto regular grid points through an interpolation algorithm to obtain regularized velocity data; Based on the regularized flow velocity data, an empirical orthogonal decomposition method is used to identify flow field modes and obtain the time coefficient and spatial distribution of each mode; By analyzing the time coefficient and spatial distribution, the dominant mode of flow velocity and flow direction change is determined, and the importance of each mode is measured by energy percentage; Using streamline tracing and flow tube visualization methods, the flow field at different times is visualized in three dimensions. The spatiotemporal evolution of the flow velocity and direction is presented through animation, showing the dynamic characteristics of the flow pattern. The time coefficient obtained by the empirical orthogonal decomposition method is used as the sample feature, and the K-means clustering method is used to classify the flow patterns. The clustering effect is evaluated by the silhouette coefficient, and the time proportion and conversion rules of various flow patterns are analyzed to show the dynamic change trend of the flow field. Analyzing the identified flow pattern through numerical simulation, and obtaining flow field velocity and pressure information by solving the Reynolds-averaged Navier-Stokes equations; Based on the process of water flow entrainment and sediment sedimentation, the transport behavior of sand under different flow patterns is simulated to obtain the suspended sediment concentration and scouring and deposition distribution, and to determine the impact mechanism of the flow pattern on the sediment dynamic process; Combining the flow pattern identification and numerical simulation results, a quantitative relationship model between the flow pattern and sand particle behavior is constructed; The quantitative relationship model is trained and validated by a machine learning algorithm, and the sand behavior is predicted based on the flow velocity data observation data.

4. The method according to claim 1, wherein The motion pattern of sand particles under different flow conditions is obtained according to the flow pattern, and the motion pattern includes sand particle suspension data, sand particle settling data, and sand particle migration data, including: According to the identified flow pattern, preset flow velocity conditions are extracted from the motion patterns under different flow velocity conditions, wherein the preset flow velocity conditions include average flow velocity and turbulence intensity, the preset flow velocity conditions are used as boundary conditions for analyzing the movement of sand particles, and the flow velocity conditions are mapped to the locations of the sand particles through an interpolation method; The DSMC direct simulation Monte Carlo method was used to perform random tracking simulations on the motion trajectory of individual sand particles under different flow conditions. Through the simulation statistics of a large number of sand particles, the statistical characteristics of the sand particles in the three motion modes of suspension, sedimentation and migration were obtained. The statistical characteristics include suspension time, sedimentation velocity and migration distance. In the DSMC simulation, the influence of sand particle characteristics on sand particle motion is taken into consideration. The equivalent particle size is introduced to correct the response of sand particle volume to hydrodynamic forces. The sphericity and roundness parameters are used to characterize the influence of sand particle shape on vertical and lateral motion. A quantitative relationship between particle characteristics and motion patterns is established. Analyze the regulatory effects of fluid viscosity and turbulence intensity on sand particle motion, determine flow conditions using the Reynolds number criterion, apply different mechanical models to sand particle motion under laminar and turbulent conditions, perform statistical analysis of velocity fluctuations under turbulent conditions, and reveal the relationship between turbulence intensity and sand particle suspension probability. In the simulation of sand particle movement, the boundary conditions and initial conditions are strictly controlled. A logarithmic velocity distribution is used to describe the velocity profile in the near-bottom region. A spectral method is used to generate a random velocity field that conforms to the energy distribution for velocity fluctuations under turbulent conditions. Using high-speed photography and particle image velocimetry, the movement of sand particles under preset flow conditions was tracked and measured to obtain real-world motion data on sand suspension, sedimentation, and migration. The accuracy of the model was verified by comparing it with the DSMC simulation results, and the model parameters were calibrated. The verified sand movement model is integrated into a two-dimensional water-sediment coupling numerical simulation program. The dynamic influence of the flow field on the sand movement is realized through the dynamic grid technology. By performing a collective statistical analysis of the migration trajectories of a large number of sand particles, the scouring and deposition distribution patterns under different flow velocity conditions are obtained.

5. The method according to claim 1, wherein The sand samples are collected and classified to obtain the physical properties of the sand particles, including particle size, density, and shape. The behavioral characteristics of the sand particles are obtained based on the different physical properties of the sand particles, including: The sand samples were collected using an equidistant grid method. The grid spacing was determined based on the average particle size and distribution uniformity of the sand samples. Sampling points were increased in areas with significant erosion and deposition changes to obtain a statistically representative sample data set. Pre-processing the collected sand samples includes removing impurities and organic matter, obtaining sand samples of different particle sizes by vibrating screening, and numbering and marking the samples of each particle size; The particle size distribution of sand particles was measured using a laser diffraction particle size analyzer, and the particle size frequency distribution curve and cumulative distribution curve were drawn to quantitatively describe the particle size composition characteristics of the sand particles. Determine the density of the sand using a pycnometer method. Add deionized water to the mark, weigh the total mass in the pycnometer, and calculate the density of the sand based on the difference in total mass in the pycnometer and the density of water. The shapes of the sand grains are quantitatively described by extracting the two-dimensional contours of the sand grains using image processing techniques, and dimensionless parameters characterizing the shapes are calculated. The dimensionless parameters include aspect ratio, roundness, and sphericity. The shapes of the sand grains are classified based on a cluster analysis method, and a mapping relationship between shape parameters and categories is established. Combined with the particle size, density and shape of the sand particles, a decision tree is used to comprehensively classify the sand samples. Classification rules are constructed through a recursive partitioning method. The depth of the tree and the number of leaf nodes are optimized using a cross-validation method. The classification accuracy is greater than a preset accuracy value, and different types of sand particle samples are obtained. According to the different types of sand particles, the instantaneous motion of sand particles in water flow is measured by the planar laser induced fluorescence method to obtain the preset motion mode of the sand particles, analyze the critical flow velocity conditions for the start-up and sedimentation of sand particles, and construct a quantitative relationship between the particle size, density and shape parameters and the sand particle motion parameters to display the behavioral characteristics of sand particles under the action of water flow, wherein the preset motion mode includes suspension, migration and rolling.

6. The method according to claim 1, wherein The accumulation and dispersion rules of sand particles obtained according to the movement behavior of different types of sand particles under varying flow velocity conditions include: For different types of sand, the initial distribution conditions of the sand and the bottom boundary conditions in the numerical simulation are determined based on the critical starting shear stress and settling velocity of the sand obtained in the flume test; Discrete the convection terms in an active upwind format and process the format of the numerical values ​​to simulate a series of preset flow conditions, including steady flow, tidal flow and wave; According to the simulation results, the analysis results of the scouring and deposition change process of sediment under different dynamic conditions are obtained; The curves of sediment concentration and scouring thickness changing with time are extracted from the analysis results. The main modes of sand movement are identified through spectrum analysis, showing the accumulation and dispersion patterns of sand under the action of water flow.

7. The method according to claim 1, wherein The sensor layout density is adjusted according to the motion pattern data and the accumulation and dispersion regularity data of the sand particles to obtain a sensor layout density scheme, wherein the sensor layout density includes the layout, spacing, and number of sensors, including: Collecting statistics on the flow field data, the accumulation and dispersion regularity data of the sand particles, and the movement pattern data; Obtain probability distribution functions of flow velocity, water depth, and sediment concentration; Combined with empirical orthogonal decomposition, the main variation modes of each parameter are extracted; Establish a data-driven model after dimensionality reduction to map flow field characteristics to sand particle motion; In response to different flow velocity conditions and the sand movement patterns described, a combination of uniform sampling and stratified sampling was used; Conduct research in a certain area and generate a large number of sampling points in the research area; The flow field and sand motion parameters at each sampling point are obtained through interpolation to form a data set for sensor layout optimization; The spatial correlation between sampling points was analyzed using the Kriging interpolation method; For areas with strong spatial correlation, the number of sensors deployed can be reduced, while for areas with drastic spatial changes, the deployment density needs to be increased; Determining the spacing of the sensor arrangements by constructing a variogram and a semi-variogram; Classifying the monitored flow velocity and sediment concentration according to data importance and difficulty in obtaining the data; For data whose importance coefficient is greater than the preset importance threshold and whose difficulty level is greater than the preset level threshold, a correlation model with other parameters is established through machine learning methods; Using observation data to train the association model to guide sensor layout; Under the premise of meeting the monitoring objectives, the sensor deployment density scheme is optimized; Obtaining the optimization result with the weighted sum of monitoring accuracy and cost as the objective function; The optimal layout scheme is found through simulated annealing and genetic heuristic search methods to control the number and layout positions of sensors; For the temporal resolution and spatial resolution of the sensor deployment density scheme, the influence of the temporal resolution and spatial resolution on data interpolation and inversion accuracy is obtained; For the time resolution, the sensor sampling frequency is required to comply with the Nyquist sampling theorem; For the stated spatial resolution, Taylor scaling was used to estimate the spatial variation scale of flow velocity and sediment concentration; After determining the sensor deployment density plan, an adaptive dynamic adjustment strategy is adopted to adjust and update the deployment plan online according to real-time feedback from monitoring data; For areas where the deviation between the observed value and the prediction of the correlation model is greater than a preset threshold, the sensor deployment density is increased to form a closed loop of monitoring-analysis-optimization; The method further includes: identifying the location and range of the sedimentation area through GIS analysis based on the sand sedimentation data, predicting the sedimentation trend using a spatial analysis method, and increasing the number of sensors deployed if the predicted sedimentation trend is a high sedimentation area, and reducing the number of sparsely deployed sensors if the predicted sedimentation trend is a low sedimentation area; calculating the path and stop points of the sand in the water flow based on the sand migration data, and arranging sensors based on the path and the stop points. If the path has intersections and turning areas, the number of sensors deployed will be increased; simulating the sand suspension data using a CFD simulation method to obtain the suspension height, velocity, and concentration distribution of the sand in the water flow; According to the sand sedimentation data, the location and range of the sedimentation area are identified through GIS analysis, and the sedimentation trend is predicted using a spatial analysis method. If the predicted sedimentation trend is a high sedimentation area, the number of sensor arrangements will be increased, and if the predicted sedimentation trend is a low sedimentation area, the number of sparse sensor arrangements will be reduced. Specifically, the sand sedimentation data is organized into a standard format, which includes sampling point coordinates, sampling time, sedimentation velocity, and sedimentation amount, and imported into a GIS platform for spatial processing; raster data layers of sedimentation velocity and sedimentation amount are generated through Kriging interpolation, and the resolution is calculated according to the density of the sampling points to perform threshold segmentation on the sedimentation velocity and sedimentation amount layers; according to sedimentation dynamics theory and empirical values, the area where the sedimentation velocity is greater than a preset velocity value and the sand sedimentation amount is greater than a preset sedimentation value is defined as a high sand sedimentation area, the boundary vector of the high sand sedimentation area is extracted, and area statistics are performed. The current scope of the high-sand deposition area is obtained through spatial analysis and spatial distribution analysis; the spatial autocorrelation of the high-sand deposition area is analyzed using GIS spatial statistical tools, and the Moran index and Getis-OrdG index are calculated to determine whether the high-sand deposition area has significant spatial agglomeration characteristics. If hot spots or cold spots exist, it indicates that the deposition process is affected by local factors and requires key monitoring; a grid of sedimentation monitoring sensors is deployed in the high-sand deposition area, and the grid density is calculated based on the spatial variability of the deposition rate; sentinel sensors are deployed at the edge of the high-sand deposition area to monitor the dynamic changes in the sand deposition range; time series analysis methods are used to perform trend extrapolation and periodic analysis on historical sedimentation data to identify the long-term evolution law and periodic fluctuation characteristics of the sand deposition process; the ARIMA model and spectral analysis method are used to quantitatively predict the sand deposition trend in the future, and the time scale can be determined according to decision-making needs; Based on the overlay analysis of the sand deposition trend prediction results and the spatial distribution map of high sand deposition areas in GIS, new high sand deposition areas and disappeared high sand deposition areas within the prediction period were identified. The sensor grid layout was adjusted to intensify monitoring in the newly added areas and reduce the number of monitoring points in the disappeared areas. A WebGIS-based sand deposition monitoring data management platform was developed to realize the real-time reception, storage, analysis and visualization of sand deposition monitoring data. The sand deposition risk level was determined based on the set sand deposition rate and range thresholds, and early warning information was generated and pushed to relevant departments. According to the data of the migration of sand particles, the path and the stop point of the sand particles in the water flow are calculated, and the sensors are arranged according to the path and the stop point. If the path has an intersection and a turning area, the number of sensors arranged will be increased. Specifically, the following steps are performed: collecting the monitoring data of sand particle migration under high flow conditions, the monitoring data of sand particle migration includes flow velocity, flow direction, sand particle concentration and particle size, using ADCP and OBS online monitoring equipment to obtain the data, and transmitting the data back in real time through wireless transmission technology; preprocessing the monitoring data of sand particle migration, using wavelet analysis method to remove high frequency noise and outliers, and selecting the optimal wavelet through cross validation. Wave basis and decomposition scale; using correlation analysis method and cross spectrum analysis method to study the quantitative relationship between the flow velocity and flow direction and the sand concentration and particle size, establish a multivariate regression model, and use the integrated learning method to optimize the generalization ability and prediction accuracy of the regression model; simulate the movement trajectory of the tracer sand according to the sand migration data, and then obtain the migration path of the sand; mine and analyze the big data of the sand movement trajectory obtained by numerical simulation, use density clustering algorithm to extract the skeleton structure of the trajectory, use spectral clustering algorithm to divide the branch structure of the trajectory, and then determine the intersection and turning point of the sand migration path; use trajectory complexity and coverage density to calculate the sand migration path. The importance of each key feature point is measured, and combined with measured data and mechanism analysis, the key nodes where sensors are preferentially deployed are finally determined, and the deployment density is proportional to the importance of the nodes; in non-critical areas of sand migration, the maximum entropy principle is used to determine the spacing of sensor deployments, and a sensor spacing deployment plan is obtained; a functional model of the deployment spacing, monitoring error and deployment cost is established through variational derivation, and then a simulated annealing algorithm is used for global optimization to obtain a sensor deployment plan that takes into account both the key nodes and ordinary areas; the sensor deployment plan that takes into account both the key nodes and ordinary areas and the sensor spacing deployment plan are combined to form a composite model. The unequally spaced sensor network topology uses a multi-hop self-organizing routing protocol to achieve communication between sensor nodes and aggregation nodes; integrating topographic and geomorphological data with hydrological and meteorological data, a comprehensive integrated model of sand migration is established, achieving a full range of simulation analysis from process mechanism to parameter identification, from static distribution to dynamic prediction; using the Bayesian network method, probabilistic reasoning and uncertainty analysis of sand migration laws are carried out, combining subjective and objective weighting methods to determine the key risk factors of sand migration, and using factor combination analysis and orthogonal experimental design to construct a multi-scenario and multi-condition sand migration early warning solution library, providing a decision-making basis for sensor deployment optimization and early warning threshold determination; The CFD simulation method is used to simulate the sand suspension data to obtain the suspension height, velocity and concentration distribution of the sand in the water flow, specifically including: capturing the influence of turbulent structure on the sand suspension process according to the LES large eddy simulation method, and combining the non-uniformity and non-spherical characteristics of the sand, using the DEM discrete element method to simulate the collision between the sand particles, and obtaining the multi-field coupling solution results between the water flow and the sand particles and between the sand particles; using the CFD simulation method to simulate the data of the sand suspension process, and obtaining the suspension height, velocity and concentration distribution of the sand particles under different flow conditions; tracking the motion trajectories of a large number of tracer particles through the Lagrangian tracer particle method, and analyzing the influence of the water flow velocity and turbulence intensity on the sand suspension characteristics; and constructing the quantitative relationship between the sand suspension parameters and the flow field characteristic quantities according to the influence law.

8. The method according to claim 1, wherein The step of formulating a sensor placement strategy based on the simulation results of the CFD simulation method and designing a sensor network based on the sensor placement strategy includes: The optimization goal is determined based on monitoring accuracy and cost, and the sensor's measurement range, resolution, and durability constraints are obtained. The placement strategy is searched through multi-objective genetic methods and ant colony methods to obtain a sensor placement solution. Generating a sensor placement strategy based on the speed magnitude obtained by the CFD simulation method and arranging the density of sensors according to the speed magnitude, wherein the greater the speed, the greater the density of sensors, and the smaller the speed, the smaller the density of sensors; The sensor network is designed according to the sensor arrangement strategy. In the process of designing the sensor network, a multi-level heterogeneous network architecture is adopted to design the topology structure and communication protocol of the sensor network.

Images