Sediment prediction method and system based on oscillation period strategy snow ablation optimization

Abstract

The invention discloses a sediment prediction method and system for oscillation period strategy snow ablation optimization, and the method comprises the steps: obtaining sediment data characteristics, enabling a drainage basin to comprise a plurality of nodes, and calculating the snow melting runoff of each node according to the sediment data characteristics; according to the snow melting runoff of each node, calculating the initial sand production concentration of each node; and constructing a sediment prediction model, inputting sediment characteristic data and corresponding actual sediment concentration to the sediment prediction model to obtain a trained sediment prediction model, inputting real-time sediment data to the trained sediment prediction model, and outputting predicted sediment concentration by the trained sediment prediction model. According to the method, historical sediment data are acquired, and a graph convolutional neural network is constructed for learning complex nonlinear mapping from snow melting-runoff generation to sediment transportation, so that the problems of weak generalization ability and strong data dependence of an existing snow melting period sediment prediction method are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water conservancy engineering, and in particular to a sediment prediction method and system with oscillation period strategy snow melting optimization. BACKGROUND

[0002] The sediment transport in high mountain and cold region basins has significant seasonal and event-driven characteristics. In particular, during the spring snowmelt period, the snowmelt rate directly affects the peak runoff and sediment yield. The existing methods have the following shortcomings: 1. The traditional energy balance model is physically reliable, but it is sensitive to small-scale heterogeneity (vegetation, surface shielding, and snow cover differences) and difficult to parameterize. 2. The simplified degree-day model has fewer parameters and faster calculation, but it cannot reflect the sudden impact of radiation and wind on snowmelt. 3. Machine learning methods are highly dependent on data and have weak generalization ability in areas with limited observations; pure data-driven models often struggle to maintain physical consistency. 4. Existing sediment transport models are mostly independent modules, lacking sufficient coupling with snowmelt-runoff, and lacking joint calibration and uncertainty quantification. Therefore, there is an urgent need for a sediment prediction method that can maintain physical correctness while utilizing data-driven learning of complex spatiotemporal patterns, and can assimilate observations in real time and quantify uncertainties, especially for optimized prediction during the snowmelt period. SUMMARY

[0003] To solve the above problems, the present application provides a sediment prediction method and system with oscillation period strategy snow melting optimization, which acquires historical sediment data and constructs a graph convolutional neural network to learn the complex nonlinear mapping between snowmelt-runoff generation and sediment transport, solving the problem of weak generalization and strong data dependence of existing snowmelt period sediment prediction methods.

[0004] To achieve the above purpose, the present application provides the following technical solutions: A sediment prediction method with oscillation period strategy snow melting optimization, comprising the following steps: S1. Acquire the basin sediment data, extract the data features of the sediment data, and obtain the sediment data features; S2. The basin includes a plurality of nodes, and the snowmelt runoff of each node is calculated according to the sediment data features; S3. Calculate the initial sediment concentration of each node according to the snowmelt runoff of each node; S4. Construct a sediment prediction model, the sediment prediction model is a graph neural network, input the sediment feature data and the corresponding actual sediment concentration into the sediment prediction model in turn, the sediment prediction model outputs the predicted sediment concentration, calculates the loss of the predicted sediment concentration according to the actual sediment concentration and the initial sediment concentration, adjusts the sediment prediction model according to the loss of the predicted sediment concentration, until the training times reach the preset threshold or the loss is less than the preset loss, the training is ended, and a trained sediment prediction model is obtained; S5. Input the real-time sediment data into the trained sediment prediction model, and the trained sediment prediction model outputs a predicted sediment concentration.

[0005] Further, in step S1, the sediment data features specifically include: Constructing a feature matrix , the feature matrix includes:

[0006] wherein, represents the air temperature of node i at time t, represents the precipitation of node i at time t, respectively represent the shortwave and longwave radiation of node i at time t, represents the wind speed of node i at time t, represents the humidity of node i at time t, represents the snow cover area of node i at time t, represents the snow water equivalent of node i at time t, represents the node terrain of node i, represents the land use of node i, represents the soil properties of node i.

[0007] Further, in step S2, the snowmelt runoff of each node is calculated according to the sediment data features, and the specific implementation manner is: for each node, the net energy is calculated according to the energy balance equation, and the specific calculation formula is:

[0008] wherein, is the net energy of node i at time t, represents the surface albedo of node i, is the sensible heat of node i, , is the air density, is the specific heat capacity at constant pressure, is the sensible heat exchange coefficient, is the wind speed of node i at time t, is the air temperature of node i at time t, is the surface temperature of the ground surface of node i at time t, is the latent heat flux of node i at time t, , wherein, is the air density, is the latent heat of vaporization of water, is the water vapor exchange coefficient, represents the surface saturation specific humidity of node i at time t, the air specific humidity of node i at time t, is the ground surface conductive flux of node i at time t; The snowmelt amount of node i at time t is calculated by the formula:

[0009] wherein, is the snowmelt amount of node i at time t, is the latent heat of ice and snow melting, is the water density, and the snowmelt runoff of node i is obtained :

[0010] wherein, represents the soil property of node i, is the infiltration function, is the land use of node i.

[0011] Further, in step S3, the initial sediment concentration of each node is calculated in the following manner: According to the soil erosion equation, we have:

[0012] wherein, is the soil erosion amount, is the runoff erosive force of node i at time t, , is the snowmelt runoff of node i at time t, is the water conservation measure of node i at time t, is the soil erodibility factor of node i, is the topographic slope length factor of node i, is the vegetation cover factor of node i, is the water conservation measure factor of node i. The initial sediment concentration of node i at time t is :

[0013] wherein, is the local runoff of node i at time t, is a small constant.

[0014] Further, before constructing the sediment prediction model in step S4, the following steps are further included: divide the basin into a graph G, wherein, node V represents a sub-basin or a section, and edge E represents the water and sediment transmission relationship.

[0015] Further, in step S4, the sediment prediction model calculates the predicted sediment concentration using the following steps: S41. Perform convolutional updates on the graph. The hidden states of each node are:

[0016] in, Let i be the set of neighboring nodes of node i, and j represent node j. Indicates the degree of node i. This represents the degree of node j. This represents the weight matrix in the l-th layer of the sediment prediction model. Let j be the hidden state of the j-th node in the l-th layer. Let be the hidden state of the i-th node in the (l+1)-th layer. Let L be the bias matrix of the l-th layer of the sediment prediction model. Indicates the activation function; S42. Using a long short-term memory recurrent neural network, temporal encoding of sediment characteristic data is performed in the time dimension:

[0017] Let be the time series of node i at time t. LSTM represents a Long Short-Term Memory recurrent neural network. Let i be the time series of node i at time t-1. This represents the sediment characteristic data of node i at time t; S43. Then the predicted sediment concentration at node i at time t. for:

[0018] For neural network functions, For the set of parameters of the neural network function, Let be the initial sediment yield concentration of node i at time t.

[0019] Furthermore, in step S4, the predicted sediment concentration loss is calculated based on the actual sediment concentration and the initial sediment yield concentration. The specific implementation method is as follows: The loss includes data error, conservation constraints, and physical consistency constraints, where data error... The calculation method is as follows:

[0020] Where N is the number of nodes, Let be the predicted sediment concentration at node i at time t. Let i be the actual sediment concentration at node i at time t; Conservation constraints The calculation method is as follows:

[0021] in, Let i be the set of neighboring nodes of node i, and j represent node j. This indicates that at time t, from node Flow to Node Volumetric flow rate, This indicates that at time t, node The concentration of sediment, This indicates that at time t, from node Total outflow volumetric flow rate This indicates that at time t, node The concentration of sediment, Let be the deposition amount at node i at time t; Physical consistency constraints The calculation method is as follows:

[0022] Let be the predicted sediment concentration at node i at time t. Let be the initial sediment yield concentration of node i at time t; The predicted loss in sediment concentration is:

[0023] in, These are the weighting coefficients.

[0024] Furthermore, step S5 also includes: After each prediction is completed, the actual sediment concentration is collected, and the sediment prediction model is updated by combining the ensemble Kalman filter with the actual sediment concentration.

[0025] Furthermore, in step S5, the trained sediment prediction model outputs the predicted sediment concentration, which also includes: providing a confidence interval for the predicted sediment concentration while outputting the predicted sediment concentration.

[0026] Through the above technical solution, the present invention has the following beneficial effects: The present invention organically combines a differentiable energy balance snow ablation module, a graph time-series deep learning model based on river network topology, and a physical constraint loss function, and supplements it with data assimilation and uncertainty quantification techniques, which effectively overcomes the shortcomings of existing methods. Furthermore, the present invention realizes integrated coupled modeling of snow ablation, runoff, and sediment transport, and uses ensemble Kalman filtering or particle filtering to achieve real-time observation assimilation and uncertainty measurement, which greatly improves the dynamic adaptability and reliability of prediction. Attached Figure Description

[0027] Figure 1 A schematic diagram of the overall process of a sediment prediction method for snowmelt optimization of an oscillation period strategy.

[0028] Figure 2 A schematic diagram of the structure of a sediment prediction system for snowmelt optimization of an oscillation period strategy in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0030] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0031] Example 1 Referring to Figure 1 A sediment prediction method for snowmelt optimization of an oscillation period strategy includes the following steps: S1. Obtain the basin sediment data, extract the data features of the sediment data, and obtain the sediment data features; S2. The basin includes a plurality of nodes, and the snowmelt runoff of each node is calculated according to the sediment data features; S3. Calculate the initial sediment concentration of each node according to the snowmelt runoff of each node; S4. Construct a sediment prediction model, the sediment prediction model is a graph neural network, input the sediment feature data and the corresponding actual sediment concentration into the sediment prediction model in turn, the sediment prediction model outputs the predicted sediment concentration, calculates the loss of the predicted sediment concentration according to the actual sediment concentration and the initial sediment concentration, adjusts the sediment prediction model according to the loss of the predicted sediment concentration, until the training times reach a preset threshold or the loss is less than a preset loss, the training is ended, and a trained sediment prediction model is obtained; S5. Input real-time sediment data into the trained sediment prediction model, and the trained sediment prediction model outputs the predicted sediment concentration.

[0032] The application provides a physically constrained graph temporal fusion sediment prediction method, which is used for optimizing prediction of basin sediment yield and transport in a snow melting period. The core idea is that a distributed energy balance snow melting module is used as a differentiable physical front end for calculating runoff and snow line changes generated by snow melting at a high spatiotemporal resolution; a graph representation is established based on river network topology, each sub-basin / section of the basin is taken as a node, and water flow and sediment routing relationship between nodes is represented by edges; a temporal deep learning module is constructed on the graph structure for learning a complex nonlinear mapping between snow melting-runoff generation and sediment transport, the module takes the output of the physical module as one of the input features, and can learn a residual model or a correction term; then a classic sediment transport equation is embedded in the network loss or an intermediate layer in a differentiable form, the network prediction is constrained by the physical constraint, and the model has physical consistency, and the output is a sediment yield and transport prediction with uncertainty measurement.

[0033] In an optional embodiment, in step S1, the sediment data features specifically include: Various data are acquired through automatic weather stations and sensors, satellites and unmanned aerial vehicles, water level meters, flow meters and turbidity meters, digital elevation models and GIS databases and other equipment, and are accessed through Internet of Things interfaces or data center APIs, and are processed through unified format conversion and spatiotemporal synchronization to form a complete input feature matrix, the feature matrix includes:

[0034] Among them, represents the air temperature of node i at time t, represents the precipitation of node i at time t, respectively represent the shortwave and longwave radiation of node i at time t, represents the wind speed of node i at time t, represents the humidity of node i at time t, represents the snow cover area of node i at time t, represents the snow water equivalent of node i at time t, represents the node terrain of node i, represents the land use of node i, represents the soil property of node i.

[0035] In an optional embodiment, in step S2, the snowmelt runoff of each node is calculated according to the sediment data features, and the specific implementation manner is that for each node, the net energy is calculated according to the energy balance equation, and the specific calculation formula is:

[0036] Among them, is the net energy of node i at time t, is the surface albedo of node i, is the sensible heat flux of node i, , is the air density, is the specific heat capacity at constant pressure, is the heat exchange coefficient, is the wind speed of node i at time t, is the air temperature of node i at time t, is the surface temperature of node i at time t, is the latent heat flux of node i at time t, , where, is the air density, is the latent heat of vaporization of water, is the water vapor exchange coefficient, is the surface saturation specific humidity of node i at time t, is the air specific humidity of node i at time t, is the surface conductive flux of node i at time t; The snowmelt amount of node i at time t is calculated by the formula:

[0037] where, is the snowmelt amount of node i at time t, is the latent heat of ice and snow melting, is the water density, and the snowmelt runoff of node i is obtained as:

[0038] where, represents the soil properties of node i, is an infiltration function, which refers to the process of snowmelt water infiltrating into the soil, and is used to describe how much snowmelt water can infiltrate into the soil and how much is left to form runoff, is the land use of node i.

[0039] The differential energy balance snow ablation is a sub-basin for each node, which helps to analyze the inflow and outflow of energy by the energy balance equation to determine the energy state and change trend of the system.

[0040] In an optional embodiment, in step S3, the initial sediment concentration of each node is calculated in the following specific manner: The sediment concentration is driven by snowmelt runoff and rainfall. The simplified and corrected universal soil loss equation is used, which has:

[0041] where,​ is the soil loss amount, is the runoff erosive force of node i at time t, , is the snowmelt runoff of node i at time t, is the water conservation measure of node i at time t, is the soil erodibility factor of node i, is the terrain slope length factor of node i, is the vegetation cover factor of node i, is the water conservation measure factor of node i; is the initial sediment concentration of node i at time t is:

[0042] wherein, is the local runoff of node i at time t, is a tiny constant to prevent zero denominator.

[0043] In an optional embodiment, before constructing the sediment prediction model in step S4, further comprising: dividing the watershed into a graph G, wherein, node V represents a sub-watershed or a section, and edge E represents a water-sediment transmission relationship.

[0044] In an optional embodiment, in step S4, the sediment prediction model calculating the predicted sediment concentration comprises the following steps: S41. Convolutionally updating the graph, each node has a hidden state:

[0045] wherein, is the set of adjacent nodes of node i, and j represents node j, represents the degree of node i, represents the degree of node j, represents the weight matrix in the lth layer network of the sediment prediction model, is the hidden state of the jth node in the lth layer, is the hidden state of the ith node in the (l+1)th layer, is the bias matrix of the lth layer of the sediment prediction model, represents an activation function; S42. Using a long short-term memory recurrent neural network to perform time series encoding on the sediment feature data in the time dimension:

[0046] is the time series of node i at time t, LSTM denotes a long short-term memory recurrent neural network, is the time series of node i at time t-1, denotes the sediment characteristic data of node i at time t; S43. the predicted sediment concentration of node i at time t is is:

[0047] is a neural network function, which is used to learn the prediction results of the physical model. It can be a feedforward neural network (FNN), a recurrent neural network (RNN), a long short-term memory network (LSTM), etc., depending on the structure of the time series , is a set of parameters of the neural network function, is a set of input features of the neural network, is a set of parameters of the neural network, which includes all the weights and biases in the neural network, is the initial sediment concentration of node i at time t. Through the training process, a large amount of historical observation data (including input features and the corresponding actual sediment concentration) is used to continuously adjust by using back propagation algorithm and other optimization methods, so that the final prediction of the whole model is as close to the true value as possible.

[0048] In an optional embodiment, in step S4, the loss of the predicted sediment concentration is calculated according to the actual sediment concentration and the initial sediment concentration, and the specific implementation manner is as follows: The loss includes data error, conservation constraint and physical consistency constraint, wherein the calculation manner of the data error is:

[0049] wherein N is the number of nodes, is the predicted sediment concentration of node i at time t, is the actual sediment concentration of node i at time t, specifically, the sediment concentration data directly obtained by using actual measurement means, including using sediment samplers, turbidity meters, optical or acoustic sensors and other instruments in the field; When simulating physical processes on computers, numerical errors will accumulate continuously due to discrete and approximate calculations, which may cause these quantities that should be conserved to gradually increase or decrease, thereby distorting and even collapsing the simulation. Therefore, conservation constraints are set to ensure the stability and authenticity of long-time simulation, and the calculation manner of the conservation constraint is:

[0050] wherein, is the set of neighboring nodes of node i, j represents node j, represents the volume flow from node to node at time t, represents the sediment concentration of node at time t, represents the total volume flow from node at time t, represents the sediment concentration of node at time t, is the deposition amount of node i at time t; The physical consistency constraint refers to a set of conditions or restrictions in a physical model or simulation that ensure that the physical phenomena described by the model conform to actual physical laws, and the calculation method of the physical consistency constraint is as follows:

[0051] is the predicted sediment concentration of node i at time t, is the initial sediment concentration of node i at time t; The loss of the predicted sediment concentration is

[0052] wherein, is a weight coefficient.

[0053] In an optional embodiment, in step S5, further comprising: After each prediction is completed, the actual sediment concentration is collected, and the sediment prediction model is updated using the ensemble Kalman filter combined with the actual sediment concentration.

[0054] When a new observation arrives, the ensemble Kalman filter (EnKF) is used for updating; The prediction step is:

[0055] The update step is:

[0056] wherein, is the predicted state, is the external driving or control input at time t, usually including boundary conditions (upstream inflow flow, downstream water level, sediment input), source and sink terms (lateral inflow, pumping, rainfall, evaporation), etc., is the updated state, is the observation value, is the Kalman gain.

[0057] The ensemble Kalman filter is an advanced method for dynamic system state estimation, especially suitable for processing nonlinear systems and large-scale data, and the ensemble Kalman filter uses the data to update the state in real time whenever new observation data arrives, ensuring the dynamics and timeliness of the model, making the state estimation more accurate.

[0058] In an optional embodiment, in step S5, the trained sediment prediction model outputs the predicted sediment concentration, and further comprises: outputting the predicted sediment concentration while giving the confidence interval of the predicted sediment concentration.

[0059] Specifically, the sediment concentration and sediment discharge of each node are outputted, and the sediment discharge of each node is calculated by the formula:

[0060] wherein, represents the total volume flow rate of the node at time t.

[0061] And the 95% confidence interval is given as the uncertainty interval:

[0062] Thus, the sediment discharge prediction of each node with uncertainty measure (confidence interval) is outputted, and support is provided for subsequent decision-making.

[0063] The present application effectively overcomes the shortcomings of existing methods by organically combining the differentiable energy balance snowmelt module, the graph time series deep learning model based on river network topology, and the physical constraint loss function, and supplemented by data assimilation and uncertainty quantification technology. Compared with the traditional energy balance model, the present method introduces learnable parameters while maintaining physical credibility, can automatically adapt to small-scale heterogeneity, and improve the parameter identifiability and applicability of the model. Compared with the simplified degree-day model, the present method comprehensively considers sudden impact factors such as radiation and wind speed, and more realistically reflects the snowmelt process. Compared with the pure machine learning method, the present method embeds the mass conservation of sediment and the physical consistency constraint in the training process, ensures that the prediction result conforms to the natural law, and enhances the generalization ability in the data scarce area through transfer learning and Bayesian hierarchical strategy. Compared with the existing independent sediment transport model, the present method realizes the integrated coupling modeling of snowmelt, runoff and sediment transport, and realizes real-time observation assimilation and uncertainty measurement by using the ensemble Kalman filter or particle filter, greatly improving the dynamic adaptability and reliability of the prediction.

[0064] ​​In conclusion, the present application can significantly improve the precision and stability of sediment prediction during snow melting period, and provide more scientific decision basis for disaster prevention and mitigation, water resource management and engineering operation in a basin, and has outstanding innovation and application value.

[0065] Example 2 Referring to Figure 2 A sediment prediction system with oscillation period strategy snow melting optimization, comprising: A data acquisition module is configured to acquire basin sediment data, extract data features of the sediment data, and obtain sediment data features. A runoff calculation module is configured to calculate snowmelt runoff of each node according to the sediment data features, wherein the basin comprises a plurality of nodes. An initial calculation module is configured to calculate the initial sediment concentration of each node according to the snowmelt runoff of each node. A sediment prediction module is configured to build a sediment prediction model, wherein the sediment prediction model is a graph neural network, the sediment feature data and the corresponding actual sediment concentration are input into the sediment prediction model in sequence, the sediment prediction model outputs the predicted sediment concentration, the loss of the predicted sediment concentration is calculated according to the actual sediment concentration and the initial sediment concentration, the sediment prediction model is adjusted according to the loss of the predicted sediment concentration, and the training is ended until the number of training reaches a preset threshold or the loss is less than a preset loss, thereby obtaining the trained sediment prediction model. A prediction output module is configured to input real-time sediment data into the trained sediment prediction model, and the trained sediment prediction model outputs the predicted sediment concentration.

[0066] The embodiments disclosed in the specification are only an example of the one-sided features of the present application, and the protection scope of the present application is not limited to this embodiment. Any other functionally equivalent embodiments fall within the protection scope of the present application. For those skilled in the art, other various corresponding changes and modifications can be made according to the above described technical solutions and concepts, and all these changes and modifications should belong to the protection scope of the claims of the present application.

Claims

1. A method for predicting sediment deposition based on snow ablation optimization using an oscillating period strategy, characterized in that, Includes the following steps: S1. Obtain watershed sediment data, extract data features from the sediment data, and obtain sediment data features; S2. The watershed includes several nodes, and the snowmelt runoff of each node is calculated based on the characteristics of sediment data; S3. Calculate the initial sediment yield concentration for each node based on the snowmelt runoff at each node; S4. Construct a sediment prediction model, which is a graph neural network. Sediment feature data and corresponding actual sediment concentrations are sequentially input into the sediment prediction model. The sediment prediction model outputs the predicted sediment concentration. The loss of the predicted sediment concentration is calculated based on the actual sediment concentration and the initial sediment production concentration. The sediment prediction model is adjusted based on the loss of the predicted sediment concentration until the number of training times reaches a preset threshold or the loss is less than the preset loss. The training ends, and a trained sediment prediction model is obtained. S5. Input real-time sediment data into the trained sediment prediction model, and the trained sediment prediction model outputs the predicted sediment concentration.

2. The method for predicting sediment deposition using an oscillation period strategy for snow ablation optimization according to claim 1, characterized in that, In step S1, the sediment data features specifically include: Constructing the feature matrix The feature matrix include: in, This represents the temperature at node i at time t. This represents the precipitation at node i at time t. These represent the shortwave and longwave radiation of node i at time t, respectively. Let represent the wind speed at node i at time t. This represents the humidity of node i at time t. Let represent the snow cover area of ​​node i at time t. This represents the snowmelt equivalent of node i at time t. Represents the node terrain of node i. Represents the land use of node i. This represents the soil properties of node i.

3. The method for predicting sediment deposition using an oscillation period strategy for snow ablation optimization according to claim 2, characterized in that, In step S2, the calculation of snowmelt runoff at each node based on sediment data characteristics is specifically implemented as follows: For each node, the net energy is calculated according to the energy balance equation, and the specific calculation formula is as follows: in, Let be the net energy of node i at time t. Represents the surface albedo of node i. For the induced heat at node i, , air density, For isobaric specific heat capacity, The heat transfer coefficient is the coefficient of performance. Let be the wind speed at node i at time t. Let be the air temperature at node i at time t. Let be the surface temperature of node i at time t. Let be the latent heat flux of node i at time t. In the formula, air density, The latent heat of vaporization of water, The water vapor exchange coefficient, This represents the surface saturation specific moisture of node i at time t. This indicates the air humidity at node i at time t. Let be the surface conduction flux of node i at time t; The formula for calculating the snow melt amount at node i at time t is: in, Let be the amount of snow melt at node i at time t. The latent heat of melting ice and snow, Given the water density, the snowmelt runoff at node i can be obtained. : in, This represents the soil properties of node i. Let be the infiltration function. Let i be the land use of node i.

4. The method for predicting sediment deposition using an oscillation period strategy for snow ablation optimization according to claim 3, characterized in that, In step S3, the initial sediment yield concentration of each node is calculated as follows: According to the soil loss equation, we have: in, Soil erosion Let i be the runoff erosion force at time t. , Let be the snowmelt runoff at node i at time t. For the water conservation measures at node i at time t, Let i be the soil erodibility factor at node i. Let i be the terrain slope length factor. Let be the vegetation cover factor of node i. For node i, the water conservation measure factor; Then the initial sediment yield concentration of node i at time t for: in, Let be the local runoff of node i at time t. It is a tiny constant.

5. The method for predicting sediment deposition using an oscillation period strategy for snow ablation optimization according to claim 1, characterized in that, In step S4, before constructing the sediment prediction model, the following steps are also included: The watershed is divided into graph G. In this context, node V represents a sub-basin or cross section, and edge E represents the water and sediment transport relationship.

6. The method for predicting sediment deposition using an oscillation period strategy for snow ablation optimization according to claim 5, characterized in that, In step S4, the sediment prediction model calculates the predicted sediment concentration using the following steps: S41. Perform convolutional updates on the graph. The hidden states of each node are: in, Let i be the set of neighboring nodes of node i, and j represent node j. Indicates the degree of node i. This represents the degree of node j. This represents the weight matrix in the l-th layer of the sediment prediction model. Let j be the hidden state of the j-th node in the l-th layer. Let be the hidden state of the i-th node in the (l+1)-th layer. Let L be the bias matrix of the l-th layer of the sediment prediction model. Indicates the activation function; S42. Using a long short-term memory recurrent neural network, temporal encoding of sediment characteristic data is performed in the time dimension: Let be the time series of node i at time t. LSTM represents a Long Short-Term Memory recurrent neural network. Let i be the time series of node i at time t-1. This represents the sediment characteristic data of node i at time t; S43. Then the predicted sediment concentration at node i at time t. for: For neural network functions, For the set of parameters of the neural network function, Let be the initial sediment yield concentration of node i at time t.

7. The method for predicting sediment deposition using an oscillating period strategy for snow ablation optimization according to claim 6, characterized in that, In step S4, the predicted sediment concentration loss is calculated based on the actual sediment concentration and the initial sediment yield concentration. The specific implementation method is as follows: The loss includes data error, conservation constraints, and physical consistency constraints, where data error... The calculation method is as follows: Where N is the number of nodes, Let be the predicted sediment concentration at node i at time t. Let i be the actual sediment concentration at node i at time t; Conservation constraints The calculation method is as follows: in, Let i be the set of neighboring nodes of node i, and j represent node j. This indicates that at time t, from node Flow to Node Volumetric flow rate, This indicates that at time t, node The concentration of sediment, This indicates that at time t, from node Total outflow volumetric flow rate This indicates that at time t, node The concentration of sediment, Let be the deposition amount at node i at time t; Physical consistency constraints The calculation method is as follows: Let be the predicted sediment concentration at node i at time t. Let be the initial sediment yield concentration of node i at time t; The predicted loss in sediment concentration is: in, These are the weighting coefficients.

8. The method for predicting sediment deposition using an oscillating period strategy for snow ablation optimization according to claim 7, characterized in that, Step S5 also includes: After each prediction is completed, the actual sediment concentration is collected, and the sediment prediction model is updated by combining the ensemble Kalman filter with the actual sediment concentration.

9. The method for predicting sediment deposition using an oscillating period strategy for snow ablation optimization according to claim 1, characterized in that, In step S5, the trained sediment prediction model outputs the predicted sediment concentration, and also includes: outputting the predicted sediment concentration while providing the confidence interval of the predicted sediment concentration.

10. A sediment prediction system for snow ablation optimization using an oscillating period strategy, characterized in that, include: The data acquisition module is used to acquire watershed sediment data, extract data features from the sediment data, and obtain sediment data features. The runoff calculation module is used to calculate the snowmelt runoff of each node in the watershed, which includes several nodes, based on the characteristics of sediment data. The initial calculation module is used to calculate the initial sediment yield concentration of each node based on the snowmelt runoff of each node; The sediment prediction module is used to construct a sediment prediction model, which is a graph neural network. Sediment feature data and the corresponding actual sediment concentration are sequentially input into the sediment prediction model. The sediment prediction model outputs the predicted sediment concentration. The loss of the predicted sediment concentration is calculated based on the actual sediment concentration and the initial sediment production concentration. The sediment prediction model is adjusted based on the loss of the predicted sediment concentration until the number of training times reaches a preset threshold or the loss is less than the preset loss. The training ends and a trained sediment prediction model is obtained. The prediction output module is used to input real-time sediment data into the trained sediment prediction model, and the trained sediment prediction model outputs the predicted sediment concentration.

Images