Prediction method of sediment cross-section erosion variation under influence of sluice

Abstract

The present application relates to the technical field of sediment cross-section erosion, and particularly relates to a method for predicting sediment cross-section erosion variation under the influence of a water gate. A current erosion and deposition topography sequence corresponding to a current time of a target gate is obtained, and future precipitation-runoff data, future water level difference between upstream and downstream, and future tide change data within a future preset time length corresponding to the target gate are obtained; based on the future precipitation-runoff data, the future water level difference between upstream and downstream, and the future tide change data, future gate scheduling data corresponding to the target gate is determined; the current erosion and deposition topography sequence, the future precipitation-runoff data, the future water level difference between upstream and downstream, the future tide change data, and the future gate scheduling data are input into a preset erosion and deposition variation prediction model, and output of sediment cross-section erosion variation within the future preset time length corresponding to the target gate is output. Timely and reliable technical support is provided for hydraulic scheduling and risk management, and safe and efficient operation of a tidal gate in an estuary is realized.

Description

Technical Field

[0001] This invention relates to the field of sediment cross-sectional scour technology, specifically to a method for predicting sediment cross-sectional scour changes under the influence of sluice gates. Background Technology

[0002] In the operation and management of estuary tide barriers, accurate prediction of downstream scouring and sedimentation changes is crucial for ensuring the safety and efficient operation of water conservancy projects. The scouring and sedimentation process in the downstream area of ​​an estuary tide barrier is significantly dynamic and complex due to the coupled influence of multiple factors, including gate scheduling, upstream inflow, downstream tide level, and sediment sources. Long-term sedimentation leads to riverbed elevation, reducing the river's flood discharge cross-section and severely inhibiting the effective flood discharge capacity of the gate, making it highly susceptible to flooding during flood season. Conversely, long-term scouring may expose or damage the bottom structure of the gate, creating a risk similar to bridge pier scouring, threatening the structural safety of the tide barrier, and consequently affecting the normal functioning of multiple functions such as flood control, drainage, and water resource utilization in the region.

[0003] Currently, the traditional method for analyzing scour and sedimentation changes in the downstream area of ​​estuary tidal gates mainly relies on numerical models such as Delft3D. These models, based on hydrodynamics and sediment transport theory, can achieve high prediction accuracy by finely depicting water flow and sediment exchange processes. However, their operation requires significant computational resources and time, and the model parameter calibration and verification processes are cumbersome, resulting in high operating costs. In practical water conservancy project management, there is often a need for rapid evaluation of scheduling strategies and rolling predictions of multiple scenarios. For example, when responding to sudden floods and formulating water resource allocation plans, it is necessary to output predictions of scour and sedimentation changes under multiple scheduling schemes within a short period of time to assist decision-making. However, traditional numerical models struggle to meet the timeliness requirements in such scenarios.

[0004] Therefore, there is an urgent need for an efficient method to predict sediment cross-sectional scour changes under the influence of sluice gates with multiple uncertain inputs, so as to provide timely and reliable technical support for hydraulic engineering scheduling and risk management, and realize the safe and efficient operation of estuary tidal gates. Summary of the Invention

[0005] In view of this, the present invention provides a method for predicting the scour change of sediment cross section under the influence of sluice gates, in order to solve the problem of the urgent need for an efficient method for predicting the scour change of sediment cross section under the influence of sluice gates that can adapt to multiple uncertain inputs.

[0006] In a first aspect, the present invention provides a method for predicting changes in sediment cross-section scour under the influence of a sluice gate, the method comprising:

[0007] Obtain the current erosion and deposition terrain sequence corresponding to the target gate at the current moment, and obtain the future precipitation-runoff data, future upstream and downstream water level difference and future tidal level change data within the future preset time period corresponding to the target gate;

[0008] Based on future precipitation-runoff data, future upstream and downstream water level differences, and future tidal level changes, determine the future gate scheduling data corresponding to the target gate;

[0009] The current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are input into the preset scour change prediction model, and the output is the scour change of the sediment cross section corresponding to the target gate within the future preset time period.

[0010] The scour and deposition change prediction method for sluice gates provided in this application acquires the current scour and deposition topography sequence corresponding to the target gate at the current moment, achieving precise focusing on a specific area, avoiding redundant interference from data across the entire region, and ensuring a strong correlation between the model input data and the target scenario. It acquires precipitation-runoff, future upstream-downstream water level differences, and future tidal level changes within a preset future timeframe, comprehensively covering the core hydrological elements affecting scour and deposition. These data represent potential future influencing factors, enabling the model to extrapolate scour and deposition changes based on forward-looking information, improving the lead time and comprehensiveness of predictions, and meeting the needs of water conservancy decision-making for predicting future trends.

[0011] Then, based on future precipitation-runoff data, future upstream and downstream water level differences, and future tidal level changes, the future gate scheduling data corresponding to the target gate is determined, changing the traditional experience-based scheduling model. By quantitatively analyzing the impact of various factors on gate scheduling, dynamic and scientific scheduling decisions are achieved. For example, when heavy precipitation is predicted to cause a high water level difference, gate opening adjustments can be planned in advance to improve flood control and disaster relief capabilities and water resource allocation efficiency.

[0012] Next, the current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream and downstream water level differences, future tidal level changes, and future gate scheduling data are input into a pre-defined scour and sedimentation change prediction model. This integrates historical topography, future environmental variables, and scheduling decision factors to form a multi-dimensional input system. Different data complement and verify each other. For example, the current topography determines the initial state of sediment, future precipitation-runoff data drives the change process, and gate scheduling regulates water flow and sediment transport, collectively improving the model's ability to simulate complex scour and sedimentation processes. The pre-defined scour and sedimentation change prediction model outputs the scour and sedimentation changes at the cross-section within a pre-defined timeframe, providing intuitive and quantitative prediction results for water conservancy project management. Managers can use the scour and sedimentation change trends to plan river dredging, dam reinforcement, and other engineering measures in advance, or optimize water resource allocation schemes. Simultaneously, the prediction results can verify the rationality of scheduling decisions, forming a virtuous cycle of "data-decision-prediction-feedback," enhancing the scientific and reliable operation of the water conservancy system. This provides timely and reliable technical support for hydraulic engineering scheduling and risk management, ensuring the safe and efficient operation of estuary tidal gates.

[0013] In one optional implementation, the preset scour change prediction model includes a scour-deposition trend classification model and a sediment scour change prediction model; the current scour-deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are input into the preset scour change prediction model, and the model outputs the sediment cross-sectional scour change corresponding to the target gate within a preset future time period, including:

[0014] The current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are input into the erosion and deposition trend classification model to determine the sediment evolution trend category corresponding to the target gate.

[0015] The sediment evolution trend category, current scour and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are input into the preset sediment scour change prediction model, and the output is the sediment cross-sectional scour change corresponding to the target gate within the future preset time period.

[0016] In one optional implementation, the current erosion and deposition topography sequence, future precipitation-runoff data, future upstream-downstream water level difference, future tidal level change data, and future gate scheduling data are input into the erosion and deposition trend classification model to determine the sediment evolution trend category corresponding to the target gate, including:

[0017] Feature extraction is performed on the current erosion and deposition terrain sequence to obtain the terrain data features corresponding to the current erosion and deposition terrain sequence;

[0018] The target fusion feature is generated by fusing future precipitation-runoff data, future upstream and downstream water level differences, future tidal level change data, and future gate scheduling data.

[0019] Topographic data features and target fusion features are input into the scour and sedimentation trend classification model to determine the sediment evolution trend category corresponding to the target gate.

[0020] In one optional implementation, future precipitation-runoff data, future upstream-downstream water level differences, future tidal level changes, and future gate scheduling data are fused to generate target fusion features, including:

[0021] Based on a preset adaptive phase space reconstruction algorithm, the chaotic features corresponding to future precipitation-runoff data, future upstream and downstream water level differences, future tidal level changes, and future gate scheduling data are analyzed. The chaotic features include the maximum Lyapunov exponent, fractal dimension, and chaotic frequency spectrum.

[0022] The chaotic features corresponding to future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are respectively input into the reinforcement learning model in the scour and sedimentation trend classification model to determine the target fusion strategy corresponding to each chaotic feature.

[0023] Based on the target fusion strategy, various chaotic features are fused to generate target fused features.

[0024] In one optional implementation, topographic data features and target fusion features are input into a sedimentation trend classification model to determine the sediment evolution trend category corresponding to the target gate, including:

[0025] Topographic data features and target fusion features are input into the quantum decision tree in the scour and sedimentation trend classification model; each decision node in the quantum decision tree is an independent quantum system.

[0026] Initialize the quantum state of each decision node in the quantum decision tree;

[0027] The input terrain data features and target fusion features are encoded using qubits to generate the current quantum state; each sub-feature in the target fusion features corresponds to one or more qubits; one qubit can represent multiple states simultaneously.

[0028] Perform quantum gate operations on the current quantum state to generate the node quantum states corresponding to each decision node;

[0029] Calculate the quantum probability amplitude of the node quantum state corresponding to each decision node;

[0030] The classification path is determined based on each quantum probability amplitude;

[0031] Based on each classification path, proceed to the next level decision node until a leaf node is reached;

[0032] Based on the previously calculated quantum probability amplitude and path information, the leaf nodes determine the sediment evolution trend category corresponding to the target gate.

[0033] In one optional implementation, a quantum gate operation is performed on the current quantum state to generate the node quantum states corresponding to each decision node, including:

[0034] Identify the current quantum state, determine the real-time dynamic characteristics corresponding to the current quantum state, and identify the core and auxiliary characteristics in the current quantum state;

[0035] Based on real-time dynamic characteristics, determine the target quantum gate combination corresponding to the current quantum state;

[0036] Entanglement calculations are performed on the core features based on the controlled NOT gates in the target quantum gate combination to generate higher-order entangled features;

[0037] Based on the control NOT gate in the target quantum gate combination, auxiliary features and core features are entangled to form a star-shaped topology;

[0038] Perform a nonlinear transformation on the current quantum state to generate the nonlinear characteristics corresponding to the current quantum state;

[0039] Based on high-order entanglement features, star topology, and nonlinear features, the node quantum states corresponding to each decision node are generated.

[0040] In one optional implementation, the sediment evolution trend category, the current scour and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are input into a preset sediment scour change prediction model, and the output is the sediment cross-sectional scour change corresponding to the target gate within a preset time period in the future, including;

[0041] Input the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the time dependency network of the preset sediment erosion change prediction model, and output the time dependency features.

[0042] Input the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the pre-set sediment erosion change prediction model's structural feature extraction network, and output the topological features.

[0043] The time-dependent features and topological features are fused to output a fused prediction feature;

[0044] Based on the integrated prediction characteristics, the change in sediment cross-section scour within a preset time period corresponding to the target gate is output.

[0045] In one optional implementation, the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are input into a time-dependent network of a preset sediment erosion change prediction model, and the output time-dependent features include:

[0046] Input the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the time dependency network of the preset sediment erosion change prediction model.

[0047] The weights in the time-dependent network determine the network. At each time step, the input data is evaluated to determine the importance of the input data at each time step to the current scour and deposition terrain sequence and the changes in sediment cross-section scour within a preset time period in the future.

[0048] Based on the importance of the input data at each time step, determine the weight information corresponding to the input data at each time step;

[0049] Based on the weight information corresponding to the input data at each time step, dynamically adjust the update gate and forget gate in the time-dependent network;

[0050] Based on the adjusted update gate and forget gate at each time step, time series features are extracted from the input data, and time-dependent features are output.

[0051] In one optional implementation, the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data are input into a pre-defined sediment erosion change prediction model's structural feature extraction network, outputting topological features, including:

[0052] Based on the current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level differences, future tidal level changes, and future gate scheduling data, a target map structure is constructed.

[0053] The graph structure is extracted using a structural feature extraction network, and the topological features are output.

[0054] In one optional implementation, based on future precipitation-runoff data, future upstream-downstream water level differences, and future tidal level changes, the future gate scheduling data corresponding to the target gate is determined, including:

[0055] Acquire historical precipitation-runoff data, historical upstream and downstream water level difference, historical tidal level change data, and historical gate scheduling data corresponding to historical precipitation-runoff data, historical upstream and downstream water level difference, and historical tidal level change data.

[0056] A cause-effect graph structure is constructed based on historical precipitation-runoff data, historical upstream and downstream water level differences, historical tidal level changes, and historical gate scheduling data.

[0057] The cause-effect graph structure is extended into a dynamic structural equation model to capture time-delayed causal relationships between variables;

[0058] By substituting future precipitation-runoff data, future upstream-downstream water level differences, and future tidal level changes into the time-delay causal relationship, the future gate scheduling data corresponding to the target gate can be determined. Attached Figure Description

[0059] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating the method for predicting changes in sediment cross-section scour under the influence of a sluice gate according to an embodiment of the present invention.

[0061] Figure 2 This is a flowchart illustrating a method for predicting sediment cross-sectional scour changes under the influence of a sluice gate, according to another embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] It should be noted that the method for predicting sediment cross-section scour changes provided in this application can be executed by a device for predicting sediment cross-section scour changes. This device can be implemented as part or all of a computer device through software, hardware, or a combination of both. The computer device can be a server or a terminal. In this application embodiment, the server can be a single server or a server cluster composed of multiple servers. The terminal in this application embodiment can be a smartphone, personal computer, tablet computer, wearable device, or other intelligent hardware device such as a smart robot. In the following method embodiments, the execution subject is always described using an electronic device as an example.

[0064] According to an embodiment of the present invention, an embodiment of a method for predicting changes in sediment cross-section scour under the influence of a sluice gate is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0065] This embodiment provides a method for predicting sediment cross-sectional scour changes under the influence of a sluice gate, which can be used in the aforementioned electronic equipment. Figure 1This is a flowchart of a method for predicting sediment cross-sectional scour changes under the influence of a sluice gate according to an embodiment of the present invention, as shown below. Figure 1 As shown, the process includes the following steps:

[0066] Step S101: Obtain the current scouring and sedimentation terrain sequence corresponding to the target gate at the current moment, and obtain the future precipitation-runoff data, future upstream and downstream water level difference and future tidal level change data within the future preset time period corresponding to the target gate.

[0067] Specifically, the electronic device can receive the current scouring and sedimentation terrain sequence corresponding to the target gate at the current moment input by the user, and obtain future precipitation-runoff data, future upstream and downstream water level difference and future tidal level change data within the future preset time period corresponding to the target gate. It can also receive the current scouring and sedimentation terrain sequence corresponding to the target gate at the current moment sent by other devices, and obtain future precipitation-runoff data, future upstream and downstream water level difference and future tidal level change data within the future preset time period corresponding to the target gate.

[0068] Optionally, the electronic equipment can also acquire topographic point cloud data corresponding to the target gate at the current moment based on sonar depth sounding technology. The topographic point cloud data is processed to generate a continuous and smooth sequence of current erosion and deposition terrain. The electronic equipment can obtain future precipitation data for the target gate within a preset time period based on a weather website, and future runoff data for the target gate within a preset time period based on a hydrological monitoring station, generating future precipitation-runoff data for the target gate within a preset time period. The electronic equipment can also predict the future upstream and downstream water level difference for the target gate within a preset time period based on a water level monitoring system, and predict future tidal level changes for the target gate within a preset time period based on a tidal level monitoring station.

[0069] This application embodiment does not specifically limit the method by which the electronic device obtains the current erosion and deposition terrain sequence corresponding to the target gate at the current moment, and obtains the future precipitation-runoff data, future upstream and downstream water level difference and future tidal level change data within a future preset time period corresponding to the target gate.

[0070] Step S102: Based on future precipitation-runoff data, future upstream and downstream water level differences, and future tidal level changes, determine the future gate scheduling data corresponding to the target gate.

[0071] Specifically, electronic devices can determine the future gate scheduling data corresponding to the target gate based on the correlation between future precipitation-runoff data, future upstream and downstream water level differences, future tidal level change data, and future gate scheduling data.

[0072] This step will be explained in detail below.

[0073] Step S103: Input the current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the preset scour change prediction model, and output the scour change of the sediment cross section corresponding to the target gate within the future preset time period.

[0074] Specifically, the electronic device can input the current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into a preset scour change prediction model. The preset scour change prediction model extracts features from the input data and, based on the extracted features, outputs the scour change of the sediment cross section corresponding to the target gate within a preset time period in the future.

[0075] This step will be explained in detail below.

[0076] The scour and deposition change prediction method for sluice gates provided in this application acquires the current scour and deposition topography sequence corresponding to the target gate at the current moment, achieving precise focusing on a specific area, avoiding redundant interference from data across the entire region, and ensuring a strong correlation between the model input data and the target scenario. It acquires precipitation-runoff, future upstream-downstream water level differences, and future tidal level changes within a preset future timeframe, comprehensively covering the core hydrological elements affecting scour and deposition. These data represent potential future influencing factors, enabling the model to extrapolate scour and deposition changes based on forward-looking information, improving the lead time and comprehensiveness of predictions, and meeting the needs of water conservancy decision-making for predicting future trends.

[0077] Then, based on future precipitation-runoff data, future upstream and downstream water level differences, and future tidal level changes, the future gate scheduling data corresponding to the target gate is determined, changing the traditional experience-based scheduling model. By quantitatively analyzing the impact of various factors on gate scheduling, dynamic and scientific scheduling decisions are achieved. For example, when heavy precipitation is predicted to cause a high water level difference, gate opening adjustments can be planned in advance to improve flood control and disaster relief capabilities and water resource allocation efficiency.

[0078] Next, the current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream and downstream water level differences, future tidal changes, and future gate scheduling data are input into a pre-defined scour and sedimentation change prediction model. This integrates historical topography, future environmental variables, and scheduling decision factors to form a multi-dimensional input system. Different data complement and verify each other. For example, the current topography determines the initial state of sediment, future precipitation-runoff data drives the change process, and gate scheduling regulates water flow and sediment transport, collectively improving the model's ability to simulate complex scour and sedimentation processes. The pre-defined scour and sedimentation change prediction model outputs the scour and sedimentation changes at the cross-section within a pre-defined timeframe, providing intuitive and quantitative prediction results for water conservancy project management. Managers can use the scour and sedimentation change trends to plan river dredging, dam reinforcement, and other engineering measures in advance, or optimize water resource allocation schemes. At the same time, the prediction results can verify the rationality of scheduling decisions, forming a virtuous cycle of "data-decision-prediction-feedback," enhancing the scientific nature and reliability of water conservancy system operation.

[0079] This embodiment provides a method for predicting sediment cross-sectional scour changes under the influence of a sluice gate, which can be used in the aforementioned electronic equipment. Figure 2 This is a flowchart of a method for predicting sediment cross-sectional scour changes under the influence of a sluice gate according to an embodiment of the present invention, as shown below. Figure 2 As shown, the process includes the following steps:

[0080] Step S201: Obtain the current scouring and sedimentation terrain sequence corresponding to the target gate at the current moment, and obtain the future precipitation-runoff data, future upstream and downstream water level difference and future tidal level change data within the future preset time period corresponding to the target gate.

[0081] Please refer to the above description of step S201 for details on this step, which will not be repeated here.

[0082] Step S202: Based on future precipitation-runoff data, future upstream and downstream water level differences, and future tidal level changes, determine the future gate scheduling data corresponding to the target gate.

[0083] Specifically, step S202 above may include the following steps:

[0084] Step S2021: Obtain historical precipitation-runoff data, historical upstream and downstream water level difference, historical tidal level change data, and historical gate scheduling data corresponding to historical precipitation-runoff data, historical upstream and downstream water level difference, and historical tidal level change data.

[0085] Specifically, the electronic device can receive historical precipitation-runoff data, historical upstream and downstream water level difference data, historical tidal level change data, and historical gate scheduling data corresponding to historical precipitation-runoff data, historical upstream and downstream water level difference data, and historical tidal level change data input by the user. It can also receive historical precipitation-runoff data, historical upstream and downstream water level difference data, historical tidal level change data, and historical gate scheduling data corresponding to historical precipitation-runoff data, historical upstream and downstream water level difference data, and historical tidal level change data sent by other devices. It can also search for historical precipitation-runoff data, historical upstream and downstream water level difference data, historical tidal level change data, and historical gate scheduling data corresponding to historical precipitation-runoff data, historical upstream and downstream water level difference data, and historical tidal level change data from the storage space.

[0086] This application does not specifically limit the method by which electronic devices acquire historical precipitation-runoff data, historical upstream and downstream water level differences, historical tidal level change data, and historical gate scheduling data corresponding to historical precipitation-runoff data, historical upstream and downstream water level differences, and historical tidal level change data.

[0087] Step S2022: Based on historical precipitation-runoff data, historical upstream and downstream water level differences, historical tidal level changes, and historical gate scheduling data, construct a causal graph structure.

[0088] Specifically, electronic devices employ a constraint-based approach, utilizing conditional independence tests (such as chi-square tests and Fisher'sz tests) and statistical relationships between variables in historical data to construct a preliminary causal network framework. For instance, it first determines whether precipitation and runoff are unconditionally independent. If not, it then tests whether they are independent under given other variables (such as soil moisture), gradually eliminating spurious associations and identifying potential causal edges. During the identification process, the algorithm identifies key thresholds through statistical data analysis. For example, when the historical distribution of precipitation intensity data exhibits obvious segmentation, the algorithm can automatically divide the threshold intervals and find that when precipitation intensity exceeds a certain threshold (such as 50 mm / h), the influence coefficient of runoff on water level difference increases significantly. This conditional causal relationship is then added to the causal graph, improving the graph structure.

[0089] In addition, a quarterly evaluation cycle is set, with major hydrological events such as typhoons and rainstorms serving as trigger points. Upon an event, a data collection process is immediately initiated, integrating various hydrological data from the event period and before and after it into a historical database. At the evaluation point, the causal discovery algorithm is re-run, merging and analyzing the new data with historical data. The current causal graph structure is compared with the results of the re-analysis to detect the existence of new causal paths. For example, after a rainstorm, a sudden increase in river sediment concentration may be observed, forming a new causal chain with precipitation intensity and runoff velocity. Then, a reward function is constructed using the actual gate scheduling effect as feedback. If the scheduling scheme guided by the current causal graph effectively reduces flood risk, the weight of the relevant causal edge is increased; if it leads to water waste or ecological problems, the weight of the corresponding edge is decreased. Through continuous iteration using reinforcement learning algorithms (such as Q-learning), the weights of the causal graph edges are made more consistent with the actual physical processes. For example, the weights of the causal edges between tide level and gate scheduling are dynamically adjusted to adapt to the differences in the impact of tide level changes on scheduling in different seasons, thereby generating the final causal graph structure.

[0090] Step S2023: Extend the cause-effect graph structure into a dynamic structural equation model to capture the time-delayed causal relationships between variables.

[0091] Specifically, the nodes in the causal graph structure (such as precipitation, runoff, water level difference, and gate scheduling) are transformed into variables in the dynamic structural equation model, the directed edges are transformed into causal paths between variables, and each variable is identified to determine endogenous variables (such as gate scheduling data and water level difference) and exogenous variables (such as precipitation and tidal changes).

[0092] For each endogenous variable, an autoregressive (AR) term and a moving average (MA) term are introduced. For example, for the target gate scheduling data yt, an AR(p) model is constructed: in For autoregressive coefficients, ∈ t This is a random error term; simultaneously, the MA(q) term is introduced to consider the impact of past errors on the current value, i.e. Capture the time-series dependencies of variables.

[0093] Then, the spatial correlation strength between monitoring points is determined by defining a spatial weight matrix W. For example, the weights are determined based on the reciprocal of the Euclidean distance, with closer monitoring points receiving higher weights. A spatial lag term is then introduced into the model for the upstream water level variable x. s,t (where s represents spatial location and t represents time), considering its spatial lag effect: Σ s W ss′ x s′，t , representing the weighted sum of water levels at all monitoring points upstream of the target gate. The time lag term is combined with the spatial lag term to form a spatiotemporal lag term, which is then incorporated into the structural equation model.

[0094] Integrating the aforementioned time delay and spatial location factors into the traditional structural equation model yields the extended dynamic structural equation model. For example, the prediction equation for the target gate scheduling data yt can be expressed as:

[0095]

[0096] Where, γ s and δ s,i For spatial location-related parameters, ω s It serves as the spatial lag coefficient, comprehensively characterizing the spatiotemporal causal relationship between variables.

[0097] Step S2024: Substitute future precipitation-runoff data, future upstream and downstream water level differences, and future tidal level change data into the time-delay causal relationship to determine the future gate scheduling data corresponding to the target gate.

[0098] Specifically, electronic devices can determine the future gate scheduling data corresponding to the target gate by substituting future precipitation-runoff data, future upstream and downstream water level differences, and future tidal level change data into time-delay causal relationships based on the set multi-objective function and constraints.

[0099] Specifically, with reducing flood risk as the core objective, the goal is quantified into indicators such as minimizing the flood-inundated area and minimizing the duration of water levels exceeding warning levels. For example, let f1 represent the flood-inundated area, and simulate the inundation range under different gate scheduling schemes using a hydraulic model, incorporating this into the objective function. For instance, In this formula, the first part sums the flood-inundated areas at each time step, while the second part, by determining whether the actual water level exceeds the warning level, performs a weighted summation based on the duration of the excess, comprehensively reflecting the degree of flood risk. Aiming to maximize the effective use of water resources, indicators include agricultural irrigation water supply satisfaction rate and industrial water supply. For example, f2 represents the total water supply in the region, calculated by statistically analyzing the actual water withdrawal of each water-using sector under different dispatch schemes. This formula calculates the sum of the ratios of actual agricultural and industrial water supply to target supply at each time step. A higher ratio indicates higher water resource utilization efficiency. It focuses on maintaining river ecological flow and protecting biological habitats, aiming to minimize the deviation of ecological flow from the ideal value. For example, a river ecological baseflow threshold is set, and f3 represents the total deviation between the actual flow and the ecological baseflow threshold.

[0100] The constraints include: Physical constraints: considering the physical limitations of the gate itself, such as the gate opening range (0≤ui≤umax, where ui is the gate opening at the i-th time step), and the extreme water levels upstream and downstream. Water balance constraints: based on the principle of water conservation within the basin, an equation is established to ensure the balance of inflow and outflow in each time period, such as It-Ot=ΔSt, where It is the inflow, Ot is the outflow, and ΔSt is the change in water storage during the time period. Water quality constraints: to meet the needs of water pollution control, the concentration of pollutants discharged is limited to not exceeding environmental standards, such as Cpollutant≤Climit.

[0101] Then, each particle is considered as a potential gate scheduling scheme. The particle's position vector corresponds to the gate opening parameter at different time steps, and the velocity vector determines the update direction and step size of the position. During initialization, the positions and velocities of particles in the feasible solution space are randomly distributed.

[0102] The multi-objective function is transformed into a fitness function, which integrates multiple objectives through a weighted summation, such as Fitness = w1f1 + w2f2 + w3f3, where w1, w2, and w3 are the weights of each objective, and w1 + w2 + w3 = 1. The weights can be determined using the analytic hierarchy process (AHP) or expert scoring.

[0103] For solutions that do not meet the constraints, a penalty term is set to reduce their fitness. For example, for each violation of a constraint, a certain penalty score is subtracted from the fitness value.

[0104] The particle velocity and position are updated based on the individual best position (pbest) and the global best position (gbest). Each particle learns from its own historical best solution and the global best solution, continuously adjusting its position in the solution space, gradually approaching the optimal scheduling scheme, and obtaining the future gate scheduling data corresponding to the target gate.

[0105] Step S203: Input the current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the preset scour change prediction model, and output the scour change of the sediment cross section corresponding to the target gate within the future preset time period.

[0106] Specifically, the preset scour change prediction model includes a scour and sedimentation trend classification model and a sediment scour change prediction model; the above step S203 may include the following steps:

[0107] Step S2031: Input the current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the scour and sedimentation trend classification model to determine the sediment evolution trend category corresponding to the target gate.

[0108] Specifically, step S2031 above may include the following steps:

[0109] Step a1: Extract features from the current alluvial-deposition terrain sequence to obtain the terrain data features corresponding to the current alluvial-deposition terrain sequence.

[0110] Specifically, the pre-defined alluvial change prediction model can calculate the average elevation, standard deviation of elevation, maximum elevation, and minimum elevation corresponding to the current alluvial-deposition terrain sequence. Then, it calculates the slope and aspect by differencing the average elevation, standard deviation of elevation, maximum elevation, and minimum elevation. The pre-defined alluvial change prediction model can also calculate the curvature characteristics (including planar curvature and profile curvature), surface roughness, and fractal dimension corresponding to the current alluvial-deposition terrain sequence. Roughness can be calculated based on the rate of change of elevation data, such as by summing the absolute values ​​of the elevation differences between adjacent points. Fractal dimension: The fractal dimension can be used to measure the complexity of the terrain. It reflects the self-similarity of the terrain at different scales and is usually calculated using methods such as box dimension. The larger the fractal dimension, the more complex the terrain.

[0111] Furthermore, the pre-defined scour and sedimentation change prediction model can calculate the changes in scour and sedimentation topography at adjacent time steps, thus obtaining the scour and sedimentation rate. By analyzing the magnitude and distribution of the scour and sedimentation rate, the speed and trend of topographic change can be understood, and active and stable scour and sedimentation areas can be identified. The areas of regions with different degrees of scour and sedimentation can be statistically analyzed, such as dividing the scour and sedimentation topography into sedimentation zones, erosion zones, and stable zones, and calculating the area of ​​each zone and its proportion of the total area to understand the spatial distribution characteristics of scour and sedimentation.

[0112] Finally, the feature values ​​such as elevation statistics, slope and aspect, curvature, roughness, fractal dimension, scour and deposition rate, and scour and deposition area statistics are arranged in a certain order to form a multi-dimensional feature vector, which serves as the topographic data feature corresponding to the current scour and deposition topographic sequence.

[0113] Step a2 involves fusing future precipitation-runoff data, future upstream-downstream water level differences, future tidal level changes, and future gate scheduling data to generate target fusion features.

[0114] Specifically, step a2 above may include the following steps:

[0115] Step a21: Based on the preset adaptive phase space reconstruction algorithm, analyze the chaotic characteristics corresponding to future precipitation-runoff data, future upstream and downstream water level differences, future tidal level change data, and future gate scheduling data.

[0116] Among them, chaotic features include the maximum Lyapunov exponent, fractal dimension, and chaotic frequency spectrum.

[0117] Specifically, for future precipitation-runoff data, future upstream and downstream water level differences, future tidal level changes, and future gate scheduling data, a pre-defined adaptive phase space reconstruction algorithm is used to construct the phase space based on Takens' theorem.

[0118] Specifically, by selecting an appropriate embedding dimension *m* and time delay *τ*, the time series data is mapped into a phase space. For example, for a precipitation-runoff data series {x(t)}, the reconstructed phase space vector is X(t) = [x(t), x(t+τ), ..., x(t+(m-1)τ)]. An adaptive algorithm dynamically adjusts *m* and *τ* to adapt to the characteristics and variations of different data.

[0119] In the reconstructed phase space, the maximum Lyapunov exponent (λmax) is calculated. The maximum Lyapunov exponent measures the separation rate of adjacent orbits in the phase space, reflecting the degree of chaos in the system. Optionally, the maximum Lyapunov exponent can be obtained by tracking the evolution of orbits in the phase space and calculating the distance growth rate between orbits. When λmax > 0, the system exhibits chaotic characteristics; the larger λmax is, the higher the degree of chaos.

[0120] Fractal dimension is used to describe the complexity of attractors in phase space. Specifically, it can be calculated using the box dimension method, dividing the phase space into boxes of different scales, counting the number of boxes N(∈) containing attractors, and then analyzing N(∈) at different scales to obtain the fractal dimension D. The larger the fractal dimension D, the more complex the structure of the attractors and the more pronounced the chaotic characteristics of the system.

[0121] Finally, spectral analysis was performed on future precipitation-runoff data, future upstream-downstream water level differences, future tidal level changes, and future gate scheduling data. For example, the time series data was converted to the frequency domain using Fast Fourier Transform (FFT) to obtain a chaotic frequency spectrum. The chaotic frequency spectrum shows the energy distribution of the system at different frequencies. By analyzing the characteristics of the frequency spectrum, such as peak positions and bandwidth, we can further understand the chaotic behavior and dynamic characteristics of the system.

[0122] Chaotic features are generated based on the maximum Lyapunov exponent, fractal dimension, and chaotic frequency spectrum.

[0123] Step a22: Input the chaotic features corresponding to future precipitation-runoff data, future upstream and downstream water level differences, future tidal level changes, and future gate scheduling data into the reinforcement learning model in the scour and sedimentation trend classification model, and determine the target fusion strategy corresponding to each chaotic feature.

[0124] Specifically, the electronic device inputs the chaotic features corresponding to future precipitation-runoff data, future upstream-downstream water level differences, future tidal level changes, and future gate scheduling data into the reinforcement learning model of the scour and sedimentation trend classification model. The reinforcement learning model constructs a state space and an action space based on the input data. The action space includes various possible chaotic feature fusion methods. Taking weighted fusion as an example, an action can be defined as an operation that assigns weights to different chaotic features. For example, action 1 = [0.3, 0.4, 0.3] means assigning weights of 0.3, 0.4, and 0.3 to the maximum Lyapunov exponent, fractal dimension, and chaotic frequency spectrum, respectively. It also includes the selection and parameter settings of other fusion algorithms such as principal component analysis (PCA) fusion and independent component analysis (ICA) fusion. Each time, the reinforcement learning model selects an action from the action space and calculates the reward function corresponding to that action. The reward function can aim to improve the accuracy of scour and sedimentation trend prediction while taking into account the effectiveness and stability of the fused features. For example, if the features fused from the model are used in a scour and sedimentation trend prediction model, a positive reward is given when the prediction result has a small error compared to the actual scour and sedimentation situation; conversely, a negative reward is given when the prediction error is large. For instance, multiple indicators can be used to quantify the reward. For example, based on the mean squared error (MSE), assuming the predicted scour and sedimentation volume is... The actual scouring and silting volume is yi, and there are n data points in a single prediction. The reward value R can be set as R = 1 / 1 + MSE, so that the smaller the MSE, the greater the reward. In addition, other indicators can be combined, such as the accuracy of the predicted trend direction (a reward is given if the predicted direction of siltation or scouring is consistent with the actual direction) and the variance of the fused features (a reward is given if the variance is small, indicating high feature stability). The reward is calculated comprehensively.

[0125] Then, based on the reward system corresponding to each action, each action is updated according to the policy function π(a|s). Here, the policy function π(a|s) represents the probability of taking action a in state s. Common policy representation methods include parameter-based policy networks, such as using neural networks to construct the policy function, with the input being the state (chaotic characteristics) and the output being the probability distribution of each action. Gradient ascent algorithms (such as the policy gradient algorithm) are used to adjust the parameters of the policy network based on reward feedback, increasing the probability of selecting actions that yield high rewards. For example, in a certain state, if taking a fusion action with specific weight allocation yields a high reward, the policy network will adjust its parameters to increase the probability of that action being selected in similar states.

[0126] Ultimately, the action with the highest reward value was determined as the target fusion strategy corresponding to each chaotic feature.

[0127] Step a23: Based on the target fusion strategy, fuse the chaotic features to generate target fusion features.

[0128] Specifically, based on a target fusion strategy, various chaotic features are fused to generate target fused features. For example, if the target fusion strategy is weighted fusion, features such as the maximum Lyapunov exponent, fractal dimension, and chaotic frequency spectrum are weighted and summed according to the weights of different chaotic features to obtain the fused feature vector.

[0129] Step a3: Input the terrain data features and target fusion features into the scour and sedimentation trend classification model to determine the sediment evolution trend category corresponding to the target gate.

[0130] Specifically, step a3 above may include the following steps:

[0131] Step a31: Input the terrain data features and target fusion features into the quantum decision tree in the scour and sedimentation trend classification model.

[0132] In this quantum decision tree, each decision node is an independent quantum system.

[0133] Specifically, topographic data features and generated target fusion features are input into the quantum decision tree of the scour and sedimentation trend classification model. Since each decision node in the quantum decision tree is an independent quantum system, these features, as input data, provide the quantum decision tree with the basic information for analyzing scour and sedimentation trends. For example, topographic data features may contain information such as slope and elevation, and target fusion features may incorporate chaotic features and other information, which together provide the basis for the decision-making process of the decision tree.

[0134] Step a32: Initialize the quantum state of each decision node in the quantum decision tree.

[0135] Specifically, each decision node in the quantum decision tree undergoes quantum state initialization. A quantum state is a state description of a quantum system, and the initialization process sets the initial conditions for subsequent quantum computation. For example, in quantum computing, the initial state of a qubit may be set to a superposition state, that is, simultaneously existing in a superposition of multiple states, providing rich possibilities for quantum computing. Through initialization, the quantum system of each decision node enters a state where computation and decision-making can be performed.

[0136] Step a33: Encode the input terrain data features and target fusion features using qubits to generate the current quantum state.

[0137] In this context, each sub-feature in the target fusion feature corresponds to one or more qubits; a qubit can represent multiple states simultaneously.

[0138] Specifically, qubits are used to encode the input terrain data features and target fusion features to generate the current quantum state.

[0139] For example, for each terrain data feature and target fusion feature, it is encoded into qubits using quantum gate operations based on their mapping relationship. For instance, for elevation features, a Hadamard gate (H-gate) is used to encode it into a superposition state of the qubits. When the H-gate operates on the |0> state, it transforms it into... Quantum encoding of elevation features is achieved by adjusting the parameters and action order of H-gates. After encoding individual features, the interrelationships between features are considered, and quantum gate operations such as controlled NOT gates (CNOT gates) are used to entangle the qubits of different features. For example, the qubits of slope and flow characteristics are entangled through CNOT gates, making the quantum states of the two features correlated. When the qubit of the slope feature is in the |0> state, the qubit of the flow characteristic undergoes a corresponding state change according to the action of the CNOT gate, thereby achieving quantum encoding of feature combinations.

[0140] Step a34: Perform a quantum gate operation on the current quantum state to generate the node quantum state corresponding to each decision node.

[0141] Specifically, step a34 above may include the following steps:

[0142] Step a341: Identify the current quantum state, determine the real-time dynamic characteristics corresponding to the current quantum state, as well as the core and auxiliary characteristics in the current quantum state.

[0143] Specifically, by performing measurement operations on quantum states, such as using projection measurement M = {P0, P1} (P0 = |0> <0>, P1 = |1> <1>), real-time dynamic characteristics are inferred based on the probability distribution of the measurement results. For example, if the probability of obtaining the |0> state changes significantly in a short period of time, real-time dynamic information such as water flow velocity and precipitation intensity can be determined by combining background knowledge of terrain data and target fusion features.

[0144] Then, information-theoretic metrics such as quantum mutual information I(ρ) are used. AB )=S(ρ A )+S(ρ B )-S(ρ AB )(ρ A ρ B Let S(ρ) be the density matrix of subsystems A and B, respectively, and S(ρ) be the von Neumann entropy. The contribution of different features to the quantum state is evaluated. For example, in terrain data features and target fusion features, the quantum mutual information between each feature and the overall quantum state is calculated. Features with quantum mutual information greater than a preset threshold are considered core features, as they have a more significant impact on the quantum state. Features with quantum mutual information less than or equal to the preset threshold are considered auxiliary features.

[0145] Step a342: Based on real-time dynamic characteristics, determine the target quantum gate combination corresponding to the current quantum state.

[0146] Specifically, electronic devices can select quantum gates based on real-time dynamic characteristics (such as the rate of change of water flow velocity, the frequency of water level fluctuations, etc.) through predefined mapping rules. For example, for features with drastic fluctuations, the Hadamard gate (H gate) is selected for state superposition enhancement; for features that require establishing correlations, the controlled NOT gate (CNOT) or Tofoli gate is selected; and for phase-sensitive features, the phase gate (P gate) or T gate is selected.

[0147] Then, the core and auxiliary features in the current quantum state are used to generate a gate sequence. For example, if core features A and B are highly correlated, a CNOT(A,B) operation is generated; if it is necessary to enhance the volatility of feature C, an H-CNOT-H sequence is generated.

[0148] Finally, quantum compilation techniques are used to optimize the gate sequence, reducing the number and depth of gates to generate the target quantum gate combination. For example, adjacent H gates are canceled out, and SWAP gates are used to adjust the positions of qubits to reduce long-distance interactions.

[0149] Step a343: Based on the control NOT gate in the target quantum gate combination, perform entanglement calculation on the core features to generate higher-order entangled features.

[0150] Specifically, electronic devices select qubits corresponding to core features (such as flow rate, slope, etc.) from real-time dynamic features.

[0151] Then, entangled chains or clusters are constructed using the controlled NOT gate (i.e., CNOT gate) in the target quantum gate combination. For example: CNOT(qubit1,qubit2); CNOT(qubit1,qubit3); H(qubit1) to construct the GHZ state; and the W state is constructed through sequential CNOT and single-bit rotation operations.

[0152] Finally, through entanglement operations, higher-order correlations between core features are encoded into quantum states, generating higher-order entangled features. For example, the second-order correlation between flow rate change and slope can be extracted by measuring the degree of entanglement after CNOT(flow rate, slope).

[0153] Step a344: Based on the control NOT gate in the target quantum gate combination, the auxiliary features and core features are entangled to form a star-shaped topology.

[0154] Specifically, each core feature is used as the center, and all related auxiliary features are connected through control NOT gates (CNOT gates). For example, with core feature A (flow rate) as the center, auxiliary features B (temperature) and C (wind direction) are connected: CNOT(A,B); CNOT(A,C).

[0155] Then, by adjusting the entanglement strength (e.g., using partial CNOT gates), the information transmission efficiency is optimized to form a star topology. For example, using square root CNOT gates. Achieve partial entanglement.

[0156] Step a345: Perform a nonlinear transformation on the current quantum state to generate the nonlinear characteristics corresponding to the current quantum state.

[0157] Specifically, electronic devices can achieve nonlinear transformations using quantum phase estimation (QPE) or quantum adiabatic evolution. For example, phase nonlinearity can be achieved by applying phase gates (P(θ)) multiple times; nonlinear transformation circuits can be constructed using controlled phase gates (CU).

[0158] Then, the linear feature is mapped to a nonlinear space. For example, a linear feature x is mapped to sin(x) or x through a quantum circuit. 2 .

[0159] Finally, nonlinear features are extracted through quantum measurements. For example, nonlinear information can be extracted by measuring the projection of quantum states onto different bases.

[0160] Step a346: Based on the high-order entanglement features, star topology, and nonlinear features, generate the node quantum states corresponding to each decision node.

[0161] Specifically, the electronic device integrates the higher-order entangled features, star topology, and nonlinear features obtained from the above processing through quantum gate operations. For example, it can use multiple control gates (such as MCX gates) to combine different features, or apply quantum Fourier transform (QFT) to perform feature space transformation.

[0162] Then, the integrated features are mapped to the node quantum state |ψ_node> through unitary transformation U: |ψ_node>=U|ψ_initial>.

[0163] Finally, the generated nodal quantum states are verified using quantum state tomography or fidelity measurements. For example, the fidelity between the generated state and the target state is calculated as F = |<ψ_target|ψ_node>|. 2 .

[0164] Step a35: Calculate the quantum probability amplitude of the node quantum state corresponding to each decision node.

[0165] Specifically, for the node quantum state |ψ>=∑_iα_i|i>, α_i is the quantum probability amplitude.

[0166] For small systems containing fewer than a certain number of qubits, electronic devices directly calculate the probability amplitude through matrix operations. For example, if |ψ>=U|0>, then α_i=<i|U|0> .

[0167] For large systems containing more than or equal to a certain number of qubits, the probability amplitude is estimated through multiple measurements. For example, by performing N measurements on |ψ>, the frequency f_i of state |i> can be statistically obtained as f_i≈|α_i|. 2 .

[0168] The quantity threshold can be 18, 15, or other values. This embodiment does not specifically limit the quantity threshold.

[0169] Step a36: Determine the classification path based on each quantum probability amplitude.

[0170] Specifically, for each possible decision path, its cumulative probability is calculated. For example, the probability of path P = {node 1 → node 2 → node 3} is P = |α_1|^2 · |α_2|^2 · |α_3|^2.

[0171] Then, the cumulative probabilities of each possible decision path are compared, and the decision path with the highest cumulative probability is selected as the classification path. For example: P_max = max(P_1, P_2, ..., P_n).

[0172] Optionally, the maximum cumulative probability can be compared with a probability threshold. If the maximum cumulative probability is lower than the probability threshold, recalculation or fusion of multiple paths can be triggered. For example, if P_max < 0.6, then a weighted fusion of the first k paths can be considered.

[0173] Step a37: Based on each classification path, proceed to the next level decision node until a leaf node is reached.

[0174] Specifically, based on the selected classification path, the quantum operations of each node are executed sequentially. After passing through a node, the quantum state is updated to the output state of that node, and the computation stops when a leaf node (without child nodes) is reached.

[0175] In step a38, the leaf nodes determine the sediment evolution trend category corresponding to the target gate based on the previously calculated quantum probability amplitude and path information.

[0176] Specifically, the leaf nodes determine the sediment evolution trend category corresponding to the target gate based on the previously calculated quantum probability amplitude and path information.

[0177] For example, a predefined mapping from leaf node states to sediment evolution trend categories. For instance: |00>→Sedimentation trend; |01>→Scrap trend; |10>→Equilibrium state; |11>→Complex evolution.

[0178] Calculate the probability of each category based on the probability amplitude of the leaf nodes. For example: P(siltation) = |α_00|²; P(scour) = |α_01|². Select the category with the highest probability as the final result. For example: if P(siltation) > P(scour), then it is determined to be a siltation trend.

[0179] Step S2032: Input the sediment evolution trend category, current scour and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the preset sediment scour change prediction model, and output the sediment cross-sectional scour change corresponding to the target gate within the future preset time period.

[0180] Specifically, step S2032 above may include the following steps:

[0181] Step b1: Input the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the time dependency network of the preset sediment erosion change prediction model, and output the time dependency features.

[0182] Specifically, step b1 above may include the following steps:

[0183] Step b11: Input the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the time dependency network of the preset sediment erosion change prediction model.

[0184] Specifically, the electronic device standardizes multi-source heterogeneous data, including sediment evolution trend categories, current alluvial-depositional topography sequences, and future precipitation-runoff data, eliminating dimensional differences. For sediment evolution trend categories, one-hot encoding or embedded vector representation is used; the current alluvial-depositional topography sequence is converted into feature vectors using point cloud data processing technology; and future precipitation-runoff time series data are normalized. Subsequently, spatiotemporal alignment technology is used to accurately match different data along the time dimension, forming structured multimodal input data, which is then input into a pre-defined time-dependent network.

[0185] Step b12: The weights in the time-dependent network are determined. At each time step, the input data is evaluated to determine the importance of the input data at each time step to the current scour and deposition terrain sequence and the changes in sediment cross-section scour within a preset time period.

[0186] Specifically, structured multimodal input data is combined into an input vector for each time step. Assuming there are n data types, m values ​​are extracted from each data type. iIf there are 1, 2, ..., n features, then the input vector xt at the t-th time step is: xt = [f1, t, f2, t, ..., fn, t]; where fi, t represents the feature vector extracted by the i-th data type at the t-th time step.

[0187] To enable the weight determination network to perceive temporal information, a time step identifier is introduced. Positional encoding can be used to encode the time step t into a vector pt, which is then concatenated with the input vector xt to obtain the final input representation Xt = [xt; pt]. This allows the network to distinguish data from different time steps and capture the temporal dependencies of the data.

[0188] Then, the input representation Xt is fed into a multi-head attention mechanism. The multi-head attention mechanism captures the relationships between data from different perspectives through multiple different "heads." First, Xt is projected onto three spaces: query, key, and value, respectively, to obtain Q. t K t and V t Then, for each head h, calculate the attention score:

[0189]

[0190] Where, d k It is the dimension of the key vector, Q t h K t h and V t h These are the query, key, and value vectors corresponding to the h-th head. Finally, the results of all heads are concatenated and projected to obtain the attention output At.

[0191] Finally, the attention output At is processed to calculate a preliminary importance score for the input data at each time step. At can be mapped to a scalar value through a linear layer, and then the score is normalized to the [0,1] interval through an activation function (such as the Sigmoid function) to obtain a preliminary importance score αt, which reflects the relative importance of each time step data in the overall data.

[0192] Step b13: Determine the weight information corresponding to the input data at each time step based on the importance of the input data at each time step.

[0193] The electronic device determines the weight information corresponding to the input data at each time step based on the calculated relationship between importance and weight information.

[0194] Step b14: Dynamically adjust the update gate and forget gate in the time-dependent network based on the weight information corresponding to the input data at each time step.

[0195] Specifically, the parameters of the update gate and forget gate in the time-dependent network are dynamically adjusted based on the weight information corresponding to the input data at each time step.

[0196] Specifically, if the weight information is greater than the weight threshold, the opening degree of the update gate is increased, allowing it to participate more in the state updates of the time-dependent network; if the weight information is less than or equal to the weight threshold, the closing degree of the forget gate is increased, reducing its impact on network memory. The Q-learning algorithm from reinforcement learning is introduced to continuously optimize the gating policy and find the gating parameter combination that minimizes the prediction error.

[0197] Step b15: Based on the adjusted update gate and forget gate at each time step, extract time series features from the input data and output time-dependent features.

[0198] Specifically, time-series features are extracted from the input data through adjusted update and forget gates. Optionally, a combination of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) is used, with CNNs extracting local features and RNNs capturing long-term dependencies in the time series. A self-attention mechanism is then used to weight and fuse the extracted features, highlighting key features and outputting time-dependent features.

[0199] Step b2 involves inputting the sediment evolution trend category, current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data into the pre-set sediment erosion change prediction model's structural feature extraction network, and outputting topological features.

[0200] Specifically, step b2 above may include the following steps:

[0201] Step b21: Based on the current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data, construct the target map structure.

[0202] Specifically, electronic devices can convert various types of data into an initial graph structure G = (V, E), where: Nodes V: contain hydrological monitoring points (such as upstream and downstream water level stations, gate locations), topographic feature points (key points of riverbed cross-sections), meteorological monitoring stations, etc., and each node contains a multi-dimensional feature vector (such as water level, flow rate, and topographic elevation). Edges E: represent the physical relationships between nodes, constructed in three ways: Spatial correlation edges: constructed based on geographical distance, such as the weight of the edge connecting adjacent water level stations being inversely proportional to the distance. Temporally dependent edges: directed edges are constructed between nodes at different time steps of the same monitoring point, forming a temporal chain. Physically causal edges: constructed based on hydrodynamic principles, such as downstream water level being affected by upstream flow rate and gate scheduling, with the direction and weight of the edges determined by causal inference algorithms (such as the PC algorithm).

[0203] Then, based on real-time hydrological data, edge weights are adjusted using an attention mechanism to generate the target graph structure. For example, edge weights between upstream and downstream water level stations are automatically enhanced during flood season. Historical information is aggregated using gated recurrent units (GRUs) to ensure that node features incorporate temporal evolution information.

[0204]

[0205] Where ht is the hidden state of node u at time t, and N(u) is the set of neighboring nodes.

[0206] Step b22: Extract features from the graph structure based on the structural feature extraction network and output the topological structure features.

[0207] Specifically, spatial dependencies between network nodes are extracted based on structural features:

[0208] in Add self-loops to the adjacency matrix. W is the degree matrix. (l) These are learnable weights. Then, a one-dimensional convolution and attention mechanism are combined to capture temporal features: O t =Attention(H t H t-1 H t-k ), where k is the size of the time window, and the attention mechanism automatically assigns weights to different time steps.

[0209] By fusing spatial dependencies and temporal features, topological structure features are output.

[0210] Step b3: Fuse the time-dependent features and topological features to output the fused prediction features.

[0211] Specifically, time-dependent features and topological features are mapped to the same dimensional space through linear transformation: FT′=WTFT, FG′=WGFG, where FT is the time-dependent feature and FG is the topological feature.

[0212] Then, the attention scores of time-dependent features to topological features are calculated: Where [;] represents a concatenation operation, W a These are learnable weights.

[0213] Then, based on the attention score of the temporal dependency features to the topological structure features, the transformed temporal dependency features and topological structure features are fused to output the fused prediction feature. The specific formula is as follows:

[0214]

[0215] in This represents element-wise multiplication.

[0216] Step b4: Based on the fusion prediction features, output the sediment cross-sectional scour change within a preset time period corresponding to the target gate.

[0217] Specifically, the preset sediment scour change prediction model can take into account the target physical constraints and, based on the integrated prediction characteristics, output the sediment cross-sectional scour change within a preset time period corresponding to the target gate.

[0218] The target physical constraints include minimizing the mean square error between the predicted and observed values. Among them, y pred For the predicted value, y obrs These are observations from the current alluvial and sedimentary topographic sequence; ensuring smooth predicted changes between adjacent time steps: Design constraints based on sediment transport equations: Where S is the sediment concentration, q is the water flow velocity, and D and E are the siltation and scouring rates, respectively.

[0219] The method for predicting sediment scour and deposition changes under the influence of sluice gates provided in this application acquires historical precipitation-runoff data, historical upstream and downstream water level differences, historical tidal level changes, and historical gate scheduling data corresponding to these data, comprehensively covering the key factors affecting sediment scour and deposition. This provides a rich and accurate data foundation for subsequent analysis, avoiding analytical biases caused by missing data. Based on historical precipitation-runoff data, historical upstream and downstream water level differences, historical tidal level changes, and historical gate scheduling data, a causal graph structure is constructed, which clearly and intuitively presents the causal relationships between various factors, helping to understand the internal mechanisms of sediment scour and deposition, making the model more interpretable, and facilitating verification and optimization of the model logic by water conservancy experts. Extending the causal graph into a dynamic structural equation model captures the time-delay causal relationships between variables, taking into account the time delay effect in the hydrological process, and improving the accuracy and reliability of the model's predictions. By substituting future precipitation-runoff data, future upstream-downstream water level differences, and future tidal level changes into the time-delay causal relationship, the future gate scheduling data corresponding to the target gate is determined, thus ensuring the accuracy of the determined future gate scheduling data.

[0220] Then, the current erosion and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level differences, future tidal level changes, and future gate scheduling data are input into the erosion and deposition trend classification model. Feature extraction is performed on the current erosion and deposition topography sequence to obtain the corresponding topographic data features, ensuring the accuracy of the obtained topographic data features. A pre-defined adaptive phase space reconstruction algorithm is used to analyze the chaotic features of future data, which can uncover complex nonlinear variation patterns in the hydrological system and capture subtle trends that are difficult to detect using traditional methods, providing more in-depth information for prediction. The chaotic features are input into a reinforcement learning model to determine the target fusion strategy, avoiding the subjectivity and limitations of manually setting fusion rules, making the fused features more representative and effective. Target fusion features are generated based on the target fusion strategy, integrating multi-faceted chaotic information, enhancing the feature's ability to express sediment evolution trends, and providing higher-quality input for subsequent classification and prediction.

[0221] Next, topographic data features and target fusion features are input into the quantum decision tree in the sedimentation trend classification model. The introduction of the quantum decision tree leverages the superposition and entanglement properties of quantum systems. Compared to traditional decision trees, it can handle multiple possibilities simultaneously, significantly improving computational efficiency and the ability to process complex data, making it particularly suitable for highly complex scenarios such as hydrological systems. From quantum state initialization and feature encoding to quantum gate operations, the input features are deeply processed. By generating high-order entangled features, constructing star-shaped topologies and nonlinear features, the complex relationships between features are fully explored, further improving classification accuracy and sensitivity to subtle changes. The quantum probability amplitude is calculated and the classification path is determined, identifying the sediment evolution trend category in probabilistic form. This not only provides classification results but also information on the reliability of the results, offering a more comprehensive reference for decision-making.

[0222] After determining the sediment evolution trend category corresponding to the target gate, the sediment evolution trend category, the current scour and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level difference, future tidal level change data, and future gate scheduling data will be input into the time dependency network of the preset sediment scour and deposition change prediction model. This will comprehensively consider the influence of multiple factors on sediment scour and deposition, avoid the one-sidedness of single-factor analysis, and make the prediction more in line with the actual situation.

[0223] The weighting network allocates weights based on data importance, adaptively highlighting the role of key factors at different time steps. For example, during flood season, the weight of precipitation-runoff data is increased, improving the model's adaptability to different hydrological conditions and its prediction accuracy. By dynamically adjusting the update and forget gates, the network effectively controls the inflow and outflow of information, extracting more valuable time-series features, enhancing the model's ability to capture time dependencies in the data, and improving prediction accuracy and stability. Based on the adjusted update and forget gates at each time step, time-series features are extracted from the input data, outputting time-dependent features, ensuring the accuracy of the output time-dependent features.

[0224] Next, based on the current scour and deposition topography sequence, future precipitation-runoff data, future upstream and downstream water level differences, future tidal changes, and future gate scheduling data, a target graph structure is constructed. This better reflects the topological structure and flow propagation patterns of the river system, providing richer structural information for prediction. Topological features are extracted from the network output using structural features, deeply mining hidden information within the graph structure and complementing the time-dependent features. Finally, the time-dependent features and topological features are fused to output a fused prediction feature. Based on the fused prediction feature, the scour and deposition changes of the sediment cross-section corresponding to the target gate within a preset future timeframe are output, achieving an organic combination of spatiotemporal information. This allows the model to comprehensively consider the impact of temporal evolution and spatial structure on sediment scour and deposition, ultimately outputting more accurate predictions of future sediment cross-section scour and deposition changes, providing a scientific and reliable basis for water conservancy project decisions and flood control scheduling.

[0225] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for predicting scour changes in sediment cross-sections under the influence of sluice gates, characterized in that, The method includes: Obtain the current erosion and deposition terrain sequence corresponding to the target gate at the current moment, and obtain the future precipitation-runoff data, future upstream and downstream water level difference and future tidal level change data within the future preset time period corresponding to the target gate; Based on the future precipitation-runoff data, the future upstream and downstream water level difference, and the future tidal level change data, determine the future gate scheduling data corresponding to the target gate; The current scour and sedimentation topography sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data are input into the preset scour change prediction model, and the scour change of the sediment cross section corresponding to the target gate within the future preset time period is output. The preset scour change prediction model includes a scour and sedimentation trend classification model and a sediment scour change prediction model. The step of inputting the current scour and sedimentation topography sequence, future precipitation-runoff data, future upstream-downstream water level difference, future tidal level change data, and future gate scheduling data into the preset scour change prediction model, and outputting the sediment cross-sectional scour change corresponding to the target gate within the future preset time period, includes: The current erosion and deposition topography sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data are input into the erosion and deposition trend classification model to determine the sediment evolution trend category corresponding to the target gate. The sediment evolution trend category, the current scour and deposition topography sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data are input into the preset sediment scour change prediction model, and the model outputs the sediment cross-sectional scour change corresponding to the target gate within the future preset time period.

2. The method according to claim 1, characterized in that, The process of inputting the current scour and sedimentation topography sequence, the future precipitation-runoff data, the future upstream-downstream water level difference, the future tidal level change data, and the future gate scheduling data into the scour and sedimentation trend classification model to determine the sediment evolution trend category corresponding to the target gate includes: Feature extraction is performed on the current erosion and deposition terrain sequence to obtain the terrain data features corresponding to the current erosion and deposition terrain sequence; The future precipitation-runoff data, the future upstream-downstream water level difference, the future tidal level change data, and the future gate scheduling data are fused together to generate target fusion features; The terrain data features and the target fusion features are input into the scour and sedimentation trend classification model to determine the sediment evolution trend category corresponding to the target gate.

3. The method according to claim 2, characterized in that, The process of fusing the future precipitation-runoff data, the future upstream-downstream water level difference, the future tidal level change data, and the future gate scheduling data to generate target fusion features includes: Based on a preset adaptive phase space reconstruction algorithm, the chaotic features corresponding to the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data are analyzed respectively; the chaotic features include the maximum Lyapunov exponent, fractal dimension, and chaotic frequency spectrum. The chaotic features corresponding to the future precipitation-runoff data, the future upstream-downstream water level difference, the future tidal level change data, and the future gate scheduling data are respectively input into the reinforcement learning model in the scour and sedimentation trend classification model to determine the target fusion strategy corresponding to each chaotic feature. Based on the target fusion strategy, the chaotic features are fused to generate target fusion features.

4. The method according to claim 2, characterized in that, The terrain data features and the target fusion features are input into the scour and sedimentation trend classification model to determine the sediment evolution trend category corresponding to the target gate, including: The terrain data features and the target fusion features are input into the quantum decision tree in the scour and siltation trend classification model; each decision node in the quantum decision tree is an independent quantum system. Quantum state initialization is performed on each decision node in the quantum decision tree; The input terrain data features and the target fusion features are encoded using qubits to generate the current quantum state; each sub-feature in the target fusion features corresponds to one or more qubits; one qubit can represent multiple states simultaneously. Perform quantum gate operations on the current quantum state to generate the node quantum state corresponding to each decision node; Calculate the quantum probability amplitude of the node quantum state corresponding to each decision node; Based on the quantum probability amplitudes described above, the classification path is determined; According to each of the classification paths, proceed to the next level decision node until the leaf node is reached; The leaf nodes determine the sediment evolution trend category corresponding to the target gate based on the previously calculated quantum probability amplitude and path information.

5. The method according to claim 4, characterized in that, The step of performing quantum gate operations on the current quantum state to generate the node quantum states corresponding to each decision node includes: The current quantum state is identified to determine the real-time dynamic characteristics corresponding to the current quantum state, as well as the core and auxiliary characteristics in the current quantum state; Based on the real-time dynamic characteristics, the target quantum gate combination corresponding to the current quantum state is determined; Entanglement calculations are performed on the core features based on the controlled NOT gates in the target quantum gate combination to generate higher-order entangled features. Based on the controlled NOT gate in the target quantum gate combination, the auxiliary feature is entangled with the core feature to form a star topology; A nonlinear transformation is performed on the current quantum state to generate the nonlinear characteristics corresponding to the current quantum state; Based on the higher-order entanglement features, the star topology, and the nonlinear features, the node quantum states corresponding to each decision node are generated.

6. The method according to claim 1, characterized in that, The method of inputting the sediment evolution trend category, the current erosion and deposition topography sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data into the preset sediment erosion change prediction model, and outputting the sediment cross-sectional erosion change corresponding to the target gate within the future preset time period, includes: The sediment evolution trend category, the current erosion and deposition topography sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data are input into the time dependency network of the preset sediment erosion change prediction model, and the time dependency features are output. The sediment evolution trend category, the current erosion and deposition topography sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data are input into the structural feature extraction network of the preset sediment erosion change prediction model, and the topological structure features are output. The time-dependent features and the topological features are fused to output a fused prediction feature; Based on the fusion prediction features, the sediment cross-sectional scour change within the future preset time period corresponding to the target gate is output.

7. The method according to claim 6, characterized in that, The process involves inputting the sediment evolution trend category, the current erosion and deposition topography sequence, the future precipitation-runoff data, the future upstream-downstream water level difference, the future tidal level change data, and the future gate scheduling data into the time-dependent network of the preset sediment erosion change prediction model, and outputting time-dependent features, including: The sediment evolution trend category, the current erosion and deposition topography sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data are input into the time dependency network of the preset sediment erosion change prediction model. The weight determination network in the time-dependent network evaluates the input data at each time step to determine the importance of the input data at each time step to the current scour and deposition terrain sequence and the scour and deposition changes of the sediment cross section within the future preset time period. Based on the importance of the input data at each time step, determine the weight information corresponding to the input data at each time step; Based on the weight information corresponding to the input data at each time step, the update gate and forget gate in the time-dependent network are dynamically adjusted; Based on the adjusted update gate and forget gate at each time step, time series features are extracted from the input data, and the time-dependent features are output.

8. The method according to claim 6, characterized in that, The process involves inputting the sediment evolution trend category, the current erosion and deposition topography sequence, the future precipitation-runoff data, the future upstream-downstream water level difference, the future tidal level change data, and the future gate scheduling data into the structural feature extraction network of the preset sediment erosion change prediction model, and outputting topological features, including: Based on the current erosion and deposition terrain sequence, the future precipitation-runoff data, the future upstream and downstream water level difference, the future tidal level change data, and the future gate scheduling data, a target map structure is constructed; Based on the structural feature extraction network, feature extraction is performed on the graph structure to output the topological structure features.

9. The method according to claim 1, characterized in that, The determination of future gate scheduling data corresponding to the target gate based on the future precipitation-runoff data, the future upstream-downstream water level difference, and the future tidal level change data includes: Acquire historical precipitation-runoff data, historical upstream and downstream water level difference, historical tidal level change data, and historical gate scheduling data corresponding to the historical precipitation-runoff data, the historical upstream and downstream water level difference, and the historical tidal level change data. Based on the historical precipitation-runoff data, the historical upstream and downstream water level difference, the historical tidal level change data, and the historical gate scheduling data, a cause-effect graph structure is constructed. The causal graph structure is extended into a dynamic structural equation model to capture the time-delayed causal relationships between variables. Substituting the future precipitation-runoff data, the future upstream-downstream water level difference, and the future tidal level change data into the time-delay causal relationship, the future gate scheduling data corresponding to the target gate is determined.

Images