A method and system for assessing and warning of the risk of insufficient channel dimensions

Abstract

This invention provides a method and system for assessing and issuing early warnings of insufficient channel scale risk. The method includes: collecting mapping and detection data for each waterway; constructing and validating water and sediment numerical models for each waterway based on the mapping and detection data; the detection data includes data from n waterway detection missions, and calculating the predicted average scour and sedimentation amplitude for each time corresponding to each detection mission based on the water and sediment numerical models; constructing multiple feature vector sets for each waterway based on the detection data and the predicted average scour and sedimentation amplitude, with each feature vector set matching the corresponding time feature factors based on the execution time of each waterway detection mission; training and validating the machine learning early warning model for each waterway using the multiple feature vector sets; and evaluating the scale alarm prediction level for each waterway using the machine learning early warning model. This invention improves the accuracy and stability of insufficient scale risk early warning.

Description

Technical Field

[0001] This invention relates to the field of waterway scale risk assessment technology, specifically to a method and system for assessing and warning of insufficient waterway scale risks. Background Technology

[0002] In waterway operation and management, the commonly used methods for risk assessment and early warning of insufficient waterway dimensions are mainly based on a combination of mechanistic models and water level prediction.

[0003] In terms of mechanistic models, two-dimensional hydro-sediment models are constructed to simulate and analyze the physical processes of water flow and sediment erosion in waterways, thereby obtaining information on the scale changes of waterways under different water conditions and operating conditions. Such models can deeply reveal the intrinsic physical mechanisms of waterway scale changes and are of great significance for understanding the laws governing waterway evolution. However, certain limitations exist in practical applications. On the one hand, the construction of two-dimensional hydro-sediment models requires a large amount of basic data, including topography, hydrology, and sediment data, which is difficult and costly to acquire. Furthermore, the accuracy and completeness of the data significantly affect the model's precision; errors or missing data may lead to significant deviations between the model simulation results and actual conditions. On the other hand, mechanistic models have limited ability to simulate waterway scale changes under complex boundary conditions and the coupled effects of multiple factors, making it difficult to comprehensively and accurately reflect the actual scale conditions of waterways under the combined influence of various factors.

[0004] Water level prediction plays a crucial role in risk assessment of insufficient channel dimensions. By establishing a water level prediction model, water level changes over a future period are predicted. Combined with parameters such as the channel's design depth, a preliminary assessment can be made as to whether the channel dimensions may be insufficient. The accuracy of the water level prediction model directly affects the accuracy of the risk assessment. Currently, commonly used water level prediction methods include time series analysis and neural network models. Time series analysis is relatively simple, but its ability to predict nonlinear and non-stationary water level changes is insufficient. While neural network models can handle complex nonlinear relationships, they require a large amount of historical data for training, and their generalization ability is greatly affected by the quality and quantity of data. Furthermore, most existing water level prediction models only consider the impact of single or a few factors such as meteorology and hydrology on water levels, and do not adequately consider the comprehensive consideration of complex factors such as water conservancy project scheduling and human activities, leading to a certain deviation between the predicted results and actual water level changes.

[0005] In summary, while the commonly used method of combining mechanistic models with water level prediction for early warning of insufficient scale can provide a reference for waterway operation and management to some extent, due to the limitations of each method, it is difficult to comprehensively and accurately assess the risk of insufficient scale in waterways in practical applications, and the accuracy and timeliness of early warning need to be improved. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for assessing and warning of insufficient channel scale risk. By combining a water and sediment numerical model with a machine learning model, it overcomes the problems of traditional mechanistic models' strong data dependence, difficulties in handling boundary conditions, and insufficient consideration of complex factors by single water level prediction models. By introducing multi-source feature factors (including real-time indicators, statistical indicators, and predictive indicators) and using the XGBoost machine learning model to train and validate on historical data, it can more comprehensively reflect the complex patterns of channel scale changes, significantly improving the accuracy and stability of insufficient scale risk warnings.

[0007] To address the aforementioned technical problems, this invention provides a method for risk assessment and early warning of insufficient waterway dimensions, comprising: Collect mapping and survey data of each waterway in the waterway; Based on mapping and survey data, numerical models of water and sediment in each channel of the waterway were constructed and verified. The detection data includes data from n channel exploration missions, and the average scouring and deposition amplitude prediction for each channel exploration mission at the corresponding time is calculated based on the water and sediment numerical model. Based on the detection data and the average scour and siltation amplitude prediction, multiple feature vector sets are constructed for each waterway. Each feature vector set includes multiple feature factors. Each feature vector set is matched with the feature factors corresponding to the execution time of each waterway detection mission. The feature factors include waterway width, water depth, water level, flow rate, weather, waterway anomalies, average scour and siltation amplitude prediction, and scale alarm prediction level. The multiple feature vector sets of each waterway are divided into training set and validation set to train and validate the machine learning early warning model of each waterway. The scale-based alarm prediction level of each waterway is evaluated using a machine learning early warning model.

[0008] In some embodiments, constructing and validating numerical models of water and sediment in each channel of the waterway includes: A two-dimensional water and sediment numerical model was constructed based on the measured river topographic map in the mapping data, and the grid was divided. The flow parameters of the two-dimensional water and sediment numerical model were calibrated and verified based on the detection data. The sediment parameters of the two-dimensional water and sediment numerical model were verified based on the detection data.

[0009] In some embodiments, the prediction of the average scouring and deposition amplitude for the corresponding time of n channel exploration missions based on the water and sediment numerical model includes: Calculate the predicted average scouring and silting amplitude for the time corresponding to the i-th channel exploration mission: The execution time of the i-th channel exploration mission is t. i Using time t i The latest mapping data is used to update the two-dimensional hydro-sediment numerical model. The updated two-dimensional hydro-sediment numerical model is then used to simulate the flow and sediment parameters, and the time t is obtained. i Predicted elevation; Time t i The predicted elevation is compared with the actual elevation measured by the i-th channel exploration mission. If the deviation meets the requirements, the updated two-dimensional water and sediment numerical model is used to simulate the flow and sediment parameters after verification, and the time t is obtained. i and t i+1 The corresponding bed surface elevation is used to calculate time t. i+1 Relative to t i The scouring and silting situation is used as a prediction of the average scouring and silting amplitude at the corresponding time of the i-th channel exploration mission; If the deviation does not meet the requirements, the verified flow parameters and sediment parameters will be optimized. Where 1≤i≤n-1, t i+1 The execution time of the (i+1)th channel exploration mission.

[0010] In some embodiments, if the deviation does not meet the requirements, a smart algorithm based on the Pareto optimal concept is used to optimize the verified flow parameters and sediment parameters. The updated two-dimensional hydro-sediment numerical model is used to simulate the flow and sediment parameters, and the time t is obtained. i Predicted elevation; Time t i The predicted elevation is compared with the actual elevation measured in the i-th channel exploration mission. If the deviation still does not meet the requirements, the Pareto optimal concept intelligent algorithm is used again to optimize the flow parameters and sediment parameters until the deviation meets the requirements.

[0011] In some embodiments, the multiple feature factors include real-time indicators, statistical indicators, and predictive indicators, respectively. The real-time indicators corresponding to the execution time of the i-th channel exploration mission include the current channel width, current shallowest water depth, current water level, weather, ship grounding, and dredging operations at the time corresponding to the i-th channel exploration mission. The statistical indicators corresponding to the execution time of the i-th channel exploration mission include the upstream predicted average water level, downstream predicted average water level, upstream predicted average flow, and shallow area predicted average scouring and silting amplitude during the period from the i-th channel exploration mission to the (i+1)-th channel exploration mission. The upstream predicted average water level is the average of the historical water level data of the upstream water level station during the period from the i-th waterway exploration mission to the (i+1)-th waterway exploration mission; The downstream predicted average water level is the average of the historical water level data of the downstream water level station during the period from the i-th waterway exploration mission to the (i+1)-th waterway exploration mission; The upstream predicted average flow is the average of the historical flow data of the upstream flow station during the period from the i-th channel exploration mission to the (i+1)-th channel exploration mission; The predicted average scour and sedimentation amplitude in shallow areas represents the average thickness of scour or sedimentation on the shallow bed surface within the time range from the i-th channel exploration mission to the (i+1)-th channel exploration mission. The predicted average scour and sedimentation amplitude in shallow areas is calculated based on the water and sediment numerical model of the waterway. The prediction indicators corresponding to the execution time of the i-th channel exploration mission include the scale alarm prediction level, which is the scale alarm level obtained based on the exploration situation of the (i+1)-th channel exploration mission.

[0012] In some embodiments, training and validating the machine learning early warning model for each waterway includes: The XGBoost model, LightGBM model, and SVM model were trained using the training set, and tested using the validation set. The accuracy of the XGBoost model, LightGBM model, and SVM model for the prediction results of each waterway was calculated, and one of the models was selected as the machine learning early warning model for each waterway based on the accuracy.

[0013] In some embodiments, training and validating the machine learning early warning model for each waterway includes: The XGBoost model was used as the machine learning early warning model for each waterway. The core hyperparameters of the XGBoost model are automatically optimized using a Bayesian optimization algorithm, with the goal of maximizing the classification accuracy on the validation set.

[0014] In some embodiments, automatically optimizing the core hyperparameters of the XGBoost model using a Bayesian optimization algorithm includes: Determine the hyperparameters to be optimized and their search range, including n_estimators, max_depth, learning_rate, subsample, colsample_bytree, reg_alpha, and reg_lambda; Define the objective function as the classification accuracy on the validation set; Initialize the Bayesian optimizer and perform stochastic exploration and directed iteration to maximize the objective function; The optimal hyperparameter combination is extracted from the optimization results and used to train the XGBoost model.

[0015] In some embodiments, the machine learning early warning model includes a SHAP interpreter, which is used to analyze the importance of feature factors in the machine learning early warning model of each waterway and quantify the contribution of feature factors to the risk of insufficient channel scale in each waterway.

[0016] In some embodiments, evaluating the scale-based alarm prediction level of each waterway using a machine learning early warning model includes: Obtain the current channel width, current shallowest water depth, current water level, weather, ship grounding, and dredging operations corresponding to the execution time of the most recent channel survey mission; Calculate the upstream predicted average water level, downstream predicted average water level, and upstream predicted average flow rate between the execution time of the most recent channel exploration mission and the time to be predicted. Calculate the predicted average scour and sedimentation amplitude in the shallow area between the execution time of the most recent channel exploration mission and the predicted time: Update the two-dimensional hydro-sediment numerical model using the latest mapping data. Simulate the updated two-dimensional hydro-sediment numerical model using verified flow and sediment parameters to obtain the predicted elevation corresponding to the execution time of the most recent channel exploration mission. Compare the predicted elevation corresponding to the execution time of the most recent channel exploration mission with the actual elevation. If the deviation meets the requirements, simulate the updated two-dimensional hydro-sediment numerical model using verified flow and sediment parameters to obtain the bed elevation corresponding to the execution time of the most recent channel exploration mission and the predicted time. Calculate the predicted average scour and sedimentation amplitude in the shallow area between the execution time of the most recent channel exploration mission and the predicted time based on the bed elevation. If the deviation does not meet the requirements, optimize the verified flow and sediment parameters. The machine learning early warning model outputs the scale alarm prediction level based on the feature factors corresponding to the execution time of the most recent channel exploration mission. The scale alarm prediction level represents the scale alarm level at the time to be predicted.

[0017] On the other hand, the present invention provides a system for implementing the aforementioned method for risk assessment and early warning of insufficient waterway dimensions, comprising: The data acquisition module is used to collect mapping and detection data from various waterways of the waterway; The water and sediment model construction and calculation module is used to construct and verify the water and sediment numerical models of each waterway based on the mapping data and exploration data; and to calculate the average scouring and silting amplitude prediction for each waterway exploration mission at the corresponding time based on the water and sediment numerical models. The feature vector construction module is used to construct the feature vector set corresponding to the execution time of each channel exploration mission in each waterway; The model training and validation module is used to divide the multiple feature vector sets of each waterway into training and validation sets, and to train and validate the machine learning early warning models for each waterway. The risk assessment module is used to evaluate the scale alarm level of each waterway using the machine learning early warning model.

[0018] The beneficial effects of this invention are as follows: 1. This invention overcomes the problems of traditional mechanistic models, such as strong data dependence, difficulty in handling boundary conditions, and insufficient consideration of complex factors by single water level prediction models, by combining a numerical water-sediment model with a machine learning model. By introducing multi-source feature factors (including real-time indicators, statistical indicators, and predictive indicators) and using the XGBoost machine learning model to train and validate on historical data, it can more comprehensively reflect the complex laws of channel scale changes and significantly improve the accuracy and stability of inadequate scale risk warning.

[0019] 2. The feature vector set constructed in this invention covers a variety of influencing factors such as channel width, water depth, water level, flow rate, weather, scour and sedimentation changes, and waterway anomalies (such as dredging construction). In particular, it introduces the shallow area predicted average scour and sedimentation amplitude calculated based on the water and sediment numerical model, which can effectively reflect the evolution trend of the waterway under complex water conditions and engineering scheduling, and improve the ability to identify the changes in waterway scale under the influence of multiple factors.

[0020] 3. This invention employs an intelligent algorithm based on the Pareto optimality concept, which simultaneously uses "water level accuracy" and "terrain accuracy" as two different objectives as optimization indicators. Through rapid iterative calculation, it automatically finds the "golden balance point" among dozens of parameter combinations that improves terrain accuracy without significantly reducing water level accuracy, thereby obtaining the optimal compromise solution parameter combination. This achieves automatic balance and coordination of the model parameters' ability to represent different physical processes.

[0021] 4. This invention introduces the SHAP interpreter into the machine learning model, which can quantify the contribution of each feature factor to the early warning result and can globally identify the most critical feature factors that lead to various abnormalities.

[0022] 5. This invention uses a Bayesian optimization algorithm to automatically optimize the core hyperparameters of the XGBoost model, further improving the model's generalization ability and classification accuracy, and avoiding the subjectivity and uncertainty of manual parameter tuning.

[0023] 6. This invention constructs machine learning early warning models for different waterways, fully considering the differences in topography, hydrology, scour and sedimentation patterns of each waterway, which can achieve refined and differentiated risk assessment and improve the pertinence and practicality of early warning results. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method of the present invention.

[0025] Figure 2 This is a schematic diagram illustrating the details of scale warning information in an embodiment of the present invention.

[0026] Figure 3 This is a calculation of the river segment grid partitioning diagram in an embodiment of the present invention.

[0027] Figure 4 This is a schematic diagram of terrain calculation in an embodiment of the present invention.

[0028] Figure 5 This is a schematic diagram of the layout of the on-site hydrological measurement and verification section in an embodiment of the present invention.

[0029] Figure 6 This is a comparison chart of the calculated and measured values ​​of the water level process at each measuring point in the embodiments of the present invention.

[0030] Figure 7 This is a comparison chart of the calculated and measured values ​​of the flow velocity and direction at each measuring point in the embodiments of the present invention.

[0031] Figure 8 This is a schematic diagram of water level verification in an embodiment of the present invention.

[0032] Figure 9 This is a schematic diagram of flow rate verification in an embodiment of the present invention.

[0033] Figure 10 This is a model verification diagram in an embodiment of the present invention.

[0034] Figure 11 This is a diagram showing the extraction range of the average scouring and silting amplitude in the shallow area in an embodiment of the present invention.

[0035] Figure 12 This is an example diagram of the XGBoost model output in an embodiment of the present invention.

[0036] Figure 13 This is a fragment diagram of the XGBoost model content in an embodiment of the present invention.

[0037] Figure 14 This is an ROC curve of a waterway in an embodiment of the present invention.

[0038] Figure 15 This is a confusion matrix result diagram of a certain waterway in an embodiment of the present invention.

[0039] Figure 16 This is a schematic diagram illustrating the feature importance of a certain waterway in an embodiment of the present invention. Detailed Implementation

[0040] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0041] like Figure 1 As shown, the present invention provides a method for risk assessment and early warning of insufficient waterway dimensions, including: S1. Collect mapping and survey data of the waterway.

[0042] The data includes data under normal scale conditions and abnormal data caused by special hydrological conditions, forming positive and negative samples. Among them, special hydrological conditions include major floods, drought reversals during the flood season, and joint operation of water conservancy projects.

[0043] The detection data includes at least the data detected in each waterway detection mission, as well as the data detected by water level stations, flow stations, and meteorological stations.

[0044] The data collected during channel survey missions includes the current channel width, water depth data, scale warning information, and channel anomalies (such as ship groundings and dredging operations) for each waterway. This data is saved after each channel survey mission as historical data for later retrieval.

[0045] The data collected by water level and flow stations include water level and flow rate data; the water level data for each waterway is determined based on the water level measurements taken at 8:00 AM daily by the inlet and outlet water level stations. The flow rate data for each waterway is determined based on the flow rate measurements taken at 8:00 AM daily by the inlet and outlet flow stations. Meteorological data is determined based on data collected by meteorological stations.

[0046] S2. Based on mapping and detection data, construct and verify numerical models of water and sediment in each channel of the waterway.

[0047] Step S2 includes: S21. Based on the channel mapping data, construct a two-dimensional water and sediment numerical model for the waterway, and subdivide the waterway using an unstructured triangular mesh. Taking a certain waterway as an example, its mesh count is 37795, the number of nodes is 19319, and the mesh scale is 40m. Figure 3 As shown. The modeling terrain data is a 1:10000 river topographic map measured in March 2022, as shown. Figure 4 As shown.

[0048] S22. Calibrate the flow parameters and validate the model: The mathematical model for fixed-bed beds was adopted on August 7, 2019 (30913m). 3The flow parameters were calibrated using measured hydrological data (50942m / s) on August 24, 2020. 3 / s), April 19, 2022 (14193m) 3 The measured hydrological data ( / s) verify the above flow parameters, as shown in Table 1. The layout of the hydrological test sections for the fixed-bed model is shown in [reference needed]. Figure 5 .

[0049] Table 1 like Figure 6 , 7 As shown, after calibration, the roughness of the waterway is between 0.030 and 0.034, the water level error is less than ±0.5m, and the average flow velocity error of each vertical line is within 0.10m / s. The calibration results meet the accuracy requirements of the "Technical Specification for Simulation Test of Water Transport Engineering" (JTS / T 231-2021).

[0050] like Figure 8 , 9 As shown, the model was validated using calibrated flow parameters. The calculated values ​​were largely consistent with the measured values ​​from the prototype. The validation errors for each water gauge value (velocity and water level) were all less than ±0.05m, the average velocity error for each vertical line was within 0.10m / s, and the deviation between the calculated and measured values ​​of the channel splitting ratio was generally within 1%. This indicates that the calibrated flow parameters can effectively reproduce the comprehensive characteristics of the river flow in the research section, meeting the accuracy requirements of the "Technical Specification for Simulation Tests of Water Transport Engineering" (JTS / T 231-2021).

[0051] S23. Verification of sediment parameters: The model's sediment parameters were validated from March 2023 to March 2024. The scouring and deposition distribution map is shown below. Figure 10 ( Figure 10 (The top side represents the measured value, and the bottom side represents the verification value).

[0052] Table 2 shows the measured and calculated scour and sedimentation volumes for each section of the waterway, verifying that the river section experienced alternating scour and sedimentation during the specified period. The calculation results indicate that the distribution trend of scour and sedimentation calculated by this model is basically consistent with the measured values, with an error within 20%, meeting the requirements of the specifications.

[0053] Table 2 The verification results show that the model's calculation of the scouring and deposition trend is basically consistent with the measured values, and further research can be carried out based on these sediment parameters.

[0054] S24. The detection data includes data from n channel exploration missions, and the average scouring and sedimentation amplitude prediction for each channel exploration mission at the corresponding time is calculated based on the water and sediment numerical model.

[0055] Step S24 includes: S241, the execution time of the i-th channel exploration mission is t. i Using time t i The latest mapping data is used to update the two-dimensional hydro-sediment numerical model. The updated two-dimensional hydro-sediment numerical model is then used to simulate the flow and sediment parameters, and the time t is obtained. i Predicted elevation; S242, Time t i The predicted elevation is compared with the actual elevation measured in the i-th channel exploration mission to determine whether the deviation between the actual elevation of each measuring point and the predicted elevation at the corresponding coordinate of the measuring point in the i-th channel exploration mission meets the requirements. Deviation = |(predicted elevation - actual elevation) / actual elevation|. If the deviation meets the requirement, i.e., the maximum deviation does not exceed 20%, then the updated two-dimensional water and sediment numerical model is used to simulate the flow and sediment parameters after verification, and the time t is obtained. i and t i+1 The corresponding bed surface elevation is used to calculate time t. i+1 Relative to t i The scouring and silting situation is used as a prediction of the average scouring and silting amplitude at the corresponding time of the i-th channel exploration mission; If the deviation does not meet the requirements, a smart algorithm based on the Pareto optimal concept will be used to optimize the verified flow and sediment parameters: Based on the verified flow and sediment parameters in steps S22 and S23, set the range of values ​​for the flow and sediment parameters (e.g., ±20%), and randomly generate multiple sets of parameter combinations for the flow and sediment parameters. The updated two-dimensional water and sediment numerical model was simulated with different parameter combinations to obtain the predicted elevation and predicted water level corresponding to different parameter combinations; Based on the predicted elevation and water level data, as well as the actual elevation and water level, an intelligent algorithm based on the Pareto optimal concept is used to solve the problem and identify a set of balanced flow and sediment parameters (i.e., the Pareto front). This means finding the "golden balance point" that improves topographic accuracy without significantly reducing water level accuracy. Through iterative search of the algorithm, the set of flow and sediment parameters that best matches this is selected as the optimized flow and sediment parameters. The updated two-dimensional hydro-sediment numerical model is used to simulate the flow and sediment parameters, and the time t is obtained. i Predicted elevation; Time t iThe predicted elevation is compared with the actual elevation measured by the i-th channel exploration mission. If the deviation still does not meet the requirements, the range of values ​​for the flow parameters and sediment parameters is adjusted, and the intelligent algorithm based on the Pareto optimal concept is used again to optimize the flow parameters and sediment parameters until the deviation meets the requirements. Where 1≤i≤n-1, t i+1 The execution time of the (i+1)th channel exploration mission.

[0056] It should be noted that although steps S22 and S23 verify that the flow and sediment parameters meet the accuracy requirements, their accuracy will decrease over time. Therefore, when the mapping and / or detection data are updated, the latest mapping and detection data can be used to update the two-dimensional water and sediment numerical model, flow parameters, and sediment parameters to better reflect the current situation and improve prediction accuracy. Since there may be a trade-off between the fitting error of the water level process and the spatial distribution error of topographic erosion and deposition, this invention employs an intelligent algorithm based on the Pareto optimality concept. This algorithm simultaneously uses "water level accuracy" and "topographic accuracy" (i.e., water depth) as optimization indicators. Through rapid iterative calculations, it automatically finds the "golden balance point" among dozens of parameter combinations that improves topographic accuracy without significantly reducing water level accuracy, thereby obtaining the optimal compromise solution parameter combination. This achieves automatic balance and coordination of the model parameters' ability to represent different physical processes.

[0057] The calculation methods for predicting the average scouring and deposition amplitude in shallow areas based on the two-dimensional water and sediment numerical model include: (1) Input condition setting 1) Boundary conditions: Flow process line: Set daily flow sequence based on digital waterway predicted water level data, such as increasing discharge during flood season and controlling discharge during non-flood season.

[0058] Sediment concentration process curve: Based on the upstream sediment inflow pattern and the sediment inflow values ​​of a certain station over the past 5 years, combined with the relationship between flow rate and sediment concentration, the suspended sediment concentration is comprehensively determined. For example, high sediment concentration (1.5~3.0 kg / m³) during the flood season. 3 Low sediment content (0.2~0.5 kg / m³) during the dry season. 3 ).

[0059] Bed sand composition: Bed sand gradation was set based on the 2019 cross-sectional bed sand gradation data.

[0060] 2) Export boundary conditions: Water level process curve: Based on long-term observation data from downstream hydrological stations, the water level change at the outlet is set. Constant water level or water level-discharge relationship processes can be used.

[0061] 3) Initial conditions: Initial water level: either initial translational or based on previous equilibrium state; Initial flow velocity: zero or small disturbance; Initial riverbed elevation: based on the actual topographic measurements taken in March 2024.

[0062] 4) Model parameters: Manning coefficient: The value after calibration; Settling velocity: based on sediment particle size (e.g., d) 50 Calculate (=0.03mm); The diffusion coefficient, recovery saturation coefficient, and other parameters were all verified.

[0063] (2) Model running process The model uses a 1-day time step to advance the calculation. Within each time step, the following operations are performed: The hydrodynamic module solves the shallow water equations and updates the flow field distribution based on the current water depth, flow velocity, and topography. Calculate the shear stress on the bed surface: Based on the velocity gradient and Manning's formula, calculate the shear force of each grid element; Determining the scouring and deposition status: If the shear stress exceeds the critical starting shear force, scouring occurs; if it is below the settlement threshold, deposition occurs. The sediment transport module calculates changes in suspended sediment concentration field and updates upstream and downstream sediment fluxes. Update riverbed elevation: Based on the Exner equation, update the riverbed elevation Zb grid by grid to achieve dynamic topographic evolution.

[0064] (3) Output target: Extraction of the predicted average scour and sedimentation amplitude in shallow areas The core output of this study is to define the scouring and sedimentation extent within the shallow zone. To achieve this, the following steps are performed during the model post-processing stage: Define the shallow area: Based on waterway maintenance standards or historical shallow point distribution, delineate the area to be maintained year-round. Specific areas include... Figure 11 As shown in the red rectangle in the middle.

[0065] Extracting sedimentation data: At the end of the simulation, read the changes in bed elevation for each shallow grid area. Positive values ​​indicate siltation, while negative values ​​indicate scouring; the average value of the change in bed elevation in each shallow zone grid is the predicted average scouring and silting amplitude in the shallow zone, in meters.

[0066] This represents the change in bed elevation at two time points in the simulation. and The change in bed surface elevation between At a certain point in time The elevation of the bed surface, At a certain point in time The elevation of the bed surface.

[0067] S3. Based on the detection data and the average scour and siltation amplitude prediction, construct multiple feature vector sets for each waterway. Each feature vector set includes multiple feature factors. Each feature vector set is matched with the feature factors corresponding to the execution time of each waterway detection mission. The feature factors include waterway width, water depth, water level, flow rate, weather, waterway anomalies, average scour and siltation amplitude prediction, and scale alarm prediction level.

[0068] Step S3 includes: This invention uses the execution time of each waterway exploration mission as a benchmark, and matches the real-time indicators of the current waterway exploration mission with the predicted indicators and statistical indicators within the time period from the current mission to the next mission to form a single complete sample data, which serves as the feature vector set of the current waterway exploration mission.

[0069] The multiple feature factors include real-time indicators, statistical indicators, and predictive indicators.

[0070] The real-time indicators corresponding to the execution time of the i-th channel exploration mission include the current channel width, current shallowest water depth, current water level, weather, ship grounding, and dredging operations at the time corresponding to the i-th channel exploration mission. Meteorology refers to the weather conditions at the time corresponding to the i-th channel exploration mission in the river, including sunny, light rain, moderate rain, and heavy rain, which are obtained from meteorological station information. Sunny, light rain, moderate rain, and heavy rain are represented by 0, 1, 2, and 3, respectively.

[0071] Ship grounding refers to whether a ship grounds during the i-th channel exploration mission. A 1 indicates the presence of a grounded ship, and a 0 indicates the absence of a grounded ship.

[0072] Dredging construction refers to whether dredging construction occurs during the i-th channel exploration mission. 1 indicates dredging construction, and 0 indicates non-dredging construction.

[0073] The statistical indicators corresponding to the execution time of the i-th channel exploration mission include the upstream predicted average water level, downstream predicted average water level, upstream predicted average flow, and shallow area predicted average scouring and silting amplitude during the period from the i-th channel exploration mission to the (i+1)-th channel exploration mission. The upstream predicted average water level is the average of the historical water level data of the upstream water level station during the period from the i-th waterway exploration mission to the (i+1)-th waterway exploration mission; The downstream predicted average water level is the average of the historical water level data of the downstream water level station during the period from the i-th waterway exploration mission to the (i+1)-th waterway exploration mission; The upstream predicted average flow is the average of the historical flow data of the upstream flow stations during the period from the i-th to the (i+1)-th channel exploration mission.

[0074] The predicted average scour and sedimentation amplitude in shallow areas represents the average thickness of scour or sedimentation on the shallow bed surface within the time range from the i-th channel exploration mission to the (i+1)-th channel exploration mission. The predicted average scour and sedimentation amplitude in shallow areas is calculated based on the water and sediment numerical model of the waterway.

[0075] The predicted indicators corresponding to the execution time of the i-th channel exploration mission include the scale alarm prediction level. The scale alarm prediction level is the scale alarm level obtained based on the exploration situation of the (i+1)-th channel exploration mission. That is, if the scale alarm level of the (i+1)-th channel exploration mission is level one, then the scale alarm prediction level corresponding to the execution time of the i-th channel exploration mission is set to level one.

[0076] The risk of insufficient channel dimensions is classified into five levels: normal, Level 1 warning, Level 2 warning, Level 3 warning, and channel dimensions alarm, represented by 0, 1, 2, 3, and 4 respectively. These levels serve as the output labels for the machine learning model. The threshold values ​​for channel dimensions risk classification are set as follows: The criteria for determining a waterway dimension alarm are: the measured water depth or width is less than the planned dimension, the measured water level is less than the historical lowest water level of the waterway, or the water level value of the control section is insufficient.

[0077] The criteria for a Level 1 warning for waterway dimensions are: the measured water depth is close to the monthly planned dimensions (the excess water depth in the upstream waterway is ≤0.2 meters, and the excess water depth in the mid-to-downstream waterway is ≤0.4 meters), and the excess width of the waterway is insufficient.

[0078] The criteria for a Level II warning for waterway dimensions are: the measured water depth is close to the monthly planned dimensions (the excess water depth in the upstream waterway is ≤0.2 meters, and the excess water depth in the mid-to-downstream waterway is ≤0.4 meters), but the waterway has a certain excess width.

[0079] The criteria for a Level III warning for waterway dimensions are: the measured water depth is close to the predicted water depth of the section, or the measured navigation width is less than the predicted navigation width of the section.

[0080] The criteria for judging normal channel dimensions are: the measured water depth or width reaches or exceeds the planned dimensions, the channel maintenance dimensions meet the planned maintenance standards approved by the superior authority, the channel condition is good, and it meets the requirements for ship navigation.

[0081] Table 3 shows a sample of the training set. Two channel surveys were conducted on November 7th and 8th, 2024. In the row for November 7th, the current channel width, current shallowest water depth, current water level, weather, ship grounding, and dredging operations are all real-time results from that survey. The upstream predicted average water level, downstream predicted average water level, and upstream predicted average flow are averages calculated based on data collected from the corresponding hydrological stations on November 7th and 8th. The shallow area predicted average scour and sedimentation amplitude is the prediction value from the mechanistic model for November 7th and 8th. The scale alarm prediction level for November 7th, 2024 is shown as 3 because the measured channel width on November 8th was 100.00m, which, when subtracted from the weekly forecast channel width of 110.00m, is less than 0, meeting the criteria for a level 3 warning. Through the above data preprocessing steps, the raw data is transformed into structured data suitable for model training, laying the foundation for subsequent model building.

[0082] Table 3 S4. Divide the feature vector sets of each waterway into training and validation sets, and train and validate the machine learning early warning models for each waterway.

[0083] Step S4 includes: S41. Train the XGBoost model, LightGBM model, and SVM model using the training set, and test the XGBoost model, LightGBM model, and SVM model using the validation set. Calculate the accuracy of the XGBoost model, LightGBM model, and SVM model for the prediction results of each waterway, and determine one of the models as the machine learning early warning model for each waterway based on the accuracy.

[0084] During the training process of the aforementioned machine learning early warning model, its performance is evaluated to ensure good generalization ability. Therefore, a suitable machine learning algorithm needs to be selected. Considering the relatively limited amount of waterway monitoring data and the fact that the early warning problem is essentially a classification problem, this invention compares the following types of machine learning models: Traditional machine learning models: Logistic Regression, Support Vector Machine (SVM), Decision Tree, Random Forest; Ensemble learning models: AdaBoost, Gradient Boosting Machine (GBM), XGBoost; Neural network model: Simple feedforward neural network.

[0085] Preliminary experimental comparisons revealed that neural network models are prone to overfitting on small sample datasets, and their training process is complex and has poor interpretability. Traditional machine learning models, such as logistic regression and support vector machines, although fast in training, have limited ability to capture nonlinear relationships. Ensemble learning models, on the other hand, perform better in classification and generalization.

[0086] Therefore, XGBoost, LightGBM, and SVM models were used to model each waterway, and the accuracy results are compared in Table 4. Overall, the XGBoost model performed the most stably and excellently, achieving no lowest accuracy in any waterway and reaching or tying for the highest accuracy in several waterways (such as waterways B, E, D, and G). The LightGBM model performed similarly to XGBoost in most cases, but slightly worse in the Tuqiao waterway. While the SVM model achieved 100% accuracy in the Zhijiang waterway, its performance in several waterways (such as waterways B and G) was inferior to the other two models, indicating poor overall stability. In summary, XGBoost is the best performing of the three models overall, maintaining high accuracy while exhibiting better generalization ability and stability.

[0087] Table 4 XGBoost model principle: XGBoost (eXtreme Gradient Boosting) is a high-efficiency machine learning algorithm based on the gradient boosting framework, particularly adept at handling structured data and exhibiting excellent performance in classification and regression tasks. Its core principles combine ensemble learning, gradient boosting, and regularization optimization.

[0088] (1) Gradient Boosting Framework XGBoost belongs to the Boosting class of algorithms. It constructs a strong learner by iteratively training multiple weak learners (usually decision trees) and combining their predictions. Its core idea is: Sequential training: Each round of newly trained trees focuses on correcting the residuals of the previous model (i.e., the error between the predicted and the true values).

[0089] Additive model: The final prediction is a weighted sum of the predictions from all the trees. in Let represent the final predicted value of the i-th sample, k represent the index of the k-th decision tree, F is the space of all possible decision trees, and K is the number of trees. It is the k-th decision tree for the i-th sample. The predicted output.

[0090] (2) Optimization of the objective function XGBoost's objective function consists of a loss function and a regularization term. The model is optimized by minimizing the objective function. loss function : Measures the difference between predicted and actual values ​​(such as mean squared error, log loss, etc.).

[0091] It is the ground truth label of the i-th sample.

[0092] It is the model's prediction for the i-th sample.

[0093] Regularization term Controlling model complexity and preventing overfitting includes: K is the number of decision trees generated in XGBoost.

[0094] It is the k-th decision tree, which represents a mapping function from sample features to output. This is a complexity penalty for a single tree, usually defined as: in: T is the number of leaf nodes in the tree. w is the weight vector of the leaf nodes. γ and λ are hyperparameters that control the intensity of regularization.

[0095] (3) Taylor expansion and second-order approximation XGBoost uses a second-order Taylor expansion to approximate the loss function, accelerating the optimization process: The objective function (approximate value) during the t-th training round. The total number of training samples.

[0096] The tree structure function represents the i-th sample. Mapped to the Leaf nodes.

[0097] No. The weight of each leaf node, which is the weight of the current tree. For the sample The predicted output.

[0098] (First-order gradient).

[0099] (Second-order gradient).

[0100] Second-order information (Hessian matrix) can more accurately guide the direction of tree splitting.

[0101] in: The first-order gradient of the i-th sample represents the rate of change of the loss function at the current predicted value. : To the front Predicted value of the wheel Find the sign of the partial derivative Loss function: measures the difference between the actual value and the predicted value. : The ground truth label of the i-th sample forward The prediction value of the round model for the i-th sample : The second gradient (second derivative / Hessian matrix) of the i-th sample, representing the rate of change of the first gradient, reflecting the curvature of the loss function.

[0102] : To the front Predicted value of the wheel Find the sign of the second-order partial derivative.

[0103] (4) Tree splitting and node gain When constructing each tree, XGBoost uses a greedy algorithm to select the optimal split point: Split gain: Measures the reduction in the objective function after splitting. in and It is the sample set of the left and right child nodes after the split. It is the minimum gain threshold for splitting.

[0104] (5) Key optimization techniques Column Block Structure: Features are pre-sorted and stored as blocks to accelerate feature selection.

[0105] Approximation algorithm: Approximate continuous features using quantiles to reduce computational load.

[0106] Parallelization: The split gain is computed in parallel along the feature dimension.

[0107] Cache optimization: Use cache-aware access to reduce memory overhead.

[0108] Missing value handling: Automatically learns the splitting direction for missing values.

[0109] S42. The core hyperparameters of the XGBoost model are automatically optimized using the Bayesian optimization algorithm, with the goal of maximizing the classification accuracy on the validation set.

[0110] Automatic optimization of the core hyperparameters of the XGBoost model using the Bayesian optimization algorithm includes: S421. Determine the hyperparameters to be optimized and their search ranges, as shown in Table 5. Select 7 key hyperparameters that have a significant impact on model performance and set reasonable search intervals (taking into account both algorithm stability and search efficiency). Finally, encapsulate these parameters and ranges into a dictionary pbounds.

[0111] Table 5 S422. Define the objective function as the classification accuracy on the validation set; The objective function, named `xgb_evaluate`, is the core of Bayesian optimization. Its input consists of seven hyperparameters defined in `pbounds`, and its output is the classification accuracy of the model on the validation set under that parameter combination. The specific logic is as follows: 1. Parameter reception: The function receives 7 hyperparameters from the Bayesian optimizer (corresponding one-to-one with the keys in pbounds), among which the integer parameters (n_estimators, max_depth) need to be converted from floating-point numbers to integers; 2. Model Construction: Combine the received hyperparameters with fixed XGBoost parameters (such as multi:softprob for multi-class classification, number of classes, random seed, etc.) to construct a temporary XGBoost classifier; 3. Model Training: Train the temporary model using the training set (encoded labels); 4. Performance evaluation: Use the temporary model to predict the validation set and calculate the accuracy of the prediction results against the true labels on the validation set; 5. Result Return: Returns the accuracy as a "score" for the current parameter combination, allowing the optimizer to judge the quality of the parameters.

[0112] S423. Initialize the Bayesian optimizer and perform stochastic exploration and directed iteration to maximize the objective function; Random exploration (init_points=5): First, randomly select 5 sets of hyperparameter combinations (all within the pbounds range), substitute them into xgb_evaluate to calculate the accuracy, and provide initial data for subsequent targeted optimization.

[0113] Directed Iteration (n_iter=25): Based on a Gaussian process, the mapping relationship between hyperparameters and accuracy is modeled. Each iteration performs the following steps: Predict the parameter combination in the pbounds space that is "most likely to yield higher accuracy"; Call xgb_evaluate to verify the actual accuracy of this parameter combination; Update the Gaussian process model to narrow down the search range for optimal parameters.

[0114] A total of 5+25=30 rounds of parameter validation were performed to finally obtain the parameter combination (optimal parameters) that maximized the xgb_evaluate output value (highest accuracy).

[0115] S424. Extract the optimal hyperparameter combination from the optimization results and use it to train the XGBoost model.

[0116] S43. The machine learning early warning model includes a SHAP interpreter. The SHAP interpreter is used to analyze the importance of feature factors in the machine learning early warning model of each waterway and quantify the contribution of feature factors to the risk of insufficient channel size in each waterway.

[0117] S5. Utilize machine learning early warning models to evaluate the scale-based alarm prediction levels for each waterway.

[0118] Step S5 includes: S51. Obtain the current channel width, current shallowest water depth, current water level, weather, ship grounding, and dredging operations corresponding to the execution time of the most recent channel exploration mission. Calculate the upstream predicted average water level, downstream predicted average water level, and upstream predicted average flow rate between the execution time of the most recent channel exploration mission and the time to be predicted. The upstream predicted average water level, downstream predicted average water level, and upstream predicted average flow rate are predicted and calculated using an LSTM model. The LSTM model can be trained and validated using the water level data and flow rate data detected in the n channel exploration missions in step S2. Calculate the predicted average scour and sedimentation amplitude in the shallow area between the execution time of the most recent channel exploration mission and the predicted time: Update the two-dimensional hydro-sediment numerical model using the latest mapping data. Simulate the updated two-dimensional hydro-sediment numerical model using verified flow and sediment parameters to obtain the predicted elevation corresponding to the execution time of the most recent channel exploration mission. Compare the predicted elevation corresponding to the execution time of the most recent channel exploration mission with the actual elevation. If the deviation meets the requirements, simulate the updated two-dimensional hydro-sediment numerical model using verified flow and sediment parameters to obtain the bed elevation corresponding to the execution time of the most recent channel exploration mission and the predicted time. Calculate the predicted average scour and sedimentation amplitude in the shallow area between the execution time of the most recent channel exploration mission and the predicted time based on the bed elevation. If the deviation does not meet the requirements, optimize the verified flow and sediment parameters. S52. The machine learning early warning model outputs a scale alarm prediction level based on the feature factors corresponding to the execution time of the most recent channel exploration mission. The scale alarm prediction level represents the scale alarm level for the predicted time. Figure 12 , 13 As shown.

[0119] The following example uses a specific waterway (data from 2023-04-03 08:22:30) as an example. Input parameters: Channel width (m): 50 Minimum water depth (m): 3.2 Current water level (m): 196.97 Forecasted average water level upstream (m): 223.55 Downstream predicted average water level (m): 195.77 Upstream predicted average flow: 4790 Predicted average scour and sedimentation amplitude in shallow areas: -0.036. Weather: 0 Ship grounded: 0 Dredging work: 0, See the output results Figure 12 and Figure 13 .

[0120] From the performance evaluation of the XGBoost model for this waterway, the ROC curve (as shown in the figure) Figure 14As shown in the image, the XGBoost model for this waterway exhibits varying discrimination capabilities across different states. The model performs exceptionally well in the "normal" state (AUC=0.95), "scale alarm" (AUC=1.00), and "Level 1 warning" (AUC=1.00) categories, demonstrating near-perfect discrimination capabilities. However, the results for "Level 2 warning" (AUC=0.86) and "Level 3 warning" (AUC=0.94) indicate that the model's discrimination capability for Level 2 warnings is lower compared to other types, but overall it possesses a reliable recognition foundation.

[0121] Based on the analysis results of the waterway confusion matrix (such as...) Figure 15 As shown in the figure, the model correctly predicted 26 samples out of 30 validation set samples, achieving an overall accuracy of 86.67%. Specific analysis indicates that the model performs best in identifying the "normal" state and the "scale alarm" category. Only one instance of misclassification occurred in the "normal" state, falling under the category of "Level 3 Warning," while all "scale alarm" samples were correctly predicted. However, the model exhibits some misclassification in warning category discrimination, primarily between "Level 1 Warning" and "Normal" states, and between "Level 2 Warning" and "Level 3 Warning." These results indicate that while the model's overall performance is good, it faces challenges in distinguishing warning categories with similar risk levels. Further improvements in sample representativeness could enhance the model's accuracy in classifying warning categories.

[0122] Based on SHAP feature importance analysis, such as Figure 16 As shown, in the XGBoost model's prediction of this waterway, the "upstream predicted average water level" has the greatest impact on the model output, significantly higher than other features, indicating its dominant role in waterway condition prediction. Following closely is "dredging operations," which is of high importance, demonstrating the crucial role of human intervention in prediction; the "upstream predicted average flow rate" also has a significant impact on the model, indicating that flow rate has a strong influence on waterway condition prediction. Next, the "shallow area predicted average scouring and silting amplitude" and "downstream predicted average water level" have smaller impacts on the model output, but still contribute to the prediction results. In contrast, "weather" and "ship grounding" have the smallest impact on the model, showing that these factors have a weak influence on waterway prediction. Overall, the model's prediction decisions mainly rely on water level features, especially in terms of upstream water levels, with the degree of influence of each feature factor varying significantly across different warning categories.

[0123] The present invention also provides a system for implementing the above-mentioned method for risk assessment and early warning of insufficient waterway dimensions, comprising: The data acquisition module is used to collect mapping and detection data from various waterways of the waterway; The water and sediment model construction and calculation module is used to construct and verify the water and sediment numerical models of each waterway based on the mapping data and exploration data; and to calculate the average scouring and silting amplitude prediction for each waterway exploration mission at the corresponding time based on the water and sediment numerical models. The feature vector construction module is used to construct the feature vector set corresponding to the execution time of each channel exploration mission in each waterway; The model training and validation module is used to divide the multiple feature vector sets of each waterway into training and validation sets, and to train and validate the machine learning early warning models for each waterway. The risk assessment module is used to evaluate the scale alarm level of each waterway using the machine learning early warning model.

[0124] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for risk assessment and early warning of insufficient waterway dimensions, characterized in that: include: Collect mapping and survey data of each waterway in the waterway; Based on mapping and detection data, numerical models of water and sediment in each waterway were constructed and verified. The detection data includes data from n channel exploration missions, and the average scouring and deposition amplitude prediction for each channel exploration mission at the corresponding time is calculated based on the water and sediment numerical model. Based on the detection data and the average scour and siltation amplitude prediction, multiple feature vector sets are constructed for each waterway. Each feature vector set includes multiple feature factors. Each feature vector set is matched with the feature factors corresponding to the execution time of each waterway detection mission. The feature factors include waterway width, water depth, water level, flow rate, weather, waterway anomalies, average scour and siltation amplitude prediction, and scale alarm prediction level. The multiple feature vector sets of each waterway are divided into training set and validation set to train and validate the machine learning early warning model of each waterway. The scale-based alarm prediction level of each waterway is evaluated using a machine learning early warning model.

2. The method for risk assessment and early warning of insufficient waterway dimensions according to claim 1, characterized in that: Constructing and validating numerical models of water and sediment in each waterway includes: A two-dimensional water and sediment numerical model was constructed based on the measured river topographic map in the mapping data, and the grid was divided. The flow parameters of the two-dimensional water and sediment numerical model were calibrated and verified based on the detection data. The sediment parameters of the two-dimensional water and sediment numerical model were verified based on the detection data.

3. The method for risk assessment and early warning of insufficient waterway dimensions according to claim 2, characterized in that: The predicted average scouring and deposition amplitude for each channel exploration mission, calculated based on the water and sediment numerical model, includes: Calculate the predicted average scouring and silting amplitude for the time corresponding to the i-th channel exploration mission: The execution time of the i-th channel exploration mission is t. i Using time t i The latest mapping data is used to update the two-dimensional hydro-sediment numerical model. The updated two-dimensional hydro-sediment numerical model is then used to simulate the flow and sediment parameters, and the time t is obtained. i Predicted elevation; Time t i The predicted elevation is compared with the actual elevation measured by the i-th channel exploration mission. If the deviation meets the requirements, the updated two-dimensional water and sediment numerical model is used to simulate the flow and sediment parameters after verification, and the time t is obtained. i and t i+1 The corresponding bed surface elevation is used to calculate time t. i+1 Relative to t i The average scouring and silting amplitude is used as the prediction of the average scouring and silting amplitude for the corresponding time of the i-th channel exploration mission. If the deviation does not meet the requirements, the verified flow parameters and sediment parameters will be optimized. Where 1≤i≤n-1, t i+1 The execution time of the (i+1)th channel exploration mission.

4. The method for risk assessment and early warning of insufficient waterway dimensions according to claim 3, characterized in that: If the deviation does not meet the requirements, a smart algorithm based on the Pareto optimal concept will be used to optimize the verified flow parameters and sediment parameters. The updated two-dimensional hydro-sediment numerical model is used to simulate the flow and sediment parameters, and the time t is obtained. i Predicted elevation; Time t i The predicted elevation is compared with the actual elevation measured in the i-th channel exploration mission. If the deviation still does not meet the requirements, the Pareto optimal concept intelligent algorithm is used again to optimize the flow parameters and sediment parameters until the deviation meets the requirements.

5. The method for risk assessment and early warning of insufficient waterway dimensions according to claim 1, characterized in that: The multiple feature factors include real-time indicators, statistical indicators, and predictive indicators. The real-time indicators corresponding to the execution time of the i-th channel exploration mission include the current channel width, current shallowest water depth, current water level, weather, ship grounding, and dredging operations at the time corresponding to the i-th channel exploration mission. The statistical indicators corresponding to the execution time of the i-th channel exploration mission include the upstream predicted average water level, downstream predicted average water level, upstream predicted average flow, and shallow area predicted average scouring and silting amplitude during the period from the i-th channel exploration mission to the (i+1)-th channel exploration mission. The upstream predicted average water level is the average of the historical water level data of the upstream water level station during the period from the i-th waterway exploration mission to the (i+1)-th waterway exploration mission; The downstream predicted average water level is the average of the historical water level data of the downstream water level station during the period from the i-th waterway exploration mission to the (i+1)-th waterway exploration mission; The upstream predicted average flow is the average of the historical flow data of the upstream flow station during the period from the i-th channel exploration mission to the (i+1)-th channel exploration mission; The predicted average scour and sedimentation amplitude in shallow areas represents the average thickness of scour or sedimentation on the shallow bed surface within the time range from the i-th channel exploration mission to the (i+1)-th channel exploration mission. The predicted average scour and sedimentation amplitude in shallow areas is calculated based on the water and sediment numerical model of the waterway. The prediction indicators corresponding to the execution time of the i-th channel exploration mission include the scale alarm prediction level, which is the scale alarm level obtained based on the exploration situation of the (i+1)-th channel exploration mission.

6. The method for risk assessment and early warning of insufficient waterway dimensions according to any one of claims 1 to 5, characterized in that: Training and validating machine learning early warning models for each waterway includes: The XGBoost model, LightGBM model, and SVM model were trained using the training set, and tested using the validation set. The accuracy of the XGBoost model, LightGBM model, and SVM model for the prediction results of each waterway was calculated, and one of the models was selected as the machine learning early warning model for each waterway based on the accuracy.

7. The method for risk assessment and early warning of insufficient waterway dimensions according to any one of claims 1 to 5, characterized in that: Training and validating machine learning early warning models for each waterway includes: The XGBoost model was used as the machine learning early warning model for each waterway. The core hyperparameters of the XGBoost model are automatically optimized using a Bayesian optimization algorithm, with the goal of maximizing the classification accuracy on the validation set.

8. The method for risk assessment and early warning of insufficient waterway dimensions according to claim 7, characterized in that: Automatic optimization of the core hyperparameters of the XGBoost model using the Bayesian optimization algorithm includes: Determine the hyperparameters to be optimized and their search range, including n_estimators, max_depth, learning_rate, subsample, colsample_bytree, reg_alpha, and reg_lambda; Define the objective function as the classification accuracy on the validation set; Initialize the Bayesian optimizer and perform stochastic exploration and directed iteration to maximize the objective function; The optimal hyperparameter combination is extracted from the optimization results and used to train the XGBoost model.

9. The method for risk assessment and early warning of insufficient waterway dimensions according to any one of claims 1 to 5, characterized in that: The evaluation of the scale-based alarm prediction level for each waterway using a machine learning early warning model includes: Obtain the current channel width, current shallowest water depth, current water level, weather, ship grounding, and dredging operations corresponding to the execution time of the most recent channel survey mission; Calculate the upstream predicted average water level, downstream predicted average water level, and upstream predicted average flow rate between the execution time of the most recent channel exploration mission and the time to be predicted. Calculate the predicted average scour and siltation amplitude in the shallow area between the execution time of the most recent channel exploration mission and the time to be predicted; The machine learning early warning model outputs the scale alarm prediction level based on the feature factors corresponding to the execution time of the most recent channel exploration mission. The scale alarm prediction level represents the scale alarm level at the time to be predicted.

10. A system for implementing the method for risk assessment and early warning of insufficient waterway dimensions as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect mapping and detection data from various waterways of the waterway; The water and sediment model construction and calculation module is used to construct and verify the water and sediment numerical models of each waterway based on the mapping data and exploration data; and to calculate the average scouring and silting amplitude prediction for each waterway exploration mission at the corresponding time based on the water and sediment numerical models. The feature vector construction module is used to construct the feature vector set corresponding to the execution time of each channel exploration mission in each waterway; The model training and validation module is used to divide the multiple feature vector sets of each waterway into training and validation sets, and to train and validate the machine learning early warning models for each waterway. The risk assessment module is used to evaluate the scale alarm level of each waterway using the machine learning early warning model.

Images