A method and system for predicting tidal flat evolution based on a mixture model

Abstract

The application provides a tidal flat evolution prediction method and system based on a hybrid model, relates to the cross technical field of dynamic geomorphology and artificial intelligence, and comprises the following steps: a tidal flat erosion and deposition evolution database under different scenarios of a target coastal zone is constructed; a geomorphology prediction model is constructed based on the database through an artificial intelligence algorithm, water dynamic characteristic data is taken as input, and seabed erosion and deposition change is taken as output; hybrid simulation is performed based on a physical mechanism model and the geomorphology prediction model, long-time evolution prediction is realized by cyclic iteration and update of terrain boundary conditions; and a business application module is configured, an interactive interface is used to receive user input parameters, and visualized results are output. The application decouples traditional water dynamic-sediment-geomorphology coupling calculation into a chain process of cooperation of a physical mechanism and AI, significantly reduces calculation complexity, realizes off-line rapid calculation of tidal flat evolution, and provides real-time technical support for coastal zone planning and disaster prevention decision-making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dynamic geomorphology and artificial intelligence, in particular to a tidal flat evolution prediction method and system based on a hybrid model. BACKGROUND

[0002] As a key transitional zone of land-sea interaction, the tidal flat plays an irreplaceable role in maintaining the balance of the coastal ecosystem and providing disaster prevention and mitigation functions. Under the dual pressures of global climate change and high-intensity human activities, the evolution mechanism of muddy tidal flats tends to be complex, and it is urgent to establish a precise and efficient prediction system to support the sustainable development of the coastal zone. Traditional prediction methods based on dynamic geomorphology have inherent defects such as high computational complexity and large resource consumption due to the need to simulate the coupling of multiple physical fields. Especially when conducting multi-scenario simulation, the traditional numerical model is time-consuming and difficult to meet the timeliness requirements of coastal zone spatial planning and disaster prevention decision-making. In addition, factors such as storm surges caused by extreme climate events and seasonal water level fluctuations further increase the risk of systematic errors in prediction results.

[0003] Therefore, breaking through the computing power bottleneck of traditional models and building a new prediction paradigm that takes into account physical mechanisms and computational efficiency has become a core challenge for fine-grained management of the coastal zone. SUMMARY

[0004] To solve the above problems existing in the prior art, the first aspect of the present application proposes a tidal flat evolution prediction method based on a hybrid model, comprising:

[0005] S1: Construct a tidal flat erosion and deposition evolution database of the target coastal zone under different scenarios;

[0006] S2: Based on the tidal flat erosion and deposition evolution database, construct a geomorphology prediction model through an artificial intelligence algorithm;

[0007] S3: Perform hybrid simulation based on the physical mechanism model and the geomorphology prediction model, comprising:

[0008] Generate water dynamic feature data through the physical mechanism model, input the water dynamic feature data into the geomorphology prediction model to generate seabed erosion and deposition changes, update the topographic boundary conditions based on the seabed erosion and deposition changes and iterate cyclically;

[0009] S4: Configure a business application module based on the hybrid simulation, and the business application module receives user input parameters, performs geomorphology speed calculation and outputs visual results.

[0010] In combination with the first aspect, in some implementations, S1 comprises:

[0011] S11: Establish a dynamic geomorphology evolution model of the target coastal zone using a dynamic geomorphology model;

[0012] S12: Based on the dynamic geomorphological evolution model, the dynamic geomorphological evolution model is calibrated by reproducing the dynamic deposition process of the tidal flat and combining historical topographic data;

[0013] S13: Based on the calibrated dynamic geomorphological evolution model, the storm surge and seasonal water level fluctuation are time-scaled to generate annual geomorphological evolution feature data set;

[0014] S14: Based on the annual geomorphological evolution feature data set, the influence of sediment supply, sea level change, tide and wave factors on the erosion and deposition evolution of the tidal flat is simulated to form a tidal flat erosion and deposition evolution database.

[0015] In combination with the first aspect, in some implementations, S12 includes: based on the dynamic geomorphological evolution model, long-term simulation is performed to output geomorphological data containing the influence of climate change and human activities;

[0016] S13 includes: based on the geomorphological data containing the influence of climate change and human activities, the characteristic parameters of storm surge and seasonal water level fluctuation are extracted for time-scale reduction;

[0017] S14 includes: based on the characteristic parameters after time-scale reduction, the contribution value of coastal engineering to geomorphological evolution is quantified.

[0018] In combination with the first aspect, in some implementations, S2 includes:

[0019] S21: Based on the tidal flat erosion and deposition evolution database, the inundation probability, average shear stress, rising tide advantage and tidal creek distance function are sorted out to construct a training data set corresponding to the seabed erosion and deposition;

[0020] S22: Based on the training data set, a convolutional neural network model is constructed with four-channel hydrodynamic features as input and seabed erosion and deposition change as output;

[0021] S23: Based on the convolutional neural network model, an encoder-decoder structure is configured and feature map cross-layer transmission is realized through jump connection to output an optimized convolutional neural network model;

[0022] S24: Based on the optimized convolutional neural network model and the tidal creek distance function, the loss function is calculated by regional weighting, and an accuracy-optimized geomorphological prediction model is output.

[0023] In combination with the first aspect, in some implementations, S23 includes: the feature maps output by each stage of the encoder are transmitted to the corresponding stage of the decoder for splicing;

[0024] S24 includes: according to the tidal creek distance function value, a weight coefficient is assigned, and the weight value of the tidal creek edge area is higher than that of the flat area.

[0025] In combination with the first aspect, in some implementations, S3 includes:

[0026] S31: Based on the initial terrain boundary conditions, the inundation probability, average shear stress, tidal dominance, and tidal channel distance function are calculated and generated through a physical mechanism model.

[0027] S32: Input the inundation probability, average shear stress, tidal dominance, and tidal channel distance function into the geomorphological prediction model, and output the seabed erosion and deposition changes for the current period;

[0028] S33: Update the topographic boundary conditions based on the seabed erosion and deposition changes in the current time period, and use the updated topographic boundary conditions as the input for the next time period. Repeat S31 to S32 until the long-duration simulation is completed.

[0029] In conjunction with the first aspect, in some implementations, S31 includes: integrating sea level rise parameters into the initial topographic boundary conditions;

[0030] S33 includes: dividing the long-duration simulation into multiple discrete time periods, with the output terrain boundary conditions of each time period serving as the input terrain boundary conditions for the next time period.

[0031] In conjunction with the first aspect, in some implementations, S4 includes:

[0032] S41: Receives coastal zone development parameters, climate parameters, and hydrodynamic parameters input by the user through an interactive interface;

[0033] S42: Based on coastal zone development parameters, climate parameters, and hydrodynamic parameters, call the hybrid simulation to perform rapid geomorphological calculations;

[0034] S43: Converts the results of terrain quick calculation into visual graphic output, and provides parameter reset, calculation trigger and history management functions.

[0035] In conjunction with the first aspect, in some implementation methods, S41 includes: verifying the threshold range of coastal zone development parameters, climate parameters and hydrodynamic parameters, and triggering warning information when they exceed their respective threshold ranges;

[0036] S43 includes: overlaying seabed erosion and deposition changes onto a geographic base map in the form of a heat map, and storing the input parameters for each calculation and the corresponding erosion and deposition evolution results.

[0037] Secondly, the present invention provides a tidal flat evolution prediction system based on a hybrid model. The system employs the method provided in any of the above embodiments, and the system includes:

[0038] The first construction module is used to build a database of tidal flat erosion and deposition evolution of the target coastal zone under different scenarios;

[0039] The second construction module is connected with the first construction module, and is used for constructing a geomorphology prediction model based on a tidal flat erosion and deposition evolution database through an artificial intelligence algorithm;

[0040] The mixed simulation execution module is connected with the second construction module, and is used for executing mixed simulation based on the physical mechanism model and the geomorphology prediction model, including:

[0041] The water dynamic characteristic data is generated through the physical mechanism model, the water dynamic characteristic data is input into the geomorphology prediction model to generate a seabed erosion and deposition change amount, and the topographic boundary condition is updated based on the seabed erosion and deposition change amount and is iterated cyclically;

[0042] The geomorphology rapid calculation execution and output module is connected with the mixed simulation execution module, and is used for configuring a business application module based on the mixed simulation, the business application module receives user input parameters, executes geomorphology rapid calculation, and outputs visualized results.

[0043] Compared with the prior art, the beneficial effects of the present application are as follows: first, a tidal flat erosion and deposition evolution database of the target coastal zone under different scenarios is constructed, which provides a physical mechanism data basis covering multiple factors for subsequent model training, and ensures the completeness of the prediction system. Secondly, based on the database, an artificial intelligence algorithm is used to construct a geomorphology prediction model, which decouples the traditional strong coupling calculation of water dynamics-sediment-geomorphology into a chain calculation of "water dynamics (sediment)-geomorphology", significantly reducing the model complexity. This process directly establishes the mapping relationship between the characteristic conditions and the geomorphology evolution, which makes the computing power demand of large-scale long-time simulation decrease by orders of magnitude.

[0044] In the mixed simulation stage, the physical mechanism model and the geomorphology prediction model form a closed loop cooperation: the physical mechanism model generates four-channel water dynamic characteristic data such as submergence probability and average shear stress, which is input into the geomorphology prediction model to quickly output high-precision seabed erosion and deposition change amount; based on the change amount, the topographic boundary condition is updated in real time and iterated cyclically, which not only inherits the accuracy of the dynamic response of the boundary of the physical mechanism, but also avoids the massive consumption of the whole coupling calculation. This "physical driving-AI rapid calculation" cycle architecture realizes the stable calculation of long-period geomorphology evolution under extreme climate events for the first time.

[0045] Finally, the business application module converts the mixed simulation capability into a business tool. After the user inputs scenario data such as coastal zone development parameters and climate parameters through the interactive interface, the system directly calls the pre-trained mixed model chain, and outputs visualized tidal flat evolution prediction results in minutes without the need for professional numerical simulation skills. This module compresses the traditional dynamic geomorphology simulation which takes several weeks into real-time rapid calculation level, completely solving the timeliness bottleneck of coastal zone management decision-making, and avoiding human operation errors through standardized operation processes.

[0046] In summary, through a four-step collaborative approach of "database construction - AI modeling - hybrid iteration - business integration", the system overcomes the limitations of traditional models in terms of computing power and timeliness while maintaining the reliability of physical mechanisms, providing real-time technical support for coastal spatial planning and disaster prevention and control. Attached Figure Description

[0047] 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.

[0048] Figure 1 The diagram shown is a flowchart of a tidal flat evolution prediction method based on a hybrid model provided by an embodiment of the present invention.

[0049] Figure 2 The image shown is a seabed erosion and deposition map of a target sea area provided by an embodiment of the present invention.

[0050] Figure 3 The diagram shown is a flowchart of the U-Net Mor calculation process provided in an embodiment of the present invention.

[0051] Figure 4 The diagram shown is a result of a hybrid simulation flowchart provided in an embodiment of the present invention.

[0052] Figure 5 The figure shown is a schematic diagram of an interactive interface provided in an embodiment of the present invention.

[0053] Figure 6 The diagram shown is a structural schematic of a tidal flat evolution prediction system based on a hybrid model provided in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0055] The specific embodiments of the present invention will be described below.

[0056] Example 1

[0057] like Figure 1 As shown, this invention proposes a method for predicting tidal flat evolution based on a hybrid model, comprising:

[0058] S1: Construct a database of tidal flat erosion and deposition evolution of the target coastal zone under different scenarios;

[0059] S2: Based on the tidal flat erosion and deposition evolution database, a geomorphology prediction model is constructed through an artificial intelligence algorithm;

[0060] S3: Perform hybrid simulation based on the physical mechanism model and the geomorphology prediction model, including:

[0061] Generate hydrodynamic feature data through the physical mechanism model, input the hydrodynamic feature data into the geomorphology prediction model to generate seabed erosion and deposition changes, update the topographic boundary conditions based on the seabed erosion and deposition changes, and iterate cyclically;

[0062] S4: Configure a business application module based on the hybrid simulation, the business application module receives user input parameters, performs geomorphology calculation, and outputs visual results.

[0063] Specifically, when constructing a database of tidal flat erosion and deposition evolution of the target coastal zone under different scenarios, first, a dynamic geomorphology model needs to be selected, and a numerical grid is established for the target area and the resolution of the tidal flat is improved. Taking the Caofedian sea area in Bohai Bay as an example, the outer grid is set to 0.9 km x 0.9 km, and the near-island area is encrypted to 0.11 km x 0.11 km. By re-enacting the historical dynamic sedimentation process, the model is calibrated in combination with the topographic DEM data to ensure the reliability of the physical mechanism. To eliminate systematic errors in extreme climate event prediction, a wave spectrum and mathematical analysis combined method is used to process multi-year wave data: the waves are divided into 64 classes (8 wave heights x 8 wave directions) at equal intervals, the representative wave height is calculated by wave energy equivalence, the wave direction uses weighted values, and each wave action duration is set to 1 day and arranged in a random sequence to eliminate the influence of tidal level. At the same time, a geomorphology acceleration factor (recommended value ≤ 50) is introduced to amplify the wave flow sand lifting effect, and the storm surge and seasonal water level fluctuation are reduced to the annual scale, directly establishing the mapping relationship between the annual scale evolution process and the characteristic conditions. Finally, the complex influence of multiple factors such as sediment supply, sea level rise, tidal wave and coastal engineering on erosion and deposition is simulated to form a database covering multiple scenarios.

[0064] When building a geomorphology prediction model based on the database, four types of characteristic parameters, including inundation probability, average shear stress, rising tide advantage, and tidal creek distance function (SDF), are arranged as inputs, and seabed erosion and deposition volume is arranged as output. The data set is divided into training set, validation set and test set according to the ratio of 7:1.5:1.5. U-Net convolutional neural network framework is selected to build U-Net Mor model, the input layer is designed as four channels, the convolution kernel size is set to 5x5, ReLU activation function is used to introduce nonlinearity, and the encoder-decoder is connected through a jump to transfer feature maps to preserve details. In the loss function calculation, in addition to the mean square error to measure the overall accuracy, the tidal creek edge area is given higher weight according to the SDF value to enhance the key geomorphology prediction capability.

[0065] In the mixed model scenario simulation stage, the sea level rise parameter is integrated into the water level boundary condition, and the long-term simulation is divided into N discrete time periods. A physical mechanism model (such as Delft3D) first generates four-channel hydrodynamic feature data for the i-th time period; a geomorphology prediction model (U-Net Mor) outputs seabed erosion and deposition amounts and updates the topography accordingly; the updated topography is input into the next time period, and the iteration is repeated until the N period simulation is completed, and the cumulative result is the long-term erosion and deposition evolution. This process decouples the traditional coupling calculation into a chain cycle of "physical mechanism generates input-AI outputs topography updates", which significantly reduces the computational load of a single calculation.

[0066] The business application module receives the user input of the coastal development parameters, climate parameters, and hydrodynamic parameters through the interactive interface, calls the pre-trained mixed model chain to perform rapid calculation. The interface integrates parameter verification function: when the input value exceeds the threshold range, an alarm is triggered; the calculation result is output in the form of a heat map superimposed on the geographic base map, and parameter reset, historical record storage and clearing functions are provided. The user only needs to select the scenario and click the calculation button to obtain the prediction result completed by the traditional model in weeks.

[0067] In the embodiments of the present application, the tidal flat erosion and deposition evolution database provides a physically complete multi-scenario data basis for AI training; the geomorphology prediction model decouples the strong coupling process into a feature mapping relationship, reducing the computational power requirement; the mixed simulation cycle architecture inherits the accuracy of the physical mechanism while avoiding the full coupling calculation consumption, and for the first time realizes long-period stable calculation including extreme events; the business application module converts professional numerical simulation into zero threshold operation, significantly reducing the decision response time. Finally, the core problems of insufficient timeliness and computational power bottleneck in coastal planning are solved.

[0068] In combination with the first aspect, in some implementations, S1 includes:

[0069] S11: using a dynamic geomorphology model to establish a dynamic geomorphology evolution model of the target coastal zone;

[0070] S12: based on the dynamic geomorphology evolution model, reenacting the dynamic deposition process of the tidal flat, and calibrating the dynamic geomorphology evolution model in combination with historical topographic data;

[0071] S13: based on the calibrated dynamic geomorphology evolution model, performing time scale reduction on storm surges and seasonal water level fluctuations to generate an annual scale geomorphology evolution feature data set;

[0072] S14: based on the annual scale geomorphology evolution feature data set, simulating the influence of sediment supply, sea level change, tides, and wave factors on erosion and deposition evolution, and forming a tidal flat erosion and deposition evolution database.

[0073] Specifically, the database construction of the tidal flat erosion and deposition evolution includes four sub-steps:

[0074] In S11, the dynamic geomorphology evolution model selection needs to consider the characteristics of the target area, for example, the muddy coast is suitable for Delft3D or XBeach with a perfect deposition module. The grid division needs to follow the principle of “coarse in the periphery and dense in the tidal flat”, and in the case of Caofadian, the grid near the island is encrypted to 10m×10m to capture the geomorphology of the tidal inlet. In S12, the model calibration needs to combine historical topographic data, for example, comparing DEMs of different years to verify the reproduction accuracy of the erosion and deposition trend. The long-term simulation length should cover the main environmental change period, for example, 50 years of simulation is adopted for Caofadian to include the cumulative effect of sea level rise.

[0075] The time scale reduction in S13 is the key to eliminate the error of extreme events: for the intermittent characteristics of storm surge, the short-time influence is removed by wave classification and random sequence arrangement; for seasonal water level fluctuation, the combination of spring tide and seasonal tide is used for simulation, and then the geomorphology acceleration factor is used to integrate the high-frequency fluctuation into the annual scale output.

[0076] Alternative solutions include using wavelet analysis to extract periodic components, or decomposing dominant modes through empirical orthogonal function.

[0077] In S14, the scenario simulation needs to cover the core variables of coastal zone management: the sediment supply scenario sets different sources (river sediment transport / beach erosion) and flux gradients, the sea level rise rate prediction range of low / medium / high scheme, the tidal process considers the change of tidal range and the modulation of astronomical tide period, the wave condition combines historical statistics and future wind field prediction, and the impact of coastal engineering is realized by modifying the boundary conditions (such as the coordinates of the breakwater). The final integration of the database is a multi-dimensional parameter matrix, and each parameter combination corresponds to a set of four-channel feature data and erosion and deposition results.

[0078] In the embodiments of the present application, the dynamic geomorphology evolution model calibration ensures the reliability of the physical mechanism; the time scale reduction suppresses the divergence of AI prediction caused by extreme events; and the multi-scenario coverage enables the database to have the ability of extrapolation prediction. For example, in the case of Caofadian, the quantitative relationship between the sea level rise rate and the erosion and deposition amount can directly support the coastal protection decision under different climate policies.

[0079] In combination with the first aspect, in some implementations, S12 includes: performing long-term simulation based on the dynamic geomorphology evolution model to output geomorphology data containing the influence of climate change and human activities;

[0080] S13 includes: extracting storm surge and seasonal water level fluctuation characteristic parameters for time scale reduction based on the geomorphology data containing the influence of climate change and human activities;

[0081] S14 includes: quantifying the contribution value of coastal engineering to geomorphology evolution based on the characteristic parameters after time scale reduction.

[0082] Specifically, the long-term simulation of S12 needs to set a sufficient time span to capture the slow-changing process, for example, a 50-year simulation can include the linear trend of sea level rise and the stepwise impact of human activities (reclamation / sand mining). Climate change factors are realized by modifying the boundary conditions: temperature rise can adjust the water body's viscosity coefficient, and the increase in storm frequency is reflected in the shift of wave spectrum energy distribution. The quantification of human activities needs to combine engineering parameters: reclamation area is converted into a grid mask, and port construction is realized by modifying the local water depth.

[0083] The feature parameter extraction of S13 needs to focus on the core indicators of storm surge and seasonal fluctuations: storm surge uses three-dimensional parameters of maximum water level increase, duration, and frequency; seasonal fluctuations are characterized by tidal level standard deviation and phase angle. In addition to the wave classification method, the event response model can be used to convert discrete storm events into equivalent continuous forces.

[0084] The contribution value quantification of coastal engineering of S14 needs to design a control experiment:

[0085] Baseline scenario: natural evolution without engineering intervention;

[0086] Engineering scenario: add boundary conditions such as embankment / siltation promoting facilities;

[0087] Contribution value = (engineering scenario erosion and deposition amount - baseline scenario erosion and deposition amount) / simulation duration.

[0088] For example, in the Caofidian case, the embankment project led to an increase of 0.12 m in the annual average deposition downstream, which can be directly input into the AI model training.

[0089] As shown in Figure 2 , for the impact of sea level rise on geomorphic evolution, a coastal geomorphic evolution database is constructed, different 50-year sea level scenarios are set, and the geomorphic evolution of the coastal tidal flat under the change of sea level rise rate is calculated.

[0090] Long-term simulation combined with engineering quantification brings three effects: revealing the cumulative geomorphic impact of human activities (such as the lag effect of reclamation leading to the shrinkage of tidal flat); separating the contribution proportion of natural evolution and human disturbance; providing the "dose-response" data set of engineering intervention for AI model. For example, the database contains the corresponding relationship between different embankment lengths and deposition amounts, supporting the comparison and selection of planning schemes.

[0091] In combination with the first aspect, in some implementations, S2 includes:

[0092] S21: Based on the tidal flat erosion and deposition evolution database, organize the submergence probability, average shear stress, flood dominance, and tidal creek distance function, and construct a training data set corresponding to the sea bed erosion and deposition;

[0093] S22: Based on the training data set, a convolutional neural network model is constructed with four-channel hydrodynamic features as input and seabed erosion and deposition change as output;

[0094] S23: Based on the convolutional neural network model, an encoder-decoder structure is configured and feature map cross-layer transmission is realized through a jump connection, and an optimized convolutional neural network model is output;

[0095] S24: Based on the optimized convolutional neural network model and the tidal creek distance function, the loss function is calculated by region weighting, and a precision optimized topographic prediction model is output.

[0096] Specifically, the training data set construction needs to strictly unify the data format: the submergence probability is a floating point matrix of the submergence time ratio of the grid unit; the average shear stress is the average value of the bed shear force; the rising tide advantage is defined as the ratio of the rising and falling tide duration difference to the period; the tidal creek distance function (SDF) is calculated by the Euclidean distance transformation to calculate the distance from the grid point to the nearest tidal creek center. Four-channel data needs to be spatially aligned and normalized to the [0, 1] interval.

[0097] As shown in Figure 3 The U-Net Mor model structure optimization focuses on three points: the encoder uses four levels of down-sampling, and the number of channels is multiplied by 8→16→32→64 at each level; the decoder is up-sampled by deconvolution, and the same size feature map of the encoder is jump-connected; the final output layer uses 1×1 convolution to compress the number of channels to 1 (erosion and deposition amount).

[0098] Alternative solutions include replacing the basic convolution block with ResNet to enhance gradient propagation, or using attention mechanism to optimize feature fusion.

[0099] The loss function weighting method divides the weight area according to the SDF value: SDF≤50m is defined as the edge of the tidal creek, 50m<SDF≤200m is the transition zone, and SDF>200m is the flat area.

[0100] In the embodiment of the present application, the four-channel input covers the core physical quantities of hydrodynamic control topography (submergence probability controls sedimentation range, shear stress determines sediment starting, rising tide advantage affects sediment transport direction, SDF represents topographic pattern); the jump connection solves the spatial information loss caused by pooling, and the prediction error of the tidal creek shape is reduced; the regionally weighted loss function makes the model focus on the sensitive area of the topography, and the prediction accuracy of the beach and trough boundary is improved.

[0101] In combination with the first aspect, in some implementations, S23 includes: passing the feature maps output by each stage of the encoder to the corresponding stage of the decoder for splicing;

[0102] S24 includes: assigning a weight coefficient according to the tidal creek distance function value, and assigning a higher weight value to the edge area of the tidal creek than to the flat area.

[0103] Specifically, the feature passing from the encoder to the decoder needs to maintain spatial dimension alignment: when the size of the feature map output by the k-th stage of the encoder is reduced due to the pooling operation, the input size of the corresponding stage of the decoder needs to be matched by center cropping or deconvolution up-sampling before the skip connection. In the Caofedian case, the cropping method is used to ensure that the size deviation is less than one percent when splicing. A 3x3 convolution operation is immediately performed after the feature map splicing to compress the number of merged channels to the original number of channels of the decoder, avoiding subsequent calculation redundancy. Alternative solutions include using a channel attention mechanism to weight and filter the encoder feature map, or using element-wise addition instead of channel splicing to reduce memory usage.

[0104] The tidal creek edge weight distribution is based on the spatial gradient characteristics of the SDF: first, the gradient modulus of the SDF field is calculated to quantify the degree of topographic abruptness, and the larger the gradient modulus represents the closer to the tidal creek boundary. The edge region is defined as the grid cell whose gradient modulus exceeds a preset threshold, and the threshold is recommended to be more than the 90th percentile value of the gradient distribution of the entire calculation domain.

[0105] This mechanism solves the core contradiction in geomorphological evolution prediction through spatially differentiated weights: the edge of the tidal creek is subject to strong hydrodynamic action, resulting in rapid erosion and deposition changes, and high weights force the model to prioritize optimizing parameters in these areas during backpropagation, preventing key geomorphic features from being submerged in overall errors; the flat area has a slow and spatially homogeneous erosion and deposition process, and the base weight can prevent the model from overfitting to weak noise. This method significantly improves the sensitivity of micro-topography evolution, such as the prediction accuracy of downstream deposition hotspots caused by jetty engineering.

[0106] Reference Figure 4 In combination with the first aspect, in some implementations, S3 includes:

[0107] S31: Based on the initial topographic boundary conditions, generate the submergence probability, mean shear stress, flood dominance, and SDF through the physical mechanism model;

[0108] S32: Input the submergence probability, mean shear stress, flood dominance, and SDF into the geomorphological prediction model to output the seabed erosion and deposition change in the current period;

[0109] S33: Update the topographic boundary conditions based on the seabed erosion and deposition change in the current period, and use the updated topographic boundary conditions as the input for the next period, and repeat S31 to S32 until the long-term simulation is completed.

[0110] Specifically, the hybrid simulation cycle architecture needs to clarify the division of labor between the physical mechanism model and the geomorphology prediction model: the physical mechanism model (such as Delft3D) is responsible for generating four-channel hydrodynamic feature data - the inundation probability is obtained by the ratio of the inundation time length of the statistical grid cell to the water level fluctuation period; the average shear stress is based on the spatial average of the flow field calculation results; the rising tide advantage is quantified by the ratio of the main tidal component and the shallow water tidal component; the tidal creek distance function (SDF) uses the Euclidean distance transformation algorithm to calculate the distance from each grid point to the nearest tidal creek center. These data need to be unified in spatial resolution and normalized before inputting into the geomorphology prediction model.

[0111] The long-term simulation splitting strategy follows the time scale characteristics of geomorphic evolution: the total time (such as 50 years) is divided into N equal or unequal time periods, and the division is based on the stability of the erosion and deposition rate, the period of external forcing change, etc. In each time period, the physical mechanism model is run based on the current topographic boundary conditions, and four-channel data is output; the geomorphology prediction model generates the seabed erosion and deposition change amount based on this, and updates the topographic boundary conditions through the "original terrain ± erosion and deposition amount" formula. The updated topography is used as the input of the physical mechanism model of the next time period, and the iteration is repeated until all N time periods are covered. The Caofidian case is divided into 10 five-year periods, and the cumulative erosion and deposition amount of each period is obtained as the long-term evolution result.

[0112] This cycle design creates a double technical effect: the physical mechanism model ensures the physical reality of the generation of hydrodynamic features, especially the response to boundary condition changes (such as the integration of sea level rise into the water level boundary) completely follows the principles of fluid mechanics; the geomorphology prediction model replaces the traditional time-consuming sediment transport calculation, reducing the time consumption of a single iteration from hours to minutes. More importantly, the feedback mechanism of the updated topography inherits the dynamic adaptability of the coupled model, avoiding the topographic divergence problem of pure data-driven models.

[0113] In combination with the first aspect, in some implementations, S31 includes: integrating a sea level rise parameter into the initial topographic boundary conditions;

[0114] S33 includes: splitting the long-term simulation into multiple discrete time periods, and the output topographic boundary conditions of each time period are used as the input topographic boundary conditions of the next time period.

[0115] Specifically, when integrating the sea level rise parameter into the initial topographic boundary conditions, the relative sea level change mechanism needs to be considered: if using the absolute sea level rise value, it is directly superimposed on the initial water depth field; if ground subsidence is involved, a vertical displacement amount needs to be added.

[0116] The discrete time period division needs to adapt to the non-uniformity of external forcing: equal-length time periods are used during the stable period of erosion and deposition rate (such as the stage of uniform sea level rise), and the time period length is shortened to capture the rapid response during the mutation event period (such as the year of frequent storm surges). When the output terrain boundary conditions of each time period are transmitted to the next time period, data format conversion and spatial interpolation need to be performed to ensure grid compatibility. Alternative solutions include using an adaptive time step algorithm to dynamically adjust the time period length according to the rate of change of the terrain.

[0117] This design solves the key difficulties of long-term simulation: dynamic integration of sea level rise parameters and boundary conditions ensures the spatio-temporal accuracy of water level driving force, avoiding systematic bias caused by fixed boundaries; discrete time period division balances computational efficiency and evolution process resolution, especially the response accuracy of sudden events is improved compared to traditional annual-scale simulation. For example, the process of tidal gully encroachment after a storm surge is completely captured in a 3-month short time period simulation, while annual simulation would miss such short-term geomorphic adjustments.

[0118] As shown in Figure 5 in combination with the first aspect, in some implementations, S4 includes:

[0119] S41: receiving the coastal zone development parameters, climate parameters, and hydrodynamic parameters input by the user through the interactive interface;

[0120] S42: based on the coastal zone development parameters, climate parameters, and hydrodynamic parameters, calling the hybrid simulation to perform geomorphic rapid calculation;

[0121] S43: converting the results of the geomorphic rapid calculation into visual graphics output, and providing parameter reset, calculation trigger, and history record management functions.

[0122] Specifically, the interactive interface of the business application module receives three types of parameters through graphical controls:

[0123] Coastal zone development parameters: reclamation range (map frame selection), port coordinates, dam length, etc.

[0124] Climate parameters: sea level rise rate options (low / medium / high), storm frequency increase / decrease percentage;

[0125] Hydrodynamic parameters: tide type (semi-diurnal tide / full-diurnal tide), wave dominant direction angle.

[0126] Input parameter immediate trigger threshold verification: if the sea level rise rate exceeds the historical extreme value, a warning box will pop up, and if the dam length exceeds a certain proportion of the coastline, the engineering rationality will be questioned.

[0127] When the geomorphology rapid calculation is performed, the system automatically maps the user parameters to the hybrid model input: the coastal development parameters are converted into a grid mask or boundary condition modification; the climate parameters drive the database to match the closest scenario combination; and the hydrodynamic parameters adjust the model initial settings. After the pre-trained hybrid model chain is called, the calculation core executes the loop iteration process, and finally outputs the seabed erosion and deposition change matrix.

[0128] The result visualization adopts the method of superimposing a heat map on a geographic base map: the erosion and deposition intensity is rendered with a red (deposition) -blue (erosion) color scale, and the tidal creek system is highlighted with black vector lines. The interface synchronously provides a parameter reset button (clear the current input), a calculation trigger button (start the rapid calculation), and a history record panel (store the input parameters and result thumbnail). The user can reproduce the complete prediction result by selecting a history record.

[0129] This module realizes "zero professional threshold" business application: the planner does not need to master the principle of numerical models, and can complete scenario setting only by map selection and parameter sliding; the input verification mechanism intercepts common sense errors; and the heat map output directly and intuitively shows the long-term geomorphic impact of the engineering scheme.

[0130] In combination with the first aspect, in some implementations, S41 includes: threshold range verification is performed on the coastal zone development parameters, climate parameters, and hydrodynamic parameters, and a warning information is triggered when each threshold range is exceeded;

[0131] S43 includes: superimposing the seabed erosion and deposition change in the form of a heat map on a geographic base map, and storing the input parameters and the corresponding erosion and deposition evolution results of each calculation.

[0132] Specifically, the parameter threshold verification is based on actual physical laws: for example, the wave direction angle is limited within the historical observation range of the target sea area; and the reclamation area cannot exceed a certain proportion of the total tidal flat area to prevent model distortion. The verification logic adopts multi-layer condition judgment: first, the data type legality is detected (for example, the numerical value type parameter cannot be a character), then the physical feasible range is compared (for example, the tidal range cannot be negative), and finally the parameter combination compatibility is checked (for example, the length of the jetty needs to match the coastal morphology). When the user input exceeds the limit, the interface displays a red warning icon next to the parameter and pops up a specific error message.

[0133] In the result visualization, the seabed erosion and deposition change heat map needs to be spatially registered with the geographic base map: a universal transverse Mercator projection coordinate system is used, and the erosion and deposition grid data is matched with the base map resolution through bilinear interpolation. The heat map color scale dynamically adapts to the erosion and deposition extreme value range, and a legend scale is added to improve readability. The history records are stored in the form of a database table, and each record contains a timestamp, a parameter set, a result map path, and metadata. When the user clicks on a record, the system automatically restores the parameter settings and reloads the result map.

[0134] The mechanism ensures the robustness of the business application: threshold verification prevents model collapse caused by non-physical input from the source; heat map geographic registration accurately locates the engineering impact area (such as the range of siltation downstream of the embankment); and the history record function supports scheme comparison and contrast. In practical application, a false siltation result that violates the conservation of mass was avoided by correcting an input sea level rise value that exceeded the threshold, demonstrating the system's immunity to incorrect input.

[0135] Embodiment 2

[0136] As shown in Figure 6 the second aspect, the present application provides a tidal flat evolution prediction system based on a hybrid model. The system uses the method provided in any of the above embodiments. The system comprises:

[0137] A first construction module for constructing a tidal flat erosion and deposition evolution database of the target coastal zone under different scenarios;

[0138] A second construction module connected to the first construction module, for constructing a geomorphology prediction model based on the tidal flat erosion and deposition evolution database through an artificial intelligence algorithm;

[0139] A hybrid simulation execution module connected to the second construction module, for executing hybrid simulation based on the physical mechanism model and the geomorphology prediction model, comprising:

[0140] Generating hydrodynamic feature data through the physical mechanism model, inputting the hydrodynamic feature data into the geomorphology prediction model to generate seabed erosion and deposition change, updating the topographic boundary condition based on the seabed erosion and deposition change and iterating cyclically;

[0141] A geomorphology rapid calculation execution and output module connected to the hybrid simulation execution module, for configuring a business application module based on the hybrid simulation, the business application module receiving user input parameters, executing geomorphology rapid calculation and outputting visual results.

[0142] The system corresponds to the method provided in Embodiment 1 described above, and will not be described again here.

[0143] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for part or all of the technical features; and these modifications or substitutions do not make the essence of the corresponding technical solution deviate from the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting tidal flat evolution based on a mixture model, characterized in that, Comprise: S1: Constructing a database of tidal flat erosion and deposition evolution of the target coastal zone under different scenarios; S2: Based on the tidal flat erosion and deposition evolution database, a geomorphology prediction model is constructed through an artificial intelligence algorithm; S2 comprises: S21: Based on the tidal flat erosion and deposition evolution database, the inundation probability, average shear stress, flood dominance and tidal creek distance function are sorted out to construct a training data set corresponding to the seabed erosion and deposition; S22: Based on the training data set, a convolutional neural network model is constructed with four-channel hydrodynamic features as input and seabed erosion and deposition change as output; S23: Based on the convolutional neural network model, an encoder-decoder structure is configured and feature maps are transmitted across layers through a jump connection to output an optimized convolutional neural network model; S24: Based on the optimized convolutional neural network model and the tidal creek distance function, the loss function is regionally weighted, and the precision optimized geomorphology prediction model is output; S3: Based on the physical mechanism model and the geomorphology prediction model, a hybrid simulation is performed, comprising: Generating hydrodynamic feature data through the physical mechanism model, inputting the hydrodynamic feature data into the geomorphology prediction model to generate seabed erosion and deposition change, updating the topographic boundary condition based on the seabed erosion and deposition change and iterating; S4: Based on the hybrid simulation, configure a business application module, the business application module receives user input parameters, performs geomorphology calculation and outputs visual results.

2. The method of predicting tidal flat evolution based on a hybrid model according to claim 1, characterized in that, S1 comprises: S11: A dynamic geomorphology model of the target coastal zone is established; S12: Based on the dynamic geomorphology evolution model, the tidal flat dynamic deposition process is reproduced, and the dynamic geomorphology evolution model is calibrated combined with historical topographic data; S13: Based on the calibrated dynamic geomorphology evolution model, the storm surge and seasonal water level fluctuation are time-scaled to generate annual geomorphology evolution feature data set; S14: Based on the annual geomorphology evolution feature data set, the influence of sediment supply, sea level change, tides and wave factors on erosion and deposition evolution is simulated to form the tidal flat erosion and deposition evolution database.

3. The tidal flat evolution prediction method based on the hybrid model according to claim 2, wherein S12 comprises: based on the dynamic geomorphology evolution model, long-term simulation is performed to output geomorphology data containing climate change and human activity influence; S13 comprises: based on the geomorphology data containing climate change and human activity influence, the characteristic parameters of storm surge and seasonal water level fluctuation are extracted for time scale reduction; S14 comprises: based on the characteristic parameters after time scale reduction, the contribution value of coastal engineering to geomorphology evolution is quantified.

4. The tidal flat evolution prediction method based on the hybrid model according to claim 1, wherein S23 comprises: the feature maps output by each stage of the encoder are transmitted to the corresponding stage of the decoder for splicing; S24 comprises: according to the tidal creek distance function value, a weight coefficient is assigned, and a higher weight value is assigned to the tidal creek edge area than to the flat area.

5. The method of predicting tidal flat evolution based on a hybrid model according to claim 1, wherein S3 Comprise: S31: Based on the initial topographic boundary condition, the inundation probability, average shear stress, flood dominance and tidal creek distance function are calculated and generated through the physical mechanism model; S32: input the submergence probability, the average shear stress, the flood dominance and the tidal creek distance function into the geomorphology prediction model, and output the seabed erosion and deposition variation in the current period; S33: update the terrain boundary condition based on the seabed erosion and deposition variation in the current period, and take the updated terrain boundary condition as the input of the next period, and execute S31 and S32 cyclically until the long-term simulation is completed.

6. The tidal flat evolution prediction method based on the hybrid model according to claim 5, characterized in that, S31 comprises: integrating the sea level rise parameter into the initial terrain boundary condition; S33 comprises: dividing the long-term simulation into multiple discrete time periods, and taking the output terrain boundary condition of each time period as the input terrain boundary condition of the next time period.

7. The method of predicting tidal flat evolution based on a hybrid model according to claim 1, wherein S4 comprises: S41: receiving the coastal zone development parameter, the climate parameter and the hydrodynamic parameter input by the user through the interactive interface; S42: calling the geomorphology quick calculation of the hybrid simulation based on the coastal zone development parameter, the climate parameter and the hydrodynamic parameter; S43: converting the result of the geomorphology quick calculation into a visualized graphic output, and providing the parameter reset, the calculation trigger and the historical record management function.

8. The tidal flat evolution prediction method based on the hybrid model according to claim 7, characterized in that, S41 comprises: performing threshold range verification on the coastal zone development parameter, the climate parameter and the hydrodynamic parameter, and triggering the warning information when the threshold range is exceeded; S43 comprises: superimposing the seabed erosion and deposition variation on the geographic base map in the form of a heat map, and storing the input parameter and the corresponding erosion and deposition evolution result of each calculation.

9. A tidal flat evolution prediction system based on a mixture model, characterized by, The system adopts the method according to any one of claims 1 to 8, and the system comprises: a first construction module configured to construct a tidal flat erosion and deposition evolution database of a target coastal zone under different scenarios; a second construction module connected with the first construction module and configured to construct a geomorphology prediction model based on the tidal flat erosion and deposition evolution database through an artificial intelligence algorithm, comprising: based on the tidal flat erosion and deposition evolution database, collating the submergence probability, the average shear stress, the flood dominance and the tidal creek distance function, and constructing a training data set corresponding to the seabed erosion and deposition; based on the training data set, constructing a convolutional neural network model taking four-channel hydrodynamic features as input and the seabed erosion and deposition variation as output; based on the convolutional neural network model, configuring an encoder-decoder structure and realizing feature map cross-layer transmission through a skip connection to output an optimized convolutional neural network model; based on the optimized convolutional neural network model and the tidal creek distance function, performing regional weighted calculation on a loss function to output a geomorphology prediction model with optimized precision; a hybrid simulation execution module connected with the second construction module and configured to execute hybrid simulation based on a physical mechanism model and the geomorphology prediction model, comprising: generating hydrodynamic feature data through the physical mechanism model, inputting the hydrodynamic feature data into the geomorphology prediction model to generate the seabed erosion and deposition variation, updating the terrain boundary condition based on the seabed erosion and deposition variation and iteratively repeating the process. The geomorphologic rapid calculation execution and output module is connected with the mixed simulation execution module, and is configured to configure a service application module based on the mixed simulation, the service application module receiving user input parameters, executing geomorphologic rapid calculation and outputting visualized results.

Images