Ocean tide level simulation method and device based on hybrid neural network model

Abstract

The invention relates to the technical field of ocean numerical simulation, in particular to an ocean tide level simulation method and device based on a hybrid neural network model, and the method comprises the steps: obtaining open boundary up-regulation parameters of a target sea area, inputting a pre-trained physical information neural network model in the hybrid neural network model, acquiring simulated tide level time sequence data of a plurality of tide observation stations; constructing a training data set, inputting the training data set into a deep neural network model in the hybrid neural network model, and training the deep neural network model; and inputting the actually measured tide level time sequence data of the plurality of tide observation stations into the trained deep neural network model, and obtaining the optimal harmonic parameters of the plurality of nodes from the deep neural network model. According to the method provided by the embodiment of the invention, by constructing and training the deep neural network model and directly inputting the actually measured tide level data of the tide observation station into the mixed neural network model, the optimal harmonic parameter is obtained, the labor and time cost in the numerical simulation process is reduced, and the numerical simulation efficiency is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of ocean numerical simulation, and particularly to an ocean tide level simulation method and device based on a hybrid neural network model. Background Art

[0002] In the numerical simulation processes of meteorology, ocean, etc., with the continuous improvement of the requirements for simulation accuracy and resolution, traditional global simulation methods are difficult to meet the needs of local high-precision simulation. Therefore, high-resolution local simulation is usually adopted currently. Local simulation requires setting "open boundaries" at the regional boundaries and providing boundary driving data that satisfies the closure of the control equations. However, these data often need to be corrected due to problems such as insufficient spatial resolution or large observation errors to meet the accuracy requirements of the numerical model.

[0003] However, in the related art, the processing of boundary driving data often depends on manual operation by humans. This process highly depends on the professional knowledge and parameter adjustment experience of operators, with high human and time costs, thus resulting in low numerical simulation efficiency. Therefore, how to reduce the human and time costs in the numerical simulation process and improve the numerical simulation efficiency has become a problem to be solved. Summary of the Invention

[0004] In view of this, the present disclosure provides an ocean tide level simulation method and device based on a hybrid neural network model to solve the problem of how to reduce the human and time costs in the numerical simulation process and improve the numerical simulation efficiency.

[0005] On the one hand, the present disclosure provides an ocean tide level simulation method based on a hybrid neural network model. The method includes: obtaining the harmonic parameters corresponding to multiple nodes on the open boundary of a target sea area, inputting the harmonic parameters into a physics-informed neural network model pre-trained in the hybrid neural network model to obtain the simulated tide level time series data corresponding to multiple tide gauges; constructing a training data set based on the harmonic parameters and the simulated tide level time series data, inputting the training data set into a deep neural network model in the hybrid neural network model, and training the deep neural network model so that the deep neural network model learns the inverse mapping relationship from the simulated tide level time series data to the harmonic parameters; inputting the measured tide level time series data of multiple tide gauges into the trained deep neural network model, and obtaining the optimal harmonic parameters of multiple nodes from the deep neural network model.

[0006] On the other hand, the present disclosure also provides an ocean tide level simulation device based on a hybrid neural network model. The device includes: a data-driven module, configured to obtain the harmonic parameters corresponding to multiple nodes on the open boundary of the target sea area, input the harmonic parameters into the physics-informed neural network model pre-trained in the hybrid neural network model, and obtain the simulated tide level time series data corresponding to multiple tide gauge stations; a model training module, configured to construct a training data set based on the harmonic parameters and the simulated tide level time series data, input the training data set into the deep neural network model in the hybrid neural network model, and train the deep neural network model so that the deep neural network model learns the inverse mapping relationship from the simulated tide level time series data to the harmonic parameters; an optimal harmonic parameter acquisition module, configured to input the measured tide level time series data of multiple tide gauge stations into the trained deep neural network model, and obtain the optimal harmonic parameters of multiple nodes from the deep neural network model.

[0007] On the other hand, the present disclosure also provides an electronic device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the above-mentioned ocean tide level simulation method based on the hybrid neural network model.

[0008] On the other hand, the present disclosure also provides a computer-readable storage medium, on which computer instructions are stored. The computer instructions are used to cause a computer to implement the above-mentioned ocean tide level simulation method based on the hybrid neural network model.

[0009] On the other hand, the present disclosure also provides a computer program product, including computer instructions, which are used to cause a computer to execute the above-mentioned ocean tide level simulation method based on the hybrid neural network model.

[0010] Through the ocean tide level simulation method and device based on the hybrid neural network model in the above embodiments of the present disclosure, by constructing and training the deep neural network model, it is enabled to have the ability to invert the harmonic parameters of each node at the open boundary according to the measured tide level data of multiple tide gauge stations. By directly inputting the measured tide level data of the tide gauge stations into the hybrid neural network model, the corresponding optimal harmonic parameters can be directly obtained, thereby avoiding the iterative adjustment process of traditional trial simulations relying on manual experience multiple times, reducing the human and time costs in the numerical simulation process, and improving the numerical simulation efficiency.

[0011] In addition, compared with the related art method of repeatedly verifying the boundary parameters by repeatedly running the numerical model, the above embodiments of the present disclosure obtain the optimal harmonic parameters at one time through the model training and inversion mechanism, effectively reducing the number of repeated simulations and resource consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] To more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the related art, the following will briefly introduce the accompanying drawings required for the description of the specific embodiments or the related art. Obviously, the accompanying drawings in the following description are some embodiments of the present disclosure. For those of ordinary skill in the art, other accompanying drawings can be obtained based on these drawings without creative efforts.

[0013] Figure 1 It is a schematic flowchart of a method for simulating ocean tide levels based on a hybrid neural network model provided by an embodiment of the present disclosure; Figure 2 It is a schematic flowchart of constructing a training data set for a method for simulating ocean tide levels based on a hybrid neural network model provided by an embodiment of the present disclosure; Figure 3 It is a schematic diagram of the DNN model structure of a method for simulating ocean tide levels based on a hybrid neural network model provided by an embodiment of the present disclosure; Figure 4 It is a schematic diagram of the structure of a device for simulating ocean tide levels based on a hybrid neural network model provided by an embodiment of the present disclosure; Figure 5 It is a schematic diagram of the structure of another device for simulating ocean tide levels based on a hybrid neural network model provided by an embodiment of the present disclosure. Specific Embodiments

[0014] In the field of numerical simulations such as meteorology and oceanography, to balance computational resources and simulation accuracy, high-resolution models are often used to finely simulate local key regions. Such local simulations require artificially setting the boundaries of the computational region, which are called open boundaries. Since open boundaries are not actual physical boundaries and cannot naturally close the model governing equations, it is necessary to additionally introduce boundary driving data (such as tidal levels or flow velocities) to ensure the model is solvable. The current mainstream approach usually relies on global ocean observations or model data (such as satellite data, global tidal databases) to obtain tidal level information and interpolates it at the open boundary positions of the target region to construct the time series data of tidal levels at the boundaries. This process involves multiple steps, including data downloading, format conversion, position matching, interpolation processing, and accuracy correction. Moreover, due to problems such as instrument errors and resolution limitations in the original observational data, significant errors are often introduced when projecting to local regions. To improve simulation accuracy, it is usually necessary to perform secondary processing and manual correction on the interpolated boundary data. On the other hand, in ocean dynamic simulations, the shallow water assumption is often adopted, and the shallow water equations are used as the governing equations to construct the tidal current motion model. Among them, the open boundary conditions are often generated by the astronomical tide harmonic method, in the form of the superposition of the amplitudes and phases of multiple specific frequency terms. This method requires obtaining the harmonic parameters (amplitudes and phases) corresponding to each astronomical tide and inputting them into the model to drive the simulation calculation. After the simulation is completed, it is necessary to compare the errors between the simulation results and the measured data, and repeat the simulation process by manually adjusting the parameters. Finally, the combination of harmonic parameters with the smallest error is obtained.

[0015] However, in the related technologies, the method for constructing ocean open boundaries still faces the following problems: 1. The entire process of boundary data preparation and correction highly depends on manual operations and professional knowledge. Among them, the harmonic parameters need to be manually downloaded, screened, and matched. The longitude and latitude interpolation also needs to be set according to the characteristics of the simulation region, and the error correction depends on the operator's understanding and experience of the main tidal components.

[0016] 2. The process of optimizing boundary data requires repeated ocean numerical simulations. For each round of simulation after adjustment, the entire model needs to be run to verify the error. The computational cost is high and the time consumption is long, making it difficult to efficiently complete the search for the optimal parameters.

[0017] 3. Since nearshore simulations usually include 8 astronomical tides, and each tide includes two adjustable parameters, amplitude and phase, the parameter dimension is as high as 16 dimensions. During the manual adjustment process, only individual parameters of the main tidal components can be roughly adjusted, and it is almost impossible to achieve global joint optimization of multiple parameters, thus limiting the simulation accuracy and the generalization ability of the model.

[0018] To solve the above problems, various embodiments of the present disclosure provide a method for simulating ocean tide levels based on a hybrid neural network model. The method includes: obtaining the harmonic parameters corresponding to multiple nodes on the open boundary of the target sea area, inputting the harmonic parameters into the physics-informed neural network model pre-trained in the hybrid neural network model, and obtaining the simulated tide level time series data corresponding to multiple tide gauges; constructing a training data set based on the harmonic parameters and the simulated tide level time series data, inputting the training data set into the deep neural network model in the hybrid neural network model, and training the deep neural network model so that the deep neural network model learns the inverse mapping relationship from the simulated tide level time series data to the harmonic parameters; inputting the measured tide level time series data of multiple tide gauges into the trained deep neural network model, and obtaining the optimal harmonic parameters of multiple nodes from the deep neural network model.

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0020] Please refer to Figure 1 , Figure 1 is a schematic flowchart of a method for simulating ocean tide levels based on a hybrid neural network model provided by an embodiment of the present disclosure. The process of the method may include the following steps: Step S101, obtain the harmonic parameters corresponding to multiple nodes on the open boundary of the target sea area, input the harmonic parameters into the physics-informed neural network model pre-trained in the hybrid neural network model, and obtain the simulated tide level time series data corresponding to multiple tide gauges.

[0021] In this embodiment, the target sea area may refer to a coastal area or a local sea area with a limited spatial range selected to meet the research and detection requirements. The target sea area has a clear calculation boundary in numerical simulation, and the non-closed part thereof is the open boundary, and an external tide level drive needs to be introduced for simulation closure.

[0022] For example, the target sea area may include, but is not limited to: the nearshore area, the offshore area, or the bay area.

[0023] Further, the nodes on the open boundary of the target sea area may refer to a series of discrete spatial points selected on the boundary of the simulation area for applying boundary conditions.

[0024] The harmonic parameters obtained separately from multiple nodes may refer to physical quantities used to describe the astronomical tidal components. The harmonic parameters may include the amplitude and phase of each astronomical tidal component.

[0025] Among them, the amplitude of the astronomical tidal component represents the maximum value of the tidal level change caused by this tidal component at the node; the phase of the astronomical tidal component represents the angular value corresponding to the time difference when the maximum high tide appears after a preset time.

[0026] Furthermore, input the harmonic parameters corresponding to each of the obtained multiple nodes into the physics-informed neural network model pre-trained in the hybrid neural network model.

[0027] Among them, the physics-informed neural networks (PINN) is a model that combines physical laws with the data-driven ability of neural networks and can be used to approximately solve the partial differential equations of physical processes.

[0028] Specifically, the core idea of the PINN model can be as follows: construct a neural network to simulate the changes of physical quantities over time and space, introduce the known physical control equations into the loss function, and during the training process, fit the data points through the neural network while making the output of the network conform to the physical control equations to constrain the rationality and interpretability of the solution.

[0029] In the embodiments of the present disclosure, the purpose of pre-training the PINN model may be to enable the PINN model to learn the forward mapping relationship from harmonic parameters to simulated tidal level time series data.

[0030] Still further, based on the pre-trained PINN model, the simulated tidal level time series data corresponding to the obtained multiple tide gauges may refer to the tidal level height sequence data calculated by the PINN model that changes over time at a specific location.

[0031] Step S102, construct a training dataset based on the harmonic parameters and the simulated tidal level time series data, input the training dataset into the deep neural network model in the hybrid neural network model, and train the deep neural network model so that the deep neural network model learns the inverse mapping relationship from the simulated tidal level time series data to the harmonic parameters.

[0032] In this embodiment, use the harmonic parameters as labels and the simulated tidal level time series data as input features to construct a training dataset based on the labels and input features.

[0033] Input the training dataset into the deep neural network model and train the deep neural network model.

[0034] Among them, the Deep Neural Networks (DNN) model is an artificial neural network structure composed of multiple layers of neurons. The DNN model consists of an input layer, multiple hidden layers, and an output layer. By pre-training the DNN model, the deep neural network model can learn the inverse mapping relationship from the simulated tide level time series data to the harmonic parameters; the trained DNN model, based on the learned inverse mapping relationship and the input tide level time series data, can deduce the corresponding optimal harmonic parameters.

[0035] Step S103: Input the measured tide level time series data of multiple tide gauging stations into the trained deep neural network model, and obtain the optimal harmonic parameters of multiple nodes from the deep neural network model.

[0036] In this embodiment, the measured tide level time series data of multiple tide gauging stations is obtained, and the measured tide level time series data is preprocessed to make the format of the measured tide level time series data consistent with the format of the simulated tide level time series data.

[0037] Input the measured tide level time series data into the trained DNN model, so that the DNN model outputs the optimal harmonic parameters of each node on the open boundary.

[0038] Through the above-mentioned embodiment of the ocean tide level simulation method and device based on the hybrid neural network model of the present disclosure, by constructing and training the deep neural network model, it enables it to have the ability to deduce the harmonic parameters of each node at the open boundary according to the measured tide level data of multiple tide gauging stations, avoiding the iterative adjustment process of traditional multiple experimental simulations relying on manual experience, reducing the human and time costs in the numerical simulation process, and improving the numerical simulation efficiency. Compared with the related technology that repeatedly verifies the boundary parameters by repeatedly running the numerical model, the above-mentioned embodiment of the present disclosure obtains the optimal harmonic parameters at one time through the model training and inverse deduction mechanism, effectively reducing the number of repeated simulations and resource consumption.

[0039] In a possible implementation manner of the above step S101, obtaining the harmonic parameters corresponding to multiple nodes on the open boundary of the target sea area includes: Through the astronomical tide component parameter random generator, obtain N b groups of harmonic parameters corresponding to each of the N T nodes on the open boundary of the target sea area; where each group of harmonic parameters includes the amplitude and phase k of the th astronomical tide component among 8 astronomical tide components.

[0040] In this embodiment, the astronomical tidal component parameter random generator may refer to a tool that, in a simulation environment, automatically generates multiple sets of astronomical tidal component harmonic parameter combinations that satisfy statistical laws based on physical constraints and empirical distributions.

[0041] Here, the astronomical tidal component parameter random generator can batch generate physically reasonable harmonic parameter samples when there is insufficient real observational data.

[0042] Furthermore, the astronomical tidal component parameter random generator can randomly generate harmonic parameters within the range of .

[0043] Among them, can be custom - set according to the historical maximum tidal level rise in the target sea area. For example, can be set to 4 times the maximum tidal level rise detected by the tide gauge station.

[0044] Furthermore, at the open boundary of the target sea area, N b discrete nodes are selected. For each node, the astronomical tidal component parameter random generator generates N T sets of harmonic parameters, and each set of harmonic parameters includes the amplitudes and phases corresponding to 8 typical astronomical tidal components respectively.

[0045] Here, the 8 typical astronomical tidal components are S2, M2, N2, K2, K1, P1, O1, and Q1 respectively. Among them, S2 represents the diurnal solar semidiurnal tide, M2 represents the principal lunar semidiurnal tide, N2 represents the major lunar ellipse semidiurnal tide, K2 represents the lunisolar semidiurnal tide, K1 represents the lunisolar diurnal tide, P1 represents the diurnal solar tide, O1 represents the principal lunar diurnal tide, and Q1 represents the major lunar ellipse diurnal tide.

[0046] Through the above - mentioned embodiment of the ocean tidal level simulation method and device based on the hybrid neural network model of the present disclosure, by generating multiple sets of harmonic parameter samples at multiple open - boundary nodes, the input space is expanded, enabling the PINN model to cover a wider distribution of harmonic parameters during the training phase, and improving the generalization ability and adaptability of the model. By generating harmonic parameters based on the preset maximum amplitude range and phase period range, non - physical simulation inputs can be effectively avoided, and the reliability of the tidal level simulation results can be ensured.

[0047] In a possible implementation manner of the above - mentioned embodiment, the pre - trained physics - informed neural network model is implemented based on the following steps: Construct the input layer and output layer of the physics - informed neural network model. The input layer includes at least the N b sets of harmonic parameters corresponding to each of the N T nodes, and the output layer includes at least the N of each of the M tide - gauging stations ​T Group of simulated tide level time series data ; Define the composite loss function of the physics-informed neural network model based on physical constraint terms and data-driven terms; among them, the physical constraint term uses the shallow water equations as the physical constraint; Minimize the composite loss function through the gradient backpropagation algorithm. When the composite loss function is lower than the preset threshold, the training of the physics-informed neural network model is completed.

[0048] In this embodiment, is the amplitude of the j th k astronomical tidal component at node b ; is the phase of the j th k astronomical tidal component at node b ; is the simulated tide level time series at the position of the i th tide observation station ( ) at time ;

[0049] Furthermore, during the pre-training process of the PINN model, the shallow water equations are introduced into the loss function of the PINN model, so that the prediction results of the PINN model not only need to fit the data, but also must satisfy the physical laws corresponding to the shallow water equations.

[0050] Here, the shallow water equations refer to a set of partial differential equations used to describe the shallow flow behavior of water bodies on the Earth's surface and can be used for tidal current modeling in shallow sea areas.

[0051] The formula of the shallow water equations is:

[0052] Among them, the expressions of each term in the shallow water equations are as follows: , F ,

[0053] Among them, is the tide level, and are the velocity components; represents the conserved variable, that is, the physical quantity that needs to evolve with time; F represents the flux term, which describes the propagation of the state quantity in space during the water body movement process; represents the source term, which represents the internal change of the system caused by non-conservative quantities, such as frictional force, Coriolis force, gravity flow caused by the bottom slope, and external force. is the bottom slope elevation, is the total water depth, is the bottom friction coefficient, is the Coriolis force coefficient. At the open boundary position , the water level drive (zero gradient for velocity) form is usually adopted as the open boundary condition .

[0054] Furthermore, the physical constraint term takes the physical control equation directly as the restriction condition for neural network training, ensuring that the network output not only fits the data but also must satisfy the physical laws. The data-driven term is the loss function of traditional supervised learning, which is used to make the neural network fit the existing observed data as much as possible.

[0055] Based on the physical constraint term and the data-driven term, a composite loss function of the physics-informed neural network model can be defined, including: a weighted sum based on the physical constraint term and the data-driven term to define the composite loss function of the PINN model.

[0056] Through the above-mentioned embodiment of the ocean tide level simulation method and device based on the hybrid neural network model of the present disclosure, by using the shallow water equation as the physical constraint term and the observed data-driven term to jointly construct a composite loss function, and using the gradient backpropagation algorithm to train the model, gradually minimizing the loss function until it meets the preset threshold, the PINN model can follow the physical laws while ensuring the data fitting accuracy, improving the accuracy of data simulation.

[0057] In a possible implementation manner of the above embodiment, constructing the input layer and output layer of the physics-informed neural network model includes: Constructing the input layer of the physics-informed neural network model, the input layer includes at least one of the following: any sampling coordinate and sampling time in the target sea area ( x, y, t ), N T groups of harmonic parameters of each node on the open boundary; wherein, the sampling coordinates in the target sea area include at least one of the following: each node on the open boundary, tide observation stations, and internal sea area points; Constructing the output layer of the physics-informed neural network model, the output layer includes at least one of the following: the simulated tide level of any sampling coordinate in the target sea area ζ ( x, y, t ), velocity components U ( x, y, t ) and V ( x, y, t ), wherein, the simulated tide level ζ ( x, y, t ), velocity components U ( x, y, t ) and V ( x, y, t ) are used to form the simulated tide level time series data .

[0058] In this embodiment, the simulated tidal level ζ ( x, y, t ) represents the water surface height change at the coordinate ( x, y ) and time t . The velocity component U ( x, y, t ) and V ( x, y, t ) can respectively represent the velocity components of the water body in the x and y directions.

[0059] Through the above-mentioned ocean tidal level simulation method and device based on the hybrid neural network model of the present disclosure, by constructing the input layer and output layer of the PINN model, the model can output the simulated tidal level and flow velocity time series data of any point in the target sea area based on the space-time coordinates and boundary harmonic parameters, meeting the requirements of tidal physical process simulation. Based on the input of coordinates and time, the PINN model can output the tidal information of any point in the target sea area, with good spatial and temporal interpolation capabilities, supporting high-quality simulation of unobserved points inside the sea area.

[0060] In a possible implementation manner of the above embodiment, based on the physical constraint term and the data-driven term, a composite loss function of the physical information neural network model is defined, including: The equation residual term constructed based on the shallow water equation, as the physical constraint term; The boundary residual term constructed based on the difference between the measured tidal level time series data and the simulated tidal level time series data, as the data-driven term; The composite loss function of the physical information neural network model is defined by the weighted sum of the physical constraint term and the data-driven term.

[0061] In this embodiment, through the preset weights in the composite loss function, it is possible to avoid a certain term from dominating the training process due to a large or small value, ensuring a reasonable coordination between physical constraints and data fitting.

[0062] Through the above-mentioned ocean tidal level simulation method and device based on the hybrid neural network model of the present disclosure, by introducing the data-driven term, the error between the model output and the true observed tidal level is constrained, effectively improving the fitting accuracy of the model for the tidal level time series data and making the simulation results closer to the true observations. By weighted fusion of the physical constraint term and the data-driven term, the complementarity between knowledge-driven and data-driven is realized.

[0063] In a possible implementation manner of the above embodiment, the equation residual term constructed based on the shallow water equation is:

[0064]

[0065] Among them, is the conservation variable vector of the shallow water equations; is the flux function composed of the simulated tidal level ζ ( x, y, t ) and the velocity components U ( x, y, t ) and V ( x, y, t ); is the source term function composed of the simulated tidal level ζ ( x, y, t ) and the velocity components U ( x, y, t ) and V ( x, y, t ); is the residual of the shallow water equations by the physics-informed neural network model at the sampling coordinates; N is the number of samples of the sampling coordinates in the target sea area; is the set of parameters to be trained for the physics-informed neural network model.

[0066] In this embodiment, is the discrete residual form of the shallow water equations and serves as the physical constraint embedded in the PNN model. represents calculating the mean of the squared residuals of the shallow water equations at all sampling points so that the physical variables learned by the neural network during the training optimization process can satisfy the shallow water equations as much as possible.

[0067] Furthermore, the equation residual represents the degree to which the shallow water equations are not satisfied.

[0068] Even further, by minimizing , the PINN model will be constrained to generate physical prediction results that automatically satisfy the shallow water equations. Among them, the physical prediction results can be the tidal level and the velocity.

[0069] Exemplarily, is the set of parameters to be trained for the physics-informed neural network model, which can include but are not limited to the weights between neurons in each layer and the bias terms of each neuron.

[0070] Through the method and device for simulating ocean tide levels based on a hybrid neural network model in the above embodiments of the present disclosure, the training process of the neural network is constrained by the residual of the shallow water equations, enabling the tide levels and flow velocities output by the model to automatically comply with physical laws and effectively improving the physical credibility of the simulation results. By jointly constructing a composite loss function with the equation residual term and the boundary residual term, the physical prior and the characteristics of the measured data are effectively fused, enhancing the overall learning effect and engineering practicability of the PINN model.

[0071] In a possible implementation manner of the above embodiment, the boundary residual term constructed based on the difference between the measured tide level time series data and the simulated tide level time series data is:

[0072] wherein, is the simulated tide level output by the physics-informed neural network model at the tide gauge station and time ; is the measured tide level at the tide gauge station and time ; represents the mean square error term between the simulated tide level and the measured tide level.

[0073] In this embodiment, a set of tide gauge stations in the target sea area is selected, as well as multiple time points corresponding to each station, and the measured tide levels at each station at the corresponding time are collected.

[0074] The spatial coordinates of each station and the time are input into the PINN model to obtain the simulated tide level of each station.

[0075] The difference between the simulated tide level and the measured tide level is determined , and the mean square error of each station is calculated as the boundary residual term .

[0076] Through the method and device for simulating ocean tide levels based on a hybrid neural network model in the above embodiments of the present disclosure, by constructing a boundary residual term and incorporating the error between the measured tide level and the simulated tide level into the loss function, the fitting ability of the model to the observed data is strengthened, thereby improving the accuracy of the physics-informed neural network model in tide level prediction.

[0077] In a possible implementation manner of the above embodiment, the composite loss function of the physics-informed neural network model is:

[0078] wherein, is a preset weight coefficient.

[0079] In this embodiment, It can also refer to the open boundary learning rate, which can control the contribution rate of the boundary observation error to the total loss in each round of training, thereby affecting the learning intensity of the boundary information during the training process.

[0080] For example, when is too large, the model is more biased towards fitting the boundary residual term to improve the accuracy of the model at the open boundary; while when is too small, the model is more biased towards fitting the physical residual term to make the model more conform to the physical equation.

[0081] In a possible implementation manner, during the training process of the model, the open boundary learning rate can be a fixed constant or can be dynamically adjusted.

[0082] For example, the dynamic adjustment method of the open boundary learning rate can include at least one of the following: dynamic adjustment based on the residual ratio, dynamic scaling adjustment, and staged adjustment.

[0083] Through the above embodiments of the ocean tide level simulation method and device based on the hybrid neural network model of the present disclosure, by using weighted control, the model can flexibly switch between physical-driven and data-driven, and is especially suitable for complex ocean modeling environments with uneven measured data quality, incomplete regional boundaries, or only partial open boundaries with observed data.

[0084] In a possible implementation manner of the above embodiment, the physical information neural network model is trained by minimizing the composite loss function through the gradient backpropagation algorithm. When the composite loss function is lower than the preset threshold, the training of the physical information neural network model is completed, including: Training the physical information neural network model through the Adam optimizer; Based on the training data set, perform a certain number of random gradient iterative updates to optimize the set of parameters to be trained in the physical information neural network model ; During the training process, if the composite loss function is lower than the preset threshold or reaches the maximum number of iterations, terminate the training of the physical information neural network model and determine the pre-trained physical information neural network model.

[0085] In this embodiment, initializing the model includes: initializing the feedforward neural network, which can be composed of multiple hidden layers, each hidden layer has multiple nodes, and the activation function can adopt Tanh. For example, the feedforward neural network is composed of 3 hidden layers, and each hidden layer has 100 nodes.

[0086] Here, the Tanh activation function can be an S-shaped activation function with an output range of [-1, 1]. Specifically, since the tidal level fluctuations oscillate up and down around a specific equilibrium point and have periodic symmetry, and the Tanh activation function is also an activation function symmetric about the origin, it helps to capture the periodic or symmetric fluctuations in physical phenomena, thus contributing to learning a representation that better conforms to physical characteristics.

[0087] In addition, since the Tanh activation function is a smooth and continuous activation function that can provide continuous derivatives over the entire domain and does not have discontinuous derivative points like the ReLU activation function, it can greatly improve the numerical stability and smoothness of the model in the calculation of the derivative term residuals.

[0088] Furthermore, in a possible implementation, during the model training process, the Adam optimizer can be used to optimize at a preset learning rate, and optimize at the preset learning rate during each training process, and each training uses a batch of a preset number of tidal level data.

[0089] For example, the preset learning rate can be 1*10 -3 , and the preset number can be 128 groups.

[0090] Furthermore, the training dataset is divided into small chunks, and 128 groups of training samples are randomly selected from the training dataset as the training input for each training iteration.

[0091] Even further, during the training process, if the composite loss function is lower than the preset threshold or the number of iterations reaches 40,000 times, terminate the training of the physics-informed neural network model and determine the pre-trained physics-informed neural network model.

[0092] Through the method and device for ocean tidal level simulation based on the hybrid neural network model in the above embodiments of the present disclosure, since the tidal level signal itself has periodicity, volatility, and symmetry, the Tanh activation function can be used to better capture the physical characteristics of tidal level changes, thereby improving the calculation accuracy of calculating the physical residuals. Using a mini-batch composed of 128 groups of tidal level data in each round of training and combining the Adam optimizer for parameter iterative update can effectively reduce the memory overhead and improve the training efficiency. Introducing the preset composite loss function threshold and the maximum number of iterations as termination conditions during the training process can, on the premise of ensuring that the model has achieved better physical consistency and fitting accuracy, timely terminate the training process, avoid overfitting the training data, and improve the stability of the model.

[0093] In a possible implementation of the above step S102, a training dataset is constructed based on the harmonic parameters and the simulated tidal level time series data, and the training dataset is input into the deep neural network model in the hybrid neural network model to train the deep neural network model, including: Take N T sets of simulated tide level time series data as input tensors, and take the harmonic parameters corresponding to each set of simulated tide level time series data as output labels to construct a training dataset; Use the training dataset to train the deep neural network model, and use a preset loss function to evaluate the deviation between the simulated harmonic parameters and the true harmonic parameters; Optimize the parameters of the deep neural network model through the Adam optimizer. When the training loss converges or reaches the maximum number of iterations, the training of the deep neural network model is completed.

[0094] In this embodiment, extract N T sets of simulated tide level time series data, and take the harmonic parameters of each set of simulated tide level time series data as output label data to construct a training dataset, and use the training dataset to train the DNN model. Among them, the training dataset includes multiple training pairs.

[0095] Optimize the set of parameters to be trained of the DNN model through the Adam optimizer When the training loss converges or reaches the maximum number of iterations, the training of the deep neural network model is completed.

[0096] Through the method and device for ocean tide level simulation based on the hybrid neural network model in the above embodiments of the present disclosure, by establishing a mapping relationship between the simulated tide level time series data and its corresponding harmonic parameters, and using the DNN model to learn this non-linear relationship, it is possible to directly predict the harmonic parameters from the tide level sequence without performing computationally complex traditional harmonic analysis, thereby greatly improving the efficiency and automation of harmonic parameter extraction.

[0097] In a possible implementation manner of the above embodiment, a batch normalization layer is connected after each hidden layer of the deep neural network model to be used for normalizing the output features of each hidden layer.

[0098] In this embodiment, the DNN model may include 5 hidden layers, and the number of neurons in each hidden layer may be 100, 100, 80, 50, and 30 respectively.

[0099] Perform batch normalization on each hidden layer, that is, perform mean and variance normalization on each dimension feature of the small batch of samples.

[0100] Through the method and device for ocean tide level simulation based on the hybrid neural network model in the above embodiments of the present disclosure, batch normalization can keep the input of each layer in a stable distribution, so that a larger learning rate can also be used for training, which helps to accelerate the convergence of the model, reduce the number of iterations, and improve the training efficiency.

[0101] In a possible implementation of the above embodiment, the number of neurons in the input layer of the deep neural network model is M, and the number of neurons in the input layer is equal to the number of tide observation stations; the number of neurons in the output layer of the deep neural network model is 2*N b , and the number of neurons in the output layer is equal to twice the number of nodes on the open boundary.

[0102] In this embodiment, the number of neurons in the input layer is M, which can represent that the model receives the simulated tide level time series data of M tide observation stations at one time. The number of neurons in the output layer of the deep neural network model is 2*N b , which can represent that each boundary node needs to predict two harmonic parameters, namely the output amplitude and phase.

[0103] Furthermore, during the training process of the DNN model, the batch size of the training data can be 215, the initial learning rate can be 0.03, and the maximum number of iterations can be 1000.

[0104] Through the method and device for ocean tide level simulation based on a hybrid neural network model in the above embodiments of the present disclosure, the number of neurons in the input layer corresponds one-to-one with the number of tide observation stations, and can completely receive and express all observed tide level information, improving the spatial coverage ability of the model. The number of neurons in the output layer is set to twice the number of boundary nodes, allowing the model to simultaneously predict the amplitude and phase information of each boundary node, thus possessing the ability of tidal harmonic decomposition.

[0105] In one embodiment, please refer to Figure 2 , Figure 2 is a schematic diagram of the training data set construction process of a method for ocean tide level simulation based on a hybrid neural network model provided by an embodiment of the present disclosure. This process may include the following steps: Step S201, obtain the harmonic parameters of multiple nodes on the open boundary; Step S202, input the harmonic parameters into the physics-informed neural network model; Step S203, obtain the simulated tide level time series data; Step S204, store the harmonic parameters and the simulated tide level time series data into the training data set; Step S205, check whether the number of the data set is greater than the preset number; if so, end the process; if not, go to step S206; Step S206, use the astronomical tidal component parameter random generator to continue generating harmonic parameters; Here, continue to generate harmonic parameters and go to step S201.

[0106] In one embodiment, please refer to Figure 3 , Figure 3Schematic diagram of the DNN model structure of a method for simulating ocean tide levels based on a hybrid neural network model provided by an embodiment of the present disclosure, where: Assume the number of tide gauge stations M = 5, and the number of nodes N on the open boundary b = 3. Then the number of neurons in the input layer of the DNN model is 5, and each neuron in the input layer corresponds to a simulated tide level time series data (i.e., the first tide level data to the fifth tide level data, that is ); the number of neurons in the output layer is 3, and each output layer corresponds to an optimal harmonic parameter of an open boundary node (i.e., the first harmonic parameter to the third harmonic parameter, that is , , ). In addition, the number of neurons in the first hidden layer of the DNN model is 6, and the number of neurons in the second hidden layer is 5.

[0107] In one embodiment, there is provided an ocean tide level simulation device 400 based on a hybrid neural network model. The ocean tide level simulation device 400 based on the hybrid neural network model corresponds one-to-one with the method for simulating ocean tide levels based on the hybrid neural network model in the above embodiment. As Figure 4 shown, the ocean tide level simulation device 400 based on the hybrid neural network model includes a data-driven module 401, a model training module 402, and an optimal harmonic parameter acquisition module 403. Among them, the detailed descriptions of each functional module are as follows: The data-driven module 401 is configured to obtain the harmonic parameters corresponding to multiple nodes on the open boundary of the target sea area, input the harmonic parameters into the physics-informed neural network model pre-trained in the hybrid neural network model, and obtain the simulated tide level time series data corresponding to multiple tide gauge stations; The model training module 402 is configured to construct a training data set based on the harmonic parameters and the simulated tide level time series data, input the training data set into the deep neural network model in the hybrid neural network model, and train the deep neural network model so that the deep neural network model learns the inverse mapping relationship from the simulated tide level time series data to the harmonic parameters; The optimal harmonic parameter acquisition module 403 is configured to input the measured tide level time series data of multiple tide gauge stations into the trained deep neural network model, and obtain the optimal harmonic parameters of multiple nodes from the deep neural network model.

[0108] In one embodiment, the data-driven module 401 is specifically configured to obtain, through an astronomical tide component parameter random generator, N b groups of harmonic parameters corresponding to each of the N T nodes on the open boundary of the target sea area ; where each group of harmonic parameters includes the amplitude k and phase of the 。

[0109] In one embodiment, the data-driven module 401 is specifically configured to construct the input layer and the output layer of the physics-informed neural network model. The input layer includes at least N b sets of harmonic parameters corresponding to each of the N T nodes, and the output layer includes at least N sets of simulated tide level time series data for each of the M tide gauge stations T ; ; Define a composite loss function for the physics-informed neural network model based on the physical constraint term and the data-driven term. Among them, the physical constraint term uses the shallow water equation as the physical constraint; Minimize the composite loss function through the gradient backpropagation algorithm. When the composite loss function is lower than the preset threshold, the training of the physics-informed neural network model is completed. In one embodiment, the data-driven module 401 is specifically configured to construct the input layer of the physics-informed neural network model. The input layer includes at least one of the following: any sampling coordinate and sampling time in the target sea area ( x, y, t ), N T sets of harmonic parameters for each node on the open boundary; among them, the sampling coordinates in the target sea area include at least one of the following: each node on the open boundary, tide gauge stations, and internal sea area points; Construct the output layer of the physics-informed neural network model. The output layer includes at least one of the following: the simulated tide level at any sampling coordinate in the target sea area ζ ( x, y, t ), flow velocity components U ( x, y, t ) and V ( x, y, t ), where the simulated tide level ζ ( x, y, t ), flow velocity components U ( x, y, t ) and V ( x, y, t ) are used to form the simulated tide level time series data .

[0110] In one embodiment, the data-driven module 401 is specifically configured to use the equation residual term constructed based on the shallow water equation as the physical constraint term; Use the boundary residual term constructed based on the difference between the measured tide level time series data and the simulated tide level time series data as the data-driven term; Define the composite loss function of the physics-informed neural network model as the weighted sum of the physical constraint term and the data-driven term.

[0111] In one embodiment, the equation residual term constructed based on the shallow water equations is:

[0112]

[0113] Wherein, is the conservation variable vector of the shallow water equations; is the flux function composed of the simulated tide level ζ ( x, y, t ) and the velocity components U ( x, y, t ) and V ( x, y, t ); is the source term function composed of the simulated tide level ζ ( x, y, t ) and the velocity components U ( x, y, t ) and V ( x, y, t ); is the residual of the shallow water equations by the physics-informed neural network model at the sampling coordinates; N is the number of sampling coordinates in the target sea area; is the set of parameters to be trained for the physics-informed neural network model.

[0114] In one embodiment, the boundary residual term constructed based on the difference between the measured tide level time series data and the simulated tide level time series data is:

[0115] Wherein, is the simulated tide level output by the physics-informed neural network model at the tide gauge station and time ; is the measured tide level at the tide gauge station and time ; characterizes the mean square error term between the simulated tide level and the measured tide level.

[0116] In one embodiment, for the data-driven module 401, the composite loss function of the physics-informed neural network model is:

[0117] Wherein, is the preset weight coefficient.

[0118] In one embodiment, the data-driven module 401 is specifically configured to train the physics-informed neural network model through the Adam optimizer; Based on the training data set, perform random gradient iterative updates a number of times to optimize the set of parameters to be trained in the physics-informed neural network model ; During the training process, if the composite loss function is lower than the preset threshold or reaches the maximum number of iterations, terminate the training of the physics-informed neural network model and determine the pre-trained physics-informed neural network model.

[0119] In one embodiment, the model training module 402 is configured to use each group of simulated tide level time series data in N T groups of simulated tide level time series data as input tensors, and use the harmonic parameters corresponding to each group of simulated tide level time series data as output labels to construct a training data set; Use the training data set to train the deep neural network model, and use a preset loss function to evaluate the deviation between the simulated harmonic parameters and the true harmonic parameters; Optimize the parameters of the deep neural network model through the Adam optimizer, and complete the training of the deep neural network model when the training loss converges or reaches the maximum number of iterations.

[0120] In one embodiment, a batch normalization layer is connected after each hidden layer in the multiple hidden layers of the deep neural network model to be used for normalizing the output features of each hidden layer.

[0121] In one embodiment, the number of neurons in the input layer of the deep neural network model is M, and the number of neurons in the input layer is equal to the number of tide gauging stations; the number of neurons in the output layer of the deep neural network model is 2*N b , and the number of neurons in the output layer is equal to twice the number of nodes on the open boundary.

[0122] It should be noted that: when the above-mentioned ocean tide level simulation device based on the hybrid neural network model implements the corresponding ocean tide level simulation method based on the hybrid neural network model, only the above-mentioned division of each program module is used for illustration. In practical applications, the above-mentioned processing can be allocated to different program modules according to needs, that is, the internal structure of the above-mentioned system is divided into different program modules to complete all or part of the above-mentioned processing. In addition, the above-mentioned system provided in the embodiment and the corresponding Figure 1 embodiment of the method shown belong to the same concept, and the specific implementation process can be seen in the method embodiment, which will not be repeated here.

[0123] The embodiments of the present disclosure also provide an electronic device having the above Figure 4 shown ocean tide level simulation device based on the hybrid neural network model.

[0124] Please refer to Figure 5 ,Figure 5 FIG. Figure 5 is a schematic structural diagram of another ocean tide level simulation device based on a hybrid neural network model provided by an embodiment of the present disclosure. As shown in Figure 5 FIG. Figure 5 , the electronic device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or in other ways as needed. The processor can process instructions executed within the electronic device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple electronic devices can be connected, and each device provides some necessary operations (such as an array of servers, a set of blade servers, or a multi-processor system). Figure 5 Here, one processor 10 is taken as an example.

[0125] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above-mentioned hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device can be a complex programmable logic device, a field-programmable gate array, a general array logic, or any combination thereof.

[0126] Among them, the memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.

[0127] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the electronic device. In addition, the memory 20 can include a high-speed random access memory and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the electronic device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0128] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0129] The electronic device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 may be connected by a bus or other means. Figure 5 Taking the connection through the bus as an example.

[0130] The input device 30 can receive input digital or character information and generate key signal inputs related to the user settings and function controls of the electronic device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED), and a haptic feedback device (e.g., a vibration motor), etc. The above display device includes, but is not limited to, a liquid crystal display, a light-emitting diode, a display, and a plasma display. In some alternative embodiments, the display device may be a touch screen.

[0131] The electronic device further includes a communication interface for the electronic device to communicate with other devices or a communication network.

[0132] The embodiments of the present disclosure also provide a computer-readable storage medium. The methods according to the embodiments of the present disclosure can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented by downloading over a network the original computer code stored in a remote storage medium or a non-transitory machine-readable storage medium and to be stored in a local storage medium, so that the methods described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0133] A part of the present disclosure can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the present disclosure through the operations of the computer. Those skilled in the art should understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible by the computer.

[0134] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. An ocean tide level simulation method based on a hybrid neural network model, characterized in that, The method includes: Obtaining the harmonic parameters corresponding to multiple nodes on the open boundary of the target sea area, inputting the harmonic parameters into the physics-informed neural network model pre-trained in the hybrid neural network model, and obtaining the simulated tide level time series data corresponding to multiple tide gauge stations; Constructing a training dataset based on the harmonic parameters and the simulated tide level time series data, inputting the training dataset into the deep neural network model in the hybrid neural network model, and training the deep neural network model so that the deep neural network model learns the inverse mapping relationship from the simulated tide level time series data to the harmonic parameters; Inputting the measured tide level time series data of the multiple tide gauge stations into the trained deep neural network model, and obtaining the optimal harmonic parameters of the multiple nodes from the deep neural network model.

2. The method according to claim 1, wherein The obtaining of the harmonic parameters corresponding to multiple nodes on the open boundary of the target sea area includes: Through the astronomical tidal component parameter random generator, obtain the N b sets of harmonic parameters corresponding to each of the N T nodes on the open boundary of the target sea area ; among them, each set of harmonic parameters includes the amplitude k and phase of the th astronomical tidal component among 8 astronomical tidal components.

3. The method according to claim 2, characterized in that The pre-trained physics-informed neural network model is implemented based on the following steps: Construct the input layer and the output layer of the cyber-physical neural network model. The input layer includes at least N b sets of harmonic parameters corresponding to each of the N T nodes among the nodes, and the output layer includes at least N sets of simulated tide level time series data for each of the M tide gauge stations T ; ; Defining a composite loss function of the physics-informed neural network model based on a physical constraint term and a data-driven term; wherein, the physical constraint term uses the shallow water equation as the physical constraint; Minimizing the composite loss function through the gradient backpropagation algorithm, and when the composite loss function is lower than a preset threshold, completing the training of the physics-informed neural network model.

4. The method according to claim 3, characterized in that, The constructing of the input layer and output layer of the physics-informed neural network model includes: Construct the input layer of the physical information neural network model, where the input layer includes at least one of the following: any sampling coordinate and sampling time in the target sea area ( x, y, t ), N T groups of harmonic parameters of each node on the open boundary; wherein, the sampling coordinates in the target sea area include at least one of the following: each node on the open boundary, tidal observation stations, and internal sea area points; Construct the output layer of the physical information neural network model, where the output layer includes at least one of the following: the simulated tide level at any sampling coordinate in the target sea area ζ ( x, y, t ), the velocity component U ( x, y, t ) and V ( x, y, t ), where the simulated tide level ζ ( x, y, t ), the velocity component U ( x, y, t ) and V ( x, y, t ) are used to form the simulated tide level time series data .

5. The method according to claim 3, characterized in that, The defining of the composite loss function of the physics-informed neural network model based on a physical constraint term and a data-driven term includes: Constructing an equation residual term based on the shallow water equation as the physical constraint term; Constructing a boundary residual term based on the difference between the measured tide level time series data and the simulated tide level time series data as the data-driven term; Defining the composite loss function of the physics-informed neural network model with the weighted sum of the physical constraint term and the data-driven term.

6. The method according to claim 5, wherein The equation residual term constructed based on the shallow water equation is: Among them, is the conservation variable vector of the shallow water equations; is the flux function composed of the simulated tidal level ζ ( x, y, t ) and the velocity components U ( x, y, t ) and V ( x, y, t ); is the source term function composed of the simulated tidal level ζ ( x, y, t ) and the velocity components U ( x, y, t ) and V ( x, y, t ); is the residual of the shallow water equations by the physics-informed neural network model at the sampling coordinates; N is the number of sampling coordinates in the target sea area; is the set of parameters to be trained of the physics-informed neural network model.

7. The method according to claim 5, wherein The boundary residual term constructed based on the difference between the measured tide level time series data and the simulated tide level time series data is: Wherein, is the simulated tide level output by the physical information neural network model at the tide gauge station and time ; is the measured tide level at the tide gauge station and time ; represents the mean square error term between the simulated tide level and the measured tide level.

8. The method according to any one of claims 5-7, characterized in that, The composite loss function of the physics-informed neural network model is: Among them, is a preset weight coefficient.

9. The method according to claim 3, wherein The minimizing of the composite loss function through the gradient backpropagation algorithm, and when the composite loss function is lower than a preset threshold, completing the training of the physics-informed neural network model includes: Training the physics-informed neural network model through the Adam optimizer; Based on the training data set, perform a number of random gradient iteration updates to optimize the set of parameters to be trained in the physics-informed neural network model ; During the training process, if the composite loss function is lower than a preset threshold or reaches the maximum number of iterations, terminate the training of the physics-informed neural network model and determine the pre-trained physics-informed neural network model.

10. The method according to claim 1, wherein The constructing of the training dataset based on the harmonic parameters and the simulated tide level time series data, inputting the training dataset into the deep neural network model in the hybrid neural network model, and training the deep neural network model includes: Take N T For each set of simulated tide level time series data among N sets of simulated tide level time series data, use it as an input tensor, and use the harmonic parameters corresponding to each set of simulated tide level time series data as output labels to construct a training dataset; Training the deep neural network model with the training dataset, and evaluating the deviation between the simulated harmonic parameters and the true harmonic parameters using a preset loss function. The parameters of the deep neural network model are optimized by the Adam optimizer, and when the training loss converges or reaches the maximum number of iterations, the training of the deep neural network model is completed.

11. The method according to claim 10, characterized in that, A batch normalization layer is connected after each of the multiple hidden layers of the deep neural network model to be used for normalizing the output features of each hidden layer.

12. The method according to claim 11, wherein The number of neurons in the input layer of the deep neural network model is M, and the number of neurons in the input layer is equal to the number of tide gauging stations; the number of neurons in the output layer of the deep neural network model is 2*N b , and the number of neurons in the output layer is equal to twice the number of nodes on the open boundary.

13. An electronic device, characterized in that, It includes: A memory for storing a computer program; A processor for implementing the steps of the method for simulating ocean tide levels based on a hybrid neural network model according to any one of claims 1 to 12 when executing the computer program.

14. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, wherein the computer program, when executed by a processor, implements the steps of the method for simulating ocean tide levels based on a hybrid neural network model according to any one of claims 1 to 12.

15. A computer program product, comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the method for simulating ocean tide levels based on a hybrid neural network model according to any one of claims 1 to 12.

Images