Marine physical information generation method, device, equipment and medium

Abstract

The invention discloses an ocean physical information generation method, device and equipment and a medium, and relates to the field of data generation, and the method comprises the steps: inputting a random noise vector used for information generation into an ocean physical information generation model to obtain generated ocean physical information; the marine physical information generation model determination method comprises the following steps: training a generative adversarial network by adopting actual marine physical information and a random noise vector for training and taking a minimum loss function as a target to obtain a marine physical information generation model; the generative adversarial network comprises a generator and a discriminator; the generator is a physical information neural network; the physical information neural network comprises an adaptive spectrum module, an attention module and a first neural network which are connected in sequence; the loss function at least comprises physical loss; the physical loss is determined by carrying out physical differential calculation on the ocean fluid in the ocean physical information output by the first neural network, the calculation cost can be reduced, and the reliability of the generated data is improved.

Description

Technical Field

[0001] The present application relates to the field of data generation, and in particular to a method, device, equipment and medium for generating ocean physical information. Background Art

[0002] At present, the generation and simulation of ocean physical information mainly rely on physical modeling and data-driven methods. Physical modeling methods (such as numerical simulation methods based on partial differential equations) can accurately describe ocean dynamic processes, but their computational cost is high and they are sensitive to complex and changing boundary conditions. Data-driven methods, whose statistical models are fitted by historical data, can quickly generate results, but when data is insufficient or environmental conditions change, their prediction effect decreases significantly, so their reliability is poor. Summary of the invention

[0003] The purpose of this application is to provide a method, device, equipment and medium for generating ocean physical information, which can reduce computing costs and improve the reliability of generated data.

[0004] To achieve the above objectives, this application provides the following solutions.

[0005] In a first aspect, the present application provides a method for generating ocean physical information, comprising: obtaining a random noise vector for information generation; inputting the random noise vector for information generation into an ocean physical information generation model to obtain generated ocean physical information.

[0006] Among them, the method for determining the ocean physical information generation model includes: obtaining actual ocean physical information and random noise vectors for training; using the actual ocean physical information and the random noise vectors for training, training the generative adversarial network with the goal of minimizing the loss function, and obtaining the ocean physical information generation model.

[0007] The generative adversarial network includes a generator and a discriminator.

[0008] The generator is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence.

[0009] The adaptive spectrum module is used to process the random noise vector used for training by fast Fourier transform; the attention module is used to operate on the processed data by using the attention mechanism; the first neural network is used to perform gradient operation on the operated data and output ocean physical information.

[0010] The discriminator is used to discriminate the ocean physical information output by the first neural network according to the actual ocean physical information to obtain a discriminant result.

[0011] The loss function at least includes: physical loss; the physical loss is determined by performing physical differential calculation on the ocean fluid in the ocean physical information output by the first neural network.

[0012] The trained generator is used as the ocean physical information generation model.

[0013] In the second aspect, the present application provides an ocean physical information generating device, including: a data acquisition module, used to obtain a random noise vector for information generation; a model determination module, used to determine the ocean physical information generating model; an information generation module, used to input the random noise vector for information generation into the ocean physical information generating model to obtain generated ocean physical information.

[0014] The model determination module includes: a data acquisition unit, which is used to acquire actual ocean physical information and random noise vectors for training; and a model training unit, which is used to train a generative adversarial network with the goal of minimizing a loss function using the actual ocean physical information and the random noise vectors for training, so as to obtain an ocean physical information generation model.

[0015] The generative adversarial network includes a generator and a discriminator.

[0016] The generator is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence.

[0017] The adaptive spectrum module is used to process the random noise vector used for training by fast Fourier transform; the attention module is used to operate on the processed data by using the attention mechanism; the first neural network is used to perform gradient operation on the operated data and output ocean physical information.

[0018] The discriminator is used to discriminate the ocean physical information output by the first neural network according to the actual ocean physical information to obtain a discriminant result.

[0019] The loss function at least includes: physical loss; the physical loss is determined by performing physical differential calculation on the fluid in the ocean physical information output by the first neural network.

[0020] The trained generator is used as the ocean physical information generation model.

[0021] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for generating ocean physical information.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned method for generating ocean physical information when executed by a processor.

[0023] According to the specific embodiments provided by the present application, the present application has the following technical effects: the present application provides a method, device, equipment and medium for generating ocean physical information, using actual ocean physical information and random noise vectors for training, training a generative adversarial network with the goal of minimizing the loss function, and obtaining an ocean physical information generation model, inputting the random noise vector used for information generation into the ocean physical information generation model to obtain generated ocean physical information, which can reduce computing costs; the generator in the generative adversarial network uses a physical information neural network, and introduces physical loss in the loss function, which can generate ocean physical information that conforms to the basic physical laws of actual ocean processes, while maintaining physical consistency, generating high-quality ocean physical information, and improving the reliability of generated data. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 This is a diagram of the application environment of a method for generating ocean physical information in one embodiment of the present application.

[0026] Figure 2 A schematic flow chart of a method for generating ocean physical information provided in accordance with an embodiment of the present application.

[0027] Figure 3 A flowchart of a method for determining an ocean physical information generation model provided in an embodiment of the present application.

[0028] Figure 4 A schematic diagram of the structure of a generative adversarial network provided in one embodiment of the present application.

[0029] Figure 5 A schematic diagram of the structure of a generator in a generative adversarial network provided in one embodiment of the present application.

[0030] Figure 6 A schematic diagram of functional modules of an ocean physical information generating device provided in another embodiment of the present application.

[0031] Figure 7 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0033] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0034] At present, although the relevant ocean physical information generation methods can meet the generation needs of ocean physical information to a certain extent, there are still some problems. Although the physical modeling method can generate data that conforms to the laws of physics, it consumes huge computing resources and is difficult to quickly respond to the needs of actual applications. Therefore, there is a problem of high computing cost. The complexity of the marine environment and the difficulty of data collection lead to limited available observation data. The statistical models of existing data-driven methods perform poorly when data is scarce. Therefore, there is a problem of data scarcity. Although traditional generative adversarial networks (GANs) can generate realistic data, due to the lack of physical constraints, the generated data often does not conform to the basic physical laws of actual ocean processes in a physical sense, which limits its reliability in practical applications. Therefore, there is a problem of lack of physical consistency. In addition, other existing models have limited generalization capabilities when processing ocean data from different geographical locations or time periods, and the generated results are not universal. Therefore, there is a problem of insufficient generalization capabilities.

[0035] In view of the above problems, this embodiment provides a new method that can combine the advantages of physical information and data-driven models to generate ocean physical information with strong physical consistency and high quality. By introducing Physics-Informed Neural Networks (PINN), not only can physical constraints be introduced in the GAN generation process, but also limited observation data can be effectively utilized to significantly improve the quality and practicality of generated data. PINN is a machine learning model that combines deep learning and physics knowledge. This embodiment combines PINN with GAN to generate ocean physical information, which solves many deficiencies in the existing technology and provides more reliable data support for marine scientific research and applications.

[0036] The method for generating ocean physical information provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the random noise vector used for information generation to the server 104. After the server 104 receives the random noise vector used for information generation, for the random noise vector used for information generation, the server 104 inputs the random noise vector used for information generation into the ocean physical information generation model to obtain the generated ocean physical information. The method for determining the ocean physical information generation model includes: obtaining actual ocean physical information and a random noise vector for training; using the actual ocean physical information and the random noise vector for training, training the generative adversarial network with the goal of minimizing the loss function, and obtaining the ocean physical information generation model. The generative adversarial network includes a generator and a discriminator. The generator is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence. The loss function at least includes: physical loss; the physical loss is determined by performing physical differential calculation on the ocean fluid in the ocean physical information output by the first neural network.

[0037] The server 104 may feed back the generated ocean physical information to the terminal 102. In addition, in some embodiments, the ocean physical information generation method may also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 may directly process the random noise vector used for information generation, or the server 104 may obtain the random noise vector used for information generation from the data storage system and process the random noise vector used for information generation.

[0038] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0039] In an exemplary embodiment, Figure 2 As shown, a method for generating ocean physical information is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, and the steps include the following steps 201 to 202.

[0040] Step 201, obtaining a random noise vector for information generation.

[0041] Step 202, input the random noise vector used for information generation into the ocean physical information generation model to obtain generated ocean physical information. The ocean physical information includes: coastline cyclone index, revetment ratio, wetland ratio, temperature and humidity field, ocean velocity field, pressure field, salinity and coastal zone resilience evaluation results, etc.

[0042] like Figure 3 As shown, the method for determining the ocean physical information generation model includes the following steps.

[0043] Step 301, obtaining actual ocean physical information and a random noise vector for training.

[0044] Step 302, using actual ocean physical information and a random noise vector for training, training the generative adversarial network with the goal of minimizing the loss function, and obtaining an ocean physical information generation model.

[0045] The generative adversarial network includes a generator and a discriminator. The generator is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence. The trained generator is used as the ocean physical information generation model.

[0046] The adaptive spectrum module is used to process the random noise vector used for training by using Fast Fourier Transform (FFT); the attention module is used to operate on the processed data by using the attention mechanism; the first neural network is used to perform gradient operation on the operated data and output ocean physical information.

[0047] The discriminator is used to discriminate the ocean physical information output by the first neural network according to the actual ocean physical information to obtain a discriminant result.

[0048] The loss function at least includes: physical loss; the physical loss is determined by performing physical differential calculation on the ocean fluid in the ocean physical information output by the first neural network.

[0049] In another exemplary embodiment of the present application, the loss function includes: generator loss and discriminator loss; the generator loss includes: generator loss itself, cross entropy loss and physical loss.

[0050] Step 302 specifically includes the following steps.

[0051] (1) Preprocess and encode the actual ocean physical information to obtain the encoded data.

[0052] Specifically: ① Process missing values ​​and outliers of actual marine physical information to obtain preprocessed data. ② One-hot encode discrete variables in the preprocessed data (such as coastal resilience assessment results) to obtain first encoded data. One-hot encoding can convert the discrete variable into an independent binary vector to adapt to the input of the generative adversarial network. ③ Normalize or standardize the continuous variables in the preprocessed data to obtain processed continuous data. The purpose of normalizing or standardizing the continuous data in this step is to convert the data into a normal distribution with a mean of 0 and a standard deviation of 1 or to scale the data to [0, 1]. In order to deal with the problem of uneven category distribution, each category is standardized separately. ④ Binary conversion encoding is performed on the processed continuous data to obtain second encoded data. This step converts the categorical variable into an independent binary vector through encoding. If the marine physical information categorical variable has n different categories, the variable will be converted into a binary vector of length n, in which only one position is 1 and the rest are 0. ⑤ The first encoded data and the second encoded data are determined as encoded data.

[0053] Among them, the process of processing missing values ​​and outliers is: using the box plot method to eliminate outliers or truncate them; for actual ocean physical information with missing values, use mean filling, median filling or regression to fill in. Processing missing values ​​helps to avoid the performance degradation of the generative adversarial network due to missing data during the training process.

[0054] Among them, the normalization processing is implemented by mode-specific normalization. Specifically: first, for each ocean physical information, the frequency of each category is counted. Assuming that a categorical variable has multiple categories, the frequency of each category is counted. For each continuous variable, its mean and standard deviation are calculated.

[0055] The process of standardization is: using the formula Implementation, where is the mean, is the variance. In the data generation phase, the generator outputs a standardized continuous variable At this time, you need to Denormalize to return to the original data scale, , is the original data.

[0056] (2) For the ath iteration in the training process, the random noise vector used for training in the ath iteration is input into the physical information neural network, and the first neural network in the physical information neural network outputs the ocean physical information of the ath iteration; wherein a>1.

[0057] (3) The encoded data and the ocean physical information of the a-th iteration are input into the discriminator, and the discriminator outputs the identification result of the a-th iteration.

[0058] (4) Determine the generator loss of the a-th iteration according to the ocean physical information of the a-th iteration output by the first neural network, determine the cross entropy loss of the a-th iteration according to the ocean physical information of the a-th iteration output by the first neural network and the encoded data, and determine the physical loss of the a-th iteration by performing physical differential calculation on the fluid in the ocean physical information of the a-th iteration output by the first neural network.

[0059] (5) Determine the generator loss of the a-th iteration based on the generator loss of the a-th iteration, the cross entropy loss of the a-th iteration, and the physical loss of the a-th iteration; determine the discriminator loss of the a-th iteration based on the identification result of the a-th iteration.

[0060] Specifically, the calculation formula of the generator loss is as follows.

[0061] .

[0062] in, represents the generator loss; Represents the loss of the generator itself; Represents the cross entropy loss, which is used to measure the difference between the ocean physical information generated by the generator and the real data distribution in some feature dimensions; Indicates physical loss; represents the i-th ocean physical information generated; m represents the number of ocean physical information generated; Represents the identification result of the i-th ocean physical information generated by the discriminator.

[0063] The physical loss is determined by performing physical differential calculation on the ocean fluid in the ocean physical information output by the first neural network. The physical differential equation calculated is the Navier-Stokes differential equation, which is expressed as follows.

[0064] .

[0065] represents the ocean fluid velocity field in the generated ocean physical information; t represents time; represents partial derivative, represents the gradient operator, ( , ), (x, y, z) represent the spatial coordinates of a point in the three-dimensional space defined by the physical quantity; Represents the kinematic viscosity of ocean fluids in the generated ocean physics information; Represents the ocean fluid density in the generated ocean physics information; Represents the ocean fluid pressure field in the generated ocean physical information; Represents the external force corresponding to the generated ocean physical information.

[0066] According to the above Navier-Stokes differential equation, the generated ocean physical information should conform to the physical laws therein, so the physical loss is taken as part of the loss function as a physical constraint. The calculation formula of the physical loss is as follows.

[0067] .

[0068] in, represents the ocean fluid velocity field in the i-th ocean physical information generated; represents the ocean fluid pressure field in the i-th generated ocean physical information; Represents the external force corresponding to the i-th generated ocean physical information.

[0069] The calculation formula of the discriminator loss is as follows.

[0070] .

[0071] represents the discriminator loss; Represents the actual i-th ocean physical information; Represents the identification result of the discriminator on the actual i-th ocean physical information.

[0072] (6) If the difference between the generator loss of the a-th iteration and the generator loss of the previous iteration is within the first set difference range, and the difference between the discriminator loss of the a-th iteration and the discriminator loss of the previous iteration is within the second set difference range, then the loss function of the a-th iteration is determined to be the smallest, and the generator after the a-th iteration is used as the ocean physical information generation model, otherwise, after updating the number of iterations, the next iteration is performed. This embodiment implements model training through forward propagation and reverse update.

[0073] In another exemplary embodiment of the present application, the generator and the discriminator are introduced in detail.

[0074] The generator in this embodiment is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence. The adaptive spectrum module is residually connected to the first neural network. The purpose of the residual connection is to prevent the degradation of the deep neural network. By introducing jump connections to alleviate the gradient vanishing problem, deeper networks can be trained more effectively, and the training efficiency, convergence speed, generalization ability and overall performance of the model are improved. The first neural network includes: a first input layer, a hidden layer and a first output layer; the hidden layer includes 16 layers of hidden neurons, wherein the last layer of hidden neurons and the middle layer of hidden neurons are residually connected to improve the training of the model. The discriminator is a second neural network, which includes: a second input layer, a hidden layer and a second output layer. The second input layer has the same structure as the first input layer, and the second output layer has a different structure from the first output layer.

[0075] Set the initial complex weights in the adaptive spectrum module ( and ) is a random floating point tensor with a shape of (dim, 2). Dim indicates the attribute of obtaining the tensor dimension. The shape of the random floating point tensor is a two-dimensional tensor. The initial complex weights are multiplied by 0.02 and scaled, and further initialized using a truncated normal distribution with a standard deviation of 0.02. The initialization threshold parameter is a random floating point tensor with a shape of (dim, 1). The shape of the random floating point tensor is a one-dimensional tensor. The initialization complex weights satisfy .

[0076] The adaptive spectrum module receives an input tensor of shape (B, N, C) which is a random noise vector, where B represents the batch size, N represents the length of the sequence, and C represents the number of features.

[0077] FFT is used to transform the random noise vector (which is a time domain signal) into a frequency domain signal. The calculation formula is as follows.

[0078] .

[0079] is the frequency domain signal; is the time domain signal; represents an imaginary unit; Represents the index of the frequency domain; Represents the index of the time domain. When performing frequency domain conversion, the length N of the sequence is the signal length in FFT.

[0080] The frequency domain weighting is performed below. The frequency domain weighting is to multiply the frequency domain signal obtained by FFT by the complex weight. The calculation formula is as follows.

[0081] .

[0082] in, represents the frequency domain weighted result; Is and The constructed weight sequence for frequency domain weighting; and Corresponding to the weights of the real and imaginary parts respectively.

[0083] The energy calculation is as follows: .in, Indicates energy.

[0084] The energy is normalized as follows: .in, represents the normalized energy, is the median energy, is a very small constant to avoid the denominator being zero. Compare and subtract the normalized energy and the threshold parameter to obtain the difference tensor. Convert the difference tensor to a floating point tensor, and then perform a sigmoid operation on the floating point tensor to convert it into a soft mask between (0~1) , the calculation formula is: , c represents the floating-point tensor converted from the difference tensor.

[0085] Then calculate the masked frequency domain signal based on the soft mask , the calculation formula is: .in, represents the amplitude in the frequency domain, Represents the phase in the frequency domain. The real part of and the imaginary part The expression is: ; .

[0086] Masked frequency domain signal It can be converted back to the time domain signal through the inverse fast Fourier transform (IFFT), and the calculation formula is: , It is the converted time domain signal, that is, the processed data output by the adaptive spectrum module.

[0087] Next, the processed data output by the adaptive spectrum module Entering the attention module, the attention module is mapped to a space three times the original feature dimension through the QKV linear layer. The calculation formula is: , , ,in: , is the query matrix, is the key matrix, is the value matrix, , and Respectively , and The corresponding linear transformation matrix is, , express OK The set of real numbers of columns, express OK The set of real numbers of columns, The value of is the total number of samples in a batch. The value of is the input feature dimension of each sample or sequence element, The value of is the new feature dimension of each sample or sequence element after linear transformation.

[0088] Compute all and The dot product between , is the transpose of the key matrix. The result of this step is a score matrix where each element represents the dot product score between an element in the query matrix and an element in the key matrix, calculated as follows.

[0089] .

[0090] .

[0091] represents the score matrix, represents the calculated attention weight, represents the query vector, represents the first key vector, Indicates key vectors, Indicates key vectors, Represents the indices corresponding to different elements of the key vector, represents the dimension of the key vector, represents the set of elements of the key vector, represents the normalized exponential function.

[0092] For each row of the score matrix (i.e. each Corresponding scores), select the k1 elements with the highest scores. The selection method is to randomly select an element as the reference value for each row of scores, divide the row of scores into two parts, one part is less than the reference value, and the other part is greater than or equal to the reference value. According to the position of the reference value, determine whether it is necessary to recursively continue searching in a certain part. Specifically, if the reference value is exactly the K1th largest score, the search ends; if the position of the reference value is greater than K1, continue to search for the K1th largest score in the part less than the reference value; if the position of the reference value is less than K1, continue to search for the (K1-left)th largest score in the part greater than or equal to the reference value, where left is the number of positions less than the reference value. Repeat the above steps until the K1th largest score is found, where the selected K1 values ​​are expressed as follows.

[0093] .

[0094] in, Indicates that the processed matrix is ​​located at Line Elements of a column; Represents the original score matrix Located in the middle Line Elements of a column; Represented in the original score matrix No. The highest score in the row A collection of elements.

[0095] According to the above The softmax function is used to convert the score into a normalized attention weight. The non-highest The scores are set to very small numbers, and their contribution in the softmax function will be close to zero. Then, these attention weights are combined with the original value matrix Multiply them together to get the weighted value matrix, which is the output of the attention module. The calculation formula is as follows.

[0096] .

[0097] in, The calculated data output by the attention module.

[0098] The calculated data is then input into the first neural network, and the calculation formula of each neuron is as follows.

[0099] .

[0100] in, represents the output of the neuron; represents weight; represents the input of the neuron; Indicates bias.

[0101] Gradient calculation is performed according to the calculation formula of each neuron, and the calculation formula is as follows.

[0102] .

[0103] .

[0104] in, Represents the output of the first hidden layer in the first neural network; Represents the output of the second hidden layer in the first neural network; Represents the output of the third hidden layer in the first neural network; represents the output of the 4th hidden layer in the first neural network; represents the output of the 5th hidden layer in the first neural network; represents the output of the 15th hidden layer in the first neural network; Represents the output of the 16th hidden layer in the first neural network.

[0105] Each gradient calculation implements one iteration: ; . represents the weight parameters between the neurons in the first input layer and the neurons in the hidden layer; represents the weight parameters between the neurons in the hidden layer and the neurons in the first output layer; represents the learning rate; Represents the partial derivative of the loss function with respect to the weights.

[0106] The structure of the generated adversarial network is as follows Figure 4 As shown, the structure of the generator in the generative adversarial network is as follows Figure 5 shown.

[0107] In another exemplary embodiment of the present application, after step 201, the ocean physical information generation method further includes: using actual ocean physical information to decode and verify the generated ocean physical information, and determining the verified ocean physical information as the finally generated ocean physical information.

[0108] Specifically, the decoding verification process is: extracting the characteristic statistics of the actual ocean physical information to obtain the first characteristic statistics; extracting the characteristic statistics of the generated ocean physical information to obtain the second characteristic statistics; determining the fidelity of the first characteristic statistics and the second characteristic statistics, and if the fidelity reaches a set degree value, the generated ocean physical information is determined as the verified ocean physical information. Among them, the characteristic statistics include: mean, standard deviation and correlation coefficient, etc.

[0109] This application compares the characteristic statistics of the generated ocean physical information and the actual ocean physical information, and can evaluate the fidelity of the generated ocean physical information in different physical quantities.

[0110] This application combines physical information neural networks with generative adversarial networks to achieve the generation of ocean physical information. It can generate high-quality ocean physical information while maintaining physical consistency, effectively solving many problems existing in the prior art.

[0111] Specifically, the ocean physical information generation method of this application can provide high-quality data generation solutions in the fields of ocean data loss, environmental monitoring, climate forecasting, etc., significantly improving the reliability and applicability of data. By introducing PINN, the constraints of physical equations are added in the data generation process to ensure that the generated data is not only realistic in statistical characteristics, but also consistent in physical laws, significantly improving the reliability and practicality of the generated data.

[0112] The ocean physical information generation method of this application can generate high-quality data when data is scarce, and use limited observation data to train a generation model with physical consistency, effectively making up for the problem of insufficient data. By combining PINN and GAN, the computational efficiency of data generation is greatly improved while maintaining physical accuracy. The physical constraints of PINN combined with the generation ability of GAN make it possible to generate ocean data that conforms to physical laws without complex numerical simulations, thereby reducing computational costs.

[0113] The ocean physical information generation method of the present application can not only generate diversified data, but also adapt well to different environmental conditions, and improve the generalization ability of the model. In this way, the generated data can better reflect the changes in different marine environments and have a wider range of applicability. PINN has its own physical correction function in the data generation process to ensure that the generated data conforms to the predetermined physical laws during the generation stage. Combined with the statistically realistic data generated by GAN, this method can automatically perform physical verification and correction during the generation process, reducing additional post-processing requirements and improving the accuracy and reliability of the data. The ocean physical information generation method combining PINN and GAN has significant advantages in improving physical consistency, enhancing scarce data processing capabilities, optimizing computing efficiency, improving generalization capabilities, and accurately verifying generated data. It can provide higher quality simulation data support for marine scientific research and applications.

[0114] In practical applications, an implementation process of the above-mentioned ocean physical information generation method can be described as follows: obtain part of the ocean physical information; input part of the ocean physical information into a generative adversarial network based on a physical information neural network for training; use the generator in the trained generative adversarial network (i.e., the ocean physical information generation model) to sample and generate high-quality ocean physical information that conforms to physical laws.

[0115] Among them, the training process of the generative adversarial network based on physical information neural network includes: first, segmenting part of the acquired ocean physical information into training set, validation set and test set. The loss function uses cross entropy loss and physical loss to ensure that the generated data conforms to the physical laws. Small batch stochastic gradient descent is used. In each training iteration, the discriminator is trained first, and then the generator is trained. The iterations are repeated until the model converges. The model performance is monitored through cross-validation and early stopping strategies to prevent overfitting. Random sampling and noise perturbation are used to ensure the diversity of data generation.

[0116] Data is generated by sampling a random noise vector from a standard normal distribution and combining it with a conditional variable before inputting it into a generator; the generator receives the random noise vector and the conditional variable and outputs ocean physical information that meets the set conditions.

[0117] Based on the same inventive concept, the embodiment of the present application also provides an ocean physical information generating device for implementing the ocean physical information generating method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the ocean physical information generating device provided below can refer to the limitations of the ocean physical information generating method above, and will not be repeated here.

[0118] In an exemplary embodiment, Figure 6As shown, a device for generating ocean physical information is provided, including: a data acquisition module 601, used to acquire a random noise vector for information generation; a model determination module 602, used to determine an ocean physical information generation model; and an information generation module 603, used to input the random noise vector for information generation into the ocean physical information generation model to obtain generated ocean physical information.

[0119] The model determination module 602 includes: a data acquisition unit for acquiring actual ocean physical information and a random noise vector for training; and a model training unit for training a generative adversarial network with the actual ocean physical information and the random noise vector for training to minimize the loss function, thereby obtaining an ocean physical information generation model.

[0120] The generative adversarial network includes a generator and a discriminator.

[0121] The generator is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence.

[0122] The adaptive spectrum module is used to process the random noise vector used for training by fast Fourier transform; the attention module is used to operate on the processed data by using the attention mechanism; the first neural network is used to perform gradient operation on the operated data and output ocean physical information.

[0123] The discriminator is used to discriminate the ocean physical information output by the first neural network according to the actual ocean physical information to obtain a discriminant result.

[0124] The loss function at least includes: physical loss; the physical loss is determined by performing physical differential calculation on the fluid in the ocean physical information output by the first neural network.

[0125] The trained generator is used as the ocean physical information generation model.

[0126] This application uses a technology that combines PINN and GAN to achieve the generation of ocean physical information. PINN is used to model the physical laws in the marine environment, and the physical consistency of the generated data is ensured by introducing physical constraints. On this basis, GAN generates realistic ocean information data through adversarial training. Specifically: a small amount of ocean information data such as coastline cyclone index, revetment ratio, wetland ratio, temperature and humidity field is obtained, and the trained model is obtained by inputting GAN based on physical information neural network for training, and then the spatiotemporal data of ocean physical information is supplemented by sampling. This application can provide high-quality data generation solutions in the fields of marine data loss, environmental monitoring, climate forecasting, etc. Combining the physical constraints of PINN with the generation capabilities of GAN can better simulate these complex processes, so that the generated data can capture complex spatiotemporal dynamic changes, which helps to realize efficient real-time prediction and monitoring models, improve response speed, and significantly improve the reliability and applicability of data.

[0127] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store random noise vectors for information generation. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for generating ocean physical information is implemented.

[0128] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0129] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0130] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0131] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0132] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0133] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, etc., but is not limited thereto.

[0134] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0135] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for generating ocean physical information, characterized in that: The method for generating ocean physical information comprises: Get the random noise vector for information generation; Inputting the random noise vector used for information generation into the ocean physical information generation model to obtain generated ocean physical information; Wherein, the method for determining the ocean physical information generation model includes: Obtain actual ocean physics information and random noise vectors for training; Using actual ocean physics information and random noise vectors for training, the generative adversarial network is trained with the goal of minimizing the loss function to obtain an ocean physics information generation model. The generative adversarial network includes a generator and a discriminator; The generator is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence; The adaptive spectrum module is used to process the random noise vector used for training by using fast Fourier transform; the attention module is used to operate on the processed data by using the attention mechanism; the first neural network is used to perform gradient operation on the operated data and output ocean physical information; The discriminator is used to discriminate the ocean physical information output by the first neural network according to the actual ocean physical information to obtain a discriminant result; The loss function at least includes: physical loss; the physical loss is determined by performing physical differential calculation on the ocean fluid in the ocean physical information output by the first neural network; The trained generator is used as the ocean physical information generation model.

2. The method for generating ocean physical information according to claim 1, characterized in that: The loss function includes: generator loss and discriminator loss; the generator loss includes: generator loss itself, cross entropy loss and physical loss.

3. The method for generating ocean physical information according to claim 2, characterized in that: Using actual ocean physics information and random noise vectors for training, the generative adversarial network is trained with the goal of minimizing the loss function to obtain an ocean physics information generation model, which specifically includes: Preprocessing and data encoding the actual ocean physical information to obtain encoded data; For the ath iteration in the training process, the random noise vector used for training in the ath iteration is input into the physical information neural network, and the first neural network in the physical information neural network outputs the ocean physical information of the ath iteration; wherein a>1; Input the encoded data and the ocean physical information of the a-th iteration into the discriminator, and the discriminator outputs the identification result of the a-th iteration; Determine the generator loss of the a-th iteration according to the ocean physical information of the a-th iteration output by the first neural network, determine the cross entropy loss of the a-th iteration according to the ocean physical information of the a-th iteration output by the first neural network and the encoded data, and perform physical differential calculation on the fluid in the ocean physical information of the a-th iteration output by the first neural network to determine the physical loss of the a-th iteration; Determine the generator loss of the a-th iteration according to the generator loss of the a-th iteration, the cross entropy loss of the a-th iteration, and the physical loss of the a-th iteration; Determine the discriminator loss of the a-th iteration according to the identification result of the a-th iteration; If the difference between the generator loss of the a-th iteration and the generator loss of the previous iteration is within the first set difference range, and the difference between the discriminator loss of the a-th iteration and the discriminator loss of the previous iteration is within the second set difference range, then the loss function of the a-th iteration is determined to be the smallest, and the generator after the a-th iteration is used as the ocean physical information generation model; otherwise, the next iteration is performed after updating the number of iterations.

4. The method for generating ocean physical information according to claim 3, characterized in that: Preprocess and encode the actual ocean physical information to obtain the encoded data, including: Process missing values ​​and outliers on actual ocean physical information to obtain preprocessed data; Performing one-hot encoding on discrete variables in the preprocessed data to obtain first encoded data; Normalizing or standardizing the continuous variables in the preprocessed data to obtain processed continuous data; Performing binary conversion encoding on the processed continuous data to obtain second encoded data; The first encoded data and the second encoded data are determined as encoded data.

5. The method for generating ocean physical information according to claim 1, characterized in that: After inputting the random noise vector used for information generation into the ocean physical information generation model to obtain the generated ocean physical information, the ocean physical information generation method further includes: The generated ocean physical information is decoded and verified using actual ocean physical information, and the verified ocean physical information is determined as the finally generated ocean physical information.

6. The method for generating ocean physical information according to claim 5, characterized in that: The generated ocean physical information is decoded and verified using actual ocean physical information, including: Extracting characteristic statistics of actual ocean physical information to obtain first characteristic statistics; Extracting characteristic statistics of the generated ocean physical information to obtain a second characteristic statistic; The fidelity of the first characteristic statistic and the second characteristic statistic is determined, and if the fidelity reaches a set degree value, the generated ocean physical information is determined as verified ocean physical information.

7. The method for generating ocean physical information according to claim 2, characterized in that: The calculation formula of the generator loss is: ； in, represents the generator loss; Represents the loss of the generator itself; represents the cross entropy loss; Indicates physical loss; represents the i-th ocean physical information generated; m represents the number of ocean physical information generated; represents the identification result of the i-th ocean physical information generated by the discriminator; The calculation formula of the physical loss is: ； in, represents the ocean fluid velocity field in the i-th ocean physical information generated; t represents time; represents the gradient operator; It means to find partial derivative; represents the ocean fluid pressure field in the i-th generated ocean physical information; Represents the kinematic viscosity of ocean fluids in the generated ocean physics information; Represents the ocean fluid density in the generated ocean physics information; represents the external force corresponding to the i-th ocean physical information generated; The discriminator loss is calculated as: ； represents the discriminator loss; Represents the actual i-th ocean physical information; Represents the identification result of the discriminator on the actual i-th ocean physical information.

8. An ocean physical information generating device, characterized in that: The ocean physical information generating device comprises: A data acquisition module, used for acquiring a random noise vector for information generation; A model determination module, used for determining a model for generating ocean physical information; An information generation module, used for inputting a random noise vector used for information generation into an ocean physical information generation model to obtain generated ocean physical information; The model determination module comprises: A data acquisition unit, used to acquire actual ocean physical information and random noise vectors for training; A model training unit, used to train the generative adversarial network with the goal of minimizing the loss function by using actual ocean physical information and a random noise vector used for training, so as to obtain an ocean physical information generation model; The generative adversarial network includes a generator and a discriminator; The generator is a physical information neural network; the physical information neural network includes: an adaptive spectrum module, an attention module and a first neural network connected in sequence; The adaptive spectrum module is used to process the random noise vector used for training by using fast Fourier transform; the attention module is used to operate on the processed data by using the attention mechanism; the first neural network is used to perform gradient operation on the operated data and output ocean physical information; The discriminator is used to discriminate the ocean physical information output by the first neural network according to the actual ocean physical information to obtain a discriminant result; The loss function at least includes: physical loss; the physical loss is determined by performing physical differential calculation on the fluid in the ocean physical information output by the first neural network; The trained generator is used as the ocean physical information generation model.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for generating ocean physical information according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating ocean physical information described in any one of claims 1 to 7 is implemented.

Images