A marine environment state prediction method, device, equipment, medium and computer program product

Abstract

The application provides a marine environment state prediction method, device, equipment, medium and computer program product, and relates to the technical field of ocean engineering. The method acquires a first marine data set through an underwater sensor group, which includes the space-time coordinates of preset sampling points and corresponding marine environment characteristics; the first marine data set is input into a preset marine environment prediction model to determine a marine environment state prediction result. The generation module of the model fits each marine environment characteristic based on the space-time coordinates to generate a second marine data set with higher resolution in the spatial dimension; the prediction module of the model extracts the space-time correlation of the second marine data set and the internal correlation between each marine environment characteristic and outputs the corresponding prediction result. The data sparsity problem is solved by the generation module first, and then the correlation between the space-time and multiple environment characteristics is captured by the prediction module, thereby significantly improving the accuracy and stability of marine environment state prediction.

Description

Technical Field

[0001] This application relates to the field of marine engineering technology, and in particular to a method, apparatus, equipment, medium, and computer program product for predicting marine environmental conditions. Background Technology

[0002] The ocean covers most of the Earth, possesses a vast ecosystem, and is rich in mineral resources. To better explore and develop marine resources and ensure the safety of maritime activities, it is typically necessary to continuously and accurately monitor and predict the state of the marine environment. In practical applications, underwater wireless sensor networks are often deployed to collect marine environmental data and use this data to predict the state of the marine environment.

[0003] However, underwater sensors are highly susceptible to seawater erosion, biofouling, and physical damage, resulting in high deployment costs, difficult maintenance, and limited lifespan. Consequently, the deployment density of underwater sensors is low, and the raw data acquired is spatially sparse. Furthermore, marine environmental data is a typical spatiotemporal data type; its trends not only depend on historical time series evolution but also interact with environmental conditions at different depths and horizontal positions, and complex physical coupling relationships exist between various environmental characteristics. Existing prediction methods typically employ single time series analysis techniques, which struggle to effectively capture the inherent correlations between these temporal, spatial, and multi-environmental characteristics simultaneously, leading to insufficient accuracy and stability in the prediction results. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, device, medium, and computer program product for predicting marine environmental conditions, aiming to solve the technical problem of how to improve the accuracy and stability of marine environmental condition prediction.

[0005] To achieve the above objectives, this application provides a method for predicting the state of the marine environment, the steps of which include: The first ocean dataset is acquired by an underwater sensor array. The first ocean dataset includes the spatiotemporal coordinates of preset sampling points and the corresponding marine environmental features. The first marine dataset is input into a preset marine environment prediction model to determine the marine environment state prediction result. The preset marine environment prediction model includes a generation module and a prediction module. The generation module is used to fit each of the marine environment features based on the spatiotemporal coordinates to generate a second marine dataset with higher resolution in the spatial dimension. The prediction module is used to extract the spatiotemporal correlation of the second marine dataset and the intrinsic correlation between each of the marine environment features, and output the corresponding marine environment state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

[0006] In one embodiment, the steps of the marine environmental state prediction method further include: Based on historical spatiotemporal coordinates and corresponding historical marine environmental characteristics, spatiotemporal feature submatrices and environmental variable submatrices are constructed respectively. The spatiotemporal feature submatrix is ​​input into a preset multilayer perceptron neural network, and the environmental variable submatrix is ​​used as the prediction target to train the preset multilayer perceptron neural network to learn the nonlinear mapping relationship between the historical spatiotemporal coordinates and each of the historical marine environmental features. The generation module is constructed based on a pre-trained multilayer perceptron neural network.

[0007] In one embodiment, the step of training the preset multilayer perceptron neural network to learn the nonlinear mapping relationship between the historical spatiotemporal coordinates and each of the historical marine environmental features further includes: In the loss function of the preset multilayer perceptron neural network, the sum of squares of all weight parameters of the network is added as a penalty term to constrain the value of each weight parameter.

[0008] In one embodiment, the steps of the marine environmental state prediction method further include: The historical second ocean dataset was determined by using historical spatiotemporal coordinates and corresponding historical marine environmental characteristics. Based on the aforementioned historical second ocean dataset, a multivariate time series matrix is ​​constructed. The multivariate time series matrix contains multiple time steps in the time dimension, and each time step contains the values ​​of the aforementioned ocean environmental features at different depths. The multivariate time series matrix is ​​input into a preset multivariate convolutional long short-term memory neural network, and the historical second ocean environmental features at the corresponding time are used as prediction targets to train the preset multivariate convolutional long short-term memory neural network to learn the spatiotemporal correlation in the historical second ocean dataset and the intrinsic correlation between the historical ocean environmental features. The prediction module is constructed based on a pre-trained multivariate convolutional long short-term memory neural network.

[0009] In one embodiment, the step of training the preset multivariate convolutional long short-term memory neural network to learn the spatiotemporal correlations in the historical second ocean dataset and the intrinsic correlations among the various historical ocean environmental features further includes: A random deactivation technique is used to randomly deactivate neurons in the preset multivariate convolutional long short-term memory neural network with a preset probability.

[0010] In one embodiment, the spatiotemporal coordinates include time coordinates, horizontal position coordinates, and depth position coordinates; the generation module has a preset multilayer perceptron neural network; the prediction module has a preset multivariate convolutional long short-term memory neural network; and the step of inputting the first marine dataset into the preset marine environment prediction model to determine the marine environment state prediction result specifically includes: The generation module determines the various depths to be predicted based on preset application requirements. The generation module extracts the horizontal position coordinates and the time coordinates from the first ocean dataset. The generation module combines the horizontal position coordinates and the time coordinates with each of the depths to be predicted to generate corresponding new spatiotemporal coordinates. The new spatiotemporal coordinates are converted into prediction environment features corresponding to each prediction depth through the preset multilayer perceptron neural network of the generation module. The generation module merges the first ocean dataset with each of the predicted environmental features, and retains the original depth coordinates and corresponding ocean environmental features of the first ocean dataset to form a second ocean dataset with higher resolution in the vertical depth direction. The spatiotemporal correlation and the intrinsic correlation are extracted from the second ocean dataset through the preset multivariate convolutional long short-term memory neural network of the prediction module. The prediction module outputs the marine environmental state prediction results based on the spatiotemporal correlation and the intrinsic correlation.

[0011] Furthermore, to achieve the above objectives, this application also provides a marine environmental state prediction device, the marine environmental state prediction device comprising: The raw data acquisition module is used to acquire a first ocean dataset, which includes the spatiotemporal coordinates of preset sampling points and the corresponding marine environmental features. The marine environment prediction module is used to input the first marine dataset into a preset marine environment prediction model to determine the marine environment state prediction result. The preset marine environment prediction model includes a generation module and a prediction module. The generation module is used to fit each of the marine environment features based on the spatiotemporal coordinates to generate a second marine dataset with higher resolution in the spatial dimension. The prediction module is used to extract the spatiotemporal correlation of the second marine dataset and the intrinsic correlation between each of the marine environment features, and output the corresponding marine environment state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

[0012] In addition, to achieve the above objectives, this application also provides a marine environmental state prediction device, which includes: a memory, a processor, and a marine environmental state prediction program stored in the memory and executable on the processor, wherein the marine environmental state prediction program is configured to implement the steps of the marine environmental state prediction method as described above.

[0013] In addition, to achieve the above objectives, this application also provides a storage medium, which is a computer-readable storage medium, on which a marine environment state prediction program is stored, and when the marine environment state prediction program is executed by a processor, it implements the steps of the marine environment state prediction method as described above.

[0014] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the marine environmental state prediction method described above.

[0015] This application provides a method, apparatus, device, medium, and computer program product for predicting marine environmental state. The method acquires a first marine dataset using an underwater sensor array, which includes the spatiotemporal coordinates of preset sampling points and corresponding marine environmental features. The first marine dataset is then input into a preset marine environmental prediction model to determine the predicted marine environmental state. Specifically, the model's generation module fits the marine environmental features based on the spatiotemporal coordinates to generate a second marine dataset with higher spatial resolution. The model's prediction module extracts the spatiotemporal correlations of the second marine dataset and the intrinsic correlations between various marine environmental features, outputting the corresponding prediction results. By first addressing the data sparsity problem through the generation module and then capturing the correlations between spatiotemporal and multiple environmental features through the prediction module, the accuracy and stability of marine environmental state prediction are significantly improved. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the marine environmental state prediction method of this application. Figure 2This is a flowchart illustrating the construction and generation module in Embodiment 2 of the marine environmental state prediction method of this application; Figure 3 This is a structural diagram of the main structure of the generation module; Figure 4 This is a flowchart illustrating the construction of the prediction module in Embodiment 2 of the marine environmental state prediction method of this application; Figure 5 This is a schematic diagram of the prediction module. Figure 6 This is a schematic diagram of the process for predicting the marine environment state in Embodiment 2 of the marine environment state prediction method of this application; Figure 7 A schematic diagram illustrating the basic framework of the marine environmental state prediction process; Figure 8 This is a schematic diagram of the module structure of Embodiment 1 of the marine environmental state prediction device of this application; Figure 9 This is a schematic diagram of the structure of Embodiment 1 of the marine environmental state prediction device of this application.

[0019] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] This application presents a marine environmental state prediction method according to a first embodiment. Please refer to [link / reference]. Figure 1 The steps of the marine environmental state prediction method include S10-S20: Step S10: Acquire a first ocean dataset by using an underwater sensor array. The first ocean dataset includes the spatiotemporal coordinates of preset sampling points and the corresponding marine environmental features. It should be understood that, in this embodiment, the executing entity of this application can be a marine environmental state prediction device or equipment, specifically a computer mounted on a surface vessel, an underwater autonomous vehicle, or a shore-based server, which can receive data transmitted back from the underwater sensor group and execute prediction programs.

[0023] It should be noted that, in this embodiment, the underwater sensor group refers to a network composed of multiple sensor nodes deployed at different locations and depths in the target sea area. Each node typically includes sensors for temperature, salinity, pressure, etc., used to collect marine environmental parameters in real time. The first marine dataset refers to the collection of raw observation data actually collected and uploaded by each sensor node corresponding to the underwater sensor group. The preset sampling point refers to the actual location of the underwater sensor group. Spatiotemporal coordinates include time coordinates and spatial coordinates. The actual coordinates refer to the specific time data corresponding to the time when the raw observation data was collected, such as year, month, day, etc., while the spatial coordinates refer to the sampling point location data corresponding to the time when the raw observation data was collected, such as horizontal position (longitude, latitude) and depth position. Marine environmental characteristics refer to physical parameters that can characterize the state of the marine environment, including but not limited to temperature, salinity, density, sound speed, etc.

[0024] It should be understood that, in this embodiment, the first ocean dataset is raw observation data without any interpolation or generation, which can truly reflect the observations of the underwater sensor group at a limited number of points in time and space. Because underwater sensors are highly susceptible to seawater erosion, biological contamination, and physical damage, their deployment costs are high, maintenance is difficult, and their lifespan is limited. Therefore, the deployment density of sensor nodes in actual networks is much lower than that of terrestrial wireless sensor networks. This results in a severe data sparsity problem in the first ocean dataset in the spatial dimension (especially in the vertical depth direction), which is a major obstacle to subsequent high-precision prediction of the marine environmental state.

[0025] It is easy to understand that in this embodiment, a limited number of actual observation samples with a limited spatial distribution can be collected at their respective preset sampling points using an underwater sensor array, serving as the first ocean dataset. Although the obtained samples are relatively sparse (low resolution), they contain real ocean environmental information and can provide a real, reliable, but low-resolution data input basis for subsequent models.

[0026] Step S20: Input the first marine dataset into a preset marine environment prediction model to determine the marine environment state prediction result. The preset marine environment prediction model includes a generation module and a prediction module. The generation module is used to fit each of the marine environment features based on the spatiotemporal coordinates to generate a second marine dataset with higher resolution in the spatial dimension. The prediction module is used to extract the spatiotemporal correlation of the second marine dataset and the intrinsic correlation between each of the marine environment features, and output the corresponding marine environment state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

[0027] It should be noted that, in this embodiment, the preset marine environment prediction model refers to a pre-trained deep learning model, including a cascaded generation module and a prediction module. The generation module is a sub-neural network that takes spatiotemporal coordinates as input and outputs environmental feature values ​​(marine environmental features) under the corresponding spatiotemporal coordinates. After training, it can fit the corresponding environmental feature values ​​according to any given spatiotemporal coordinates, thereby generating a high-resolution second marine dataset. The second marine dataset refers to a dataset with higher depth resolution obtained by dense sampling interpolation in the spatial dimension through the generation module based on the first marine dataset. The prediction module is another sub-neural network that takes high-resolution multivariate data as input and outputs the prediction results of the environmental state at future times. The neural network, after training, can extract the spatiotemporal correlations and intrinsic correlations between feature values ​​of various stages in a given high-resolution second ocean dataset, and generate high-precision prediction results of ocean environmental status based on these correlations. Spatiotemporal correlations include temporal correlations and spatial correlations. Temporal correlations refer to the patterns of change of observation data at the same location (preset sampling point) over time, while spatial correlations refer to the patterns of mutual influence between observation data at different horizontal and depth locations. Intrinsic correlations refer to the physical coupling relationships between different types of observation data, such as temperature and salinity being coupled through the seawater density equation, jointly affecting the sound speed profile, etc. The prediction results of ocean environmental status refer to the predicted environmental feature values ​​at various horizontal locations and depths at future times.

[0028] It should be understood that the preset marine environment prediction model provided in this embodiment is a model constructed based on the problem of marine data sparsity and the multi-dimensional data correlation in time and space. It can be decomposed into two stages for solution: the generation stage and the prediction stage. In the generation stage, the goal is to construct a high-resolution marine dataset for each marine environmental feature based on the original observation data. In the prediction stage, the goal is to use the generated high-resolution marine dataset to predict future environmental change trends.

[0029] It is worth noting that, in this embodiment, due to the instability of the underwater sensor array, the actual collected observation data often has missing or anomalies in certain dimensions. Correspondingly, the data in the actual observation data (the first ocean dataset) may be incomplete. Therefore, in this embodiment, when predicting the changing trends of marine environmental characteristics, the goal is to minimize the prediction error, as shown in the following expression: ; in, This represents the actual data of the marine environmental characteristics at a future time t, while This represents the predicted result of the marine environmental characteristics at a future time t.

[0030] In this embodiment, the prediction task can be easily decomposed into two sub-tasks: "generating high-resolution data" and "capturing multidimensional correlations." Using models constructed by the generation and prediction modules respectively for these sub-tasks, a sparse first ocean dataset is first fed into the generation module. This module infers unsampled ocean environmental features based on the sparse observation data, resulting in a second ocean dataset with higher resolution. Subsequently, the prediction module analyzes the temporal variation (temporal correlation), spatial variation (spatial correlation), and physical coupling (intrinsic correlation) patterns between different types of ocean environmental features in the second ocean dataset. Finally, it outputs a high-precision prediction of the ocean environmental state at future times.

[0031] This application provides a method for predicting the state of the marine environment. A first marine dataset is acquired using an underwater sensor array, including the spatiotemporal coordinates of preset sampling points and corresponding marine environmental features. This first marine dataset is then input into a preset marine environment prediction model to determine the predicted marine environment state. Specifically, the model's generation module fits the marine environmental features based on the spatiotemporal coordinates to generate a second marine dataset with higher spatial resolution. The model's prediction module extracts the spatiotemporal correlations of the second marine dataset and the intrinsic correlations between various marine environmental features, outputting the corresponding prediction results. By first addressing the data sparsity problem through the generation module and then capturing the correlations between spatiotemporal and multi-environmental features through the prediction module, the accuracy and stability of marine environment state prediction are significantly improved.

[0032] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 as well as Figure 7 The steps of the marine environmental state prediction method further include: Step S201: Based on historical spatiotemporal coordinates and corresponding historical marine environmental characteristics, construct spatiotemporal feature submatrices and environmental variable submatrices respectively; It should be understood that, in this embodiment, please refer to Figure 2 and Figure 3 , Figure 2 This describes a flowchart of the module building process provided in this embodiment. Figure 3 This is a schematic diagram illustrating the main structure of the generation module provided in this embodiment.

[0033] It should be noted that, in this embodiment, historical spatiotemporal coordinates refer to the coordinate data corresponding to data points collected over a period of time that have been labeled with time and location, such as longitude, latitude, depth, and time; historical marine environmental features refer to the measured environmental parameter values ​​corresponding to historical spatiotemporal coordinates, such as temperature and salinity; the spatiotemporal feature submatrix refers to organizing these historical spatiotemporal coordinates into a matrix form, with each row representing the multidimensional coordinates of a sampling point; the environmental variable submatrix refers to organizing the corresponding environmental feature values ​​into a matrix form, with each row representing multiple environmental feature values ​​corresponding to a preset sampling point.

[0034] It is easy to understand that, in this embodiment, the spatiotemporal feature submatrix and the environmental variable submatrix are used to provide standardized input-output pairs for the subsequent training generation module. The spatiotemporal feature submatrix serves as the input to the generation module, while the environmental feature variable submatrix can serve as the prediction target. Together, they constitute the training dataset.

[0035] In practice, a historical observation matrix containing all data can be constructed using historical spatiotemporal coordinates and historical marine environmental characteristics. This historical observation matrix can then be further divided into a spatiotemporal feature submatrix and an environmental variable submatrix, as shown in the following expression: ; in, For historical observation matrix, For spatiotemporal eigenmatrices, This is a submatrix of environment variables.

[0036] It is worth noting that in this embodiment, in middle, This represents the historical spatiotemporal coordinates of a historical observation, containing spatiotemporal information, and uniquely identifies a set of feature vectors corresponding to a set of marine environmental characteristics. Among them, time information in spatiotemporal information This is used to indicate time parameters such as the month of collection contained in historical observation information, while spatial information is included in spatiotemporal information. It can be represented as: ; in, Indicates longitude. Indicates latitude, This represents the depth at the i-th sampling point.

[0037] If we take Represents the historical observation matrix The amount of historical observation information in the spatiotemporal feature submatrix Includes Spatiotemporal coordinates of historical observation information, environmental variable submatrix Includes The marine environmental characteristics of historical observation information. At this point, if we consider... This indicates the number of spatiotemporal features in historical observation information. Representing the eigenvector The number of marine environmental features in the matrix, then the spatiotemporal feature submatrix Depend on Composed of elements, the environment variable submatrix Depend on Each feature vector consists of 1 element. It can be represented as: ; in, Let i be the i-th feature in the feature vector.

[0038] Step S202: Input the spatiotemporal feature submatrix into a preset multilayer perceptron neural network, and use the environmental variable submatrix as the prediction target to train the preset multilayer perceptron neural network to learn the nonlinear mapping relationship between the historical spatiotemporal coordinates and each of the historical marine environmental features. Step S203: Construct the generation module based on the trained preset multilayer perceptron neural network.

[0039] It should be noted that in this embodiment, the preset multilayer perceptron (MLP) neural network refers to a feedforward neural network, which is the main structure of the generation module, such as... Figure 3 As shown, it can be composed of an input layer, multiple hidden layers and an output layer. Each hidden layer contains several neurons. The neurons between adjacent layers are fully connected by weights and biases, and nonlinear activation functions (such as ReLU activation function) are used to introduce nonlinearity. Nonlinear mapping relationship refers to the complex functional relationship between spatiotemporal coordinates (longitude, latitude, depth, time) and marine environmental characteristics (temperature, salinity, etc.). This relationship is not a simple linear interpolation, but a nonlinear surface determined by marine physical processes.

[0040] It should be understood that traditional linear interpolation methods cannot fit common nonlinear vertical structures in the ocean (such as thermoclines and salinity strata), while MLP can automatically learn such nonlinear mapping relationships from historical data by virtue of its powerful nonlinear identification capabilities, and construct a mapping relationship between the sparse first ocean dataset and the high-resolution second ocean dataset, as expressed below: ; in, This represents a high-resolution second ocean dataset; This represents the sparse original observation dataset, i.e., the first ocean dataset; Represents the nonlinear mapping function of the MLP; This represents the model parameters of the MLP.

[0041] It is easy to understand that, in this embodiment, the spatiotemporal feature submatrix constructed based on the first ocean dataset can be used as the core input, that is, replacing the above. And use the environmental variable submatrix constructed based on the first ocean dataset as the prediction target, that is, as mentioned above. The prediction target is to train the above MLP, and the training objective is to make the generated predictions more accurate. The error between the MLP and the actual environmental variable submatrix is ​​minimized. This allows the MLP to learn the implicit nonlinear relationship between spatiotemporal coordinates and environmental features, i.e., the nonlinear mapping relationship. After sufficient training, the MLP can output high-precision estimates of environmental features based on arbitrarily given, even unsampled, spatiotemporal coordinates, thereby constructing a high-resolution ocean dataset. At this point, the final required generative module can be built based on the trained MLP.

[0042] In practical implementation, MLP uses spatiotemporal feature submatrices. The spatiotemporal information in the matrix is ​​used as input, with the environmental variable submatrix as the input. The marine environmental characteristics corresponding to the mid-spatial-temporal coordinates are used as the prediction target, and are represented by sub-matrices for each environmental variable. Each marine environmental feature Learning nonlinear transformation functions , to represent its origin from the input (spatiotemporal feature submatrix) The spatiotemporal dependency from output to output is shown in the following formula: ; in, This represents a set of MLP parameters used to fit a nonlinear function. MLPs with different neural network hyperparameters (such as the number of hidden layers, the number of hidden neurons, activation functions, etc.) have different representational capabilities.

[0043] It is worth noting that the above It means In deep learning, a real-dimensional vector space corresponds to the input of the model, i.e., a vector space with... Samples with each feature. And the above... It represents a one-dimensional real space, corresponding to the output of the model, which is a specific scalar value. It is commonly used in regression tasks (such as predicting a continuous value) and binary classification tasks (the output is processed by Sigmoid to output a probability value).

[0044] Subsequently, This represents an environment variable submatrix based on MLP. Each feature The set of all nonlinear transformation functions learned. Represented as features Among all the constructed nonlinear transformation functions, the MLP model that best approximates the nonlinear relationship between historical spatiotemporal coordinates and historical marine environmental characteristics can be obtained through cross-validation. At this point, the MLP's network structure is similar to... All of these are already fixed. Finally, it can be based on... A generation module is constructed to facilitate the subsequent generation of a high-resolution second ocean dataset.

[0045] Furthermore, in this embodiment, the step of training the preset multilayer perceptron neural network to learn the nonlinear mapping relationship between the historical spatiotemporal coordinates and each of the historical marine environmental features further includes: Step S2021: In the loss function of the preset multilayer perceptron neural network, the sum of squares of all weight parameters of the network is added as a penalty term to constrain the value of each weight parameter.

[0046] It should be noted that, in this embodiment, the loss function of the preset multilayer perceptron neural network is usually composed of a prediction error term and a regularization term. The prediction error term can be the mean squared error, and the regularization term can be L2 regularization, which is the sum of squares of all weight parameters. The weight parameters refer to each element in the weight matrix connecting each layer in the MLP. The penalty term is the L2 regularization term.

[0047] It's easy to understand that with limited training data, MLPs can become overly complex, and perfectly fitting noise in the training data can lead to decreased generalization ability. To address this issue, L2 regularization can be added to prevent overfitting. In this embodiment, during training, L2 regularization can be used as a penalty term to force large weights to be as small as possible, thereby limiting model complexity and improving its predictive performance on unseen data. Ultimately, this ensures that the generation module will not experience abnormal fluctuations due to noise in the training data when generating high-resolution data, guaranteeing the physical rationality and reliability of the generated high-resolution dataset.

[0048] Furthermore, in this embodiment, the steps of the marine environmental state prediction method further include: It should be understood that, in this embodiment, please refer to Figure 4 and Figure 5 , Figure 4 This describes a flowchart of the process for constructing the prediction module provided in this embodiment. Figure 5 This describes a schematic diagram of the prediction module provided in this embodiment.

[0049] Step S204: Determine the historical second ocean dataset by using historical spatiotemporal coordinates and corresponding historical ocean environmental characteristics; It should be noted that, in this embodiment, the historical second ocean dataset refers to the dataset obtained by using a pre-trained generation module to generate spatiotemporal coordinates within a historical period at high resolution.

[0050] It is easy to understand that in order to train the prediction module, a sufficiently long time series of high-resolution data is required. In this embodiment, the historical raw data (historical spatiotemporal coordinates and corresponding historical marine environmental features) can be enhanced by the trained generation module to obtain a high-resolution historical second ocean dataset, which is then used as the training input for the prediction module.

[0051] Step S205: Based on the historical second ocean dataset, construct a multivariate time series matrix. The multivariate time series matrix contains multiple time steps in the time dimension, and each time step contains the values ​​of each of the ocean environmental features at different depths. It should be noted that, in this embodiment, the multivariate time series matrix refers to a multidimensional tensor, specifically a three-dimensional tensor, whose dimensions include the number of time steps, the number of spatial features, and the number of variables; a time step refers to the observation moments arranged in chronological order, such as consecutive months.

[0052] It is easy to understand that, in order to predict the state of the marine environment, it is necessary to first organize the high-resolution marine dataset into an input format suitable for the subsequent pre-defined multivariate convolutional long short-term memory neural network. In this embodiment, the environmental feature values ​​of all horizontal positions and depths in the historical second marine dataset can be flattened or preserved in matrix form to form frames corresponding to the time steps. Then, multiple consecutive frames are stacked to form the input sequence that the subsequent neural network can process, which is the multivariate time series matrix, and its expression is as follows: ; In deep learning, the above expression can represent a space of a three-dimensional real tensor, whose shape is the first dimension (dimension 0). The second dimension (dimension 1) And the third dimension (dimension 2) Generally speaking, in deep learning, It can represent the number of samples input into the network for processing in one operation; It can represent the number of nodes in a neural network; This can represent the query feature dimension. Here, Specifically, regarding the length of the time series, It can be represented by the polymorphic time series matrix at the t-th time series, by It consists of marine environmental characteristics at different depths, and its specific expression is as follows: ; It is worth noting that, in this embodiment, the time step length used for time series prediction in the variable time series matrix may differ for different prediction tasks. Furthermore, The number of samples and the amount of marine environmental characteristics are also related to application requirements.

[0053] Step S206: Input the multivariate time series matrix into a preset multivariate convolutional long short-term memory neural network, and use the historical second ocean environmental features at the corresponding time as the prediction target to train the preset multivariate convolutional long short-term memory neural network to learn the spatiotemporal correlation in the historical second ocean dataset and the intrinsic correlation between each historical ocean environmental feature. Step S207: Construct the prediction module based on the trained preset multivariate convolutional long short-term memory neural network.

[0054] It should be understood that correlation extraction and multivariate time series prediction are the core of the prediction module. The purpose of these two parts is to capture the multivariate time series matrix. Relationships between various data Including spatial correlation Time correlation and intrinsic connections To achieve accurate prediction of marine environmental conditions.

[0055] It should be noted that, in this embodiment, the preset multivariate convolutional long short-term memory neural network (MVC-LSTM) refers to a deep learning network specifically designed for the aforementioned correlation extraction and multivariate spatiotemporal sequence prediction, and its structure is as follows: Figure 5 As shown, it includes an input layer, several long short-term memory convolutional layers ConvLSTM (ConvLSTM1~ConvLSTMm), and a 3D (three-dimensional) convolutional layer Conv3D. Multivariate temporal observations are stacked into a fixed-dimensional representation, and ConvLSTM and Conv3D are coupled into the same framework to extract the spatiotemporal correlations between various marine environmental features and the intrinsic correlations between different types of features.

[0056] Specifically, in MVC-LSTM, the ConvLSTM layer is used to capture the spatiotemporal correlations between input data and the intrinsic correlations between multiple variables, while the Conv3D layer is used to handle the interactions between spatial dimensions. The number and size of convolutional kernels in the ConvLSTM and Conv3D layers are determined based on the specific application scenario and the characteristics of the input data to improve predictive ability.

[0057] It is worth noting that in this embodiment, ConvLSTM replaces the fully connected structure from the input layer to the hidden layer and from the hidden layer to the hidden layer in the LSTM network with convolution operations, so that it can extract spatial features while having the ability of traditional LSTM networks to extract temporal features.

[0058] It should be noted that, in this embodiment, the ConvLSTM proposed in this embodiment can be established using the following formula: ; ; ; ; ; in, , and These represent the three key components of an LSTM unit (i.e., the input gate, the forget gate, and the output gate), and * represents the convolution operator. This represents the Hadamard product. Represents the ogistic sigma-oid activation function; , and All of them are generated by the sigma-oid activation function to a value located at (0, 1). , and These represent the input information, hidden state, and memory cell of the LSTM memory cell, respectively. The weight matrix is ​​used to mix various types of information into the current state. This represents the bias term when mixing various types of information. The above equations enable ConvLSTM to effectively construct long-term dependencies in time series data.

[0059] It should be noted that, in this embodiment, This is used to control the proportion of newly input information incorporated into the current state, and can be controlled by... The value is set so that only important information can enter the current state; Used to determine the proportion of historical information that can remain in the current state; These three units are used to control the output of the LSTM memory units, determining the proportion of the current output information that can be passed to the next layer. They effectively maintain long-term information correlation while preventing gradient explosion or vanishing gradients, thus stabilizing network performance.

[0060] In this embodiment, it is easy to understand that a high-resolution multivariate time series matrix from multiple past time steps can be used as the input sequence, and the data at the corresponding time step (i.e., the next time step) (historical second marine environmental features) can be used as the prediction target. A preset multivariate convolutional long short-term memory neural network is trained on this network. By stacking multiple high-resolution sequences, the temporal variation characteristics of each spatial point can be captured, and key information can be extracted as input data for training the model. This allows the model to learn the spatiotemporal correlations within the marine dataset and the intrinsic correlations between various marine environmental features. At this point, the required prediction module can be constructed based on the trained MVC-LSTM.

[0061] Specifically, in the input layer, high-resolution features at a specified location in the generated high-resolution dataset at time t can be stacked to form a dataset containing... The variable time series matrix of each element .in, This represents the value of the i-th variable at the j-th depth. This is achieved by using a multivariate time series matrix. The element corresponding to the i-th variable (such as different marine environmental characteristics like temperature and salinity) at the j-th depth. Comparing it to a pixel in a single-channel image, it can represent a variable time series matrix. Consider a pixel with a width of 1 and a height of 1. The number of channels is The tensor is a three-dimensional tensor. The size and number of variables used in the prediction together determine the tensor.

[0062] In ConvLSTM layers, convolutional kernels of different sizes can be used to capture the interactions and spatiotemporal dependencies between variables at different scales (time and space). Larger kernels can learn trends over larger regions and longer time periods. Larger kernels also mean that more information about the interactions between variables can be learned each time. Smaller kernels primarily capture proximity in the spatial and temporal dimensions. In different applications, network depth, the number of kernels, and kernel size all need to be customized, and different tasks may require different settings to achieve optimal prediction accuracy. After multiple convolution operations, the size of each ConvLSTM input gradually decreases. Furthermore, since convolution operations end when they reach the edges, edge data in each tensor has less impact on the output than data located at the center of the tensor. In convolution operations, data at the center participates in multiple operations, but edge data may only participate in one, leading to a loss of edge information.

[0063] In the Conv3D layer, the multivariate spatiotemporal features learned by the ConvLSTM layer can be used as input, and more global spatiotemporal relationships between different types of features can be extracted. In addition, the 3D convolution transforms the number of output channels and maps the prediction results to an output space with the same shape as the input.

[0064] In the output layer, the prediction results of Conv3D can be returned, with the same shape as the input tensor.

[0065] The MVC-LSTM constructed using the above structure can preserve all spatial information and interactions between multivariate observations throughout the prediction process. By stacking multiple ConvLSTM layers and Conv3D layers, the entire structure possesses a powerful ability to represent the spatiotemporal dependencies and interactions of various features. Therefore, the MVC-LSTM constructed in this embodiment can exhibit good performance in complex spatiotemporal data prediction tasks.

[0066] It is worth noting that, in this embodiment, to ensure that the output of the i-th layer, i.e., the input of the (i+1)-th layer, always has the same size as the input layer tensor and to preserve edge information, this embodiment can also employ padding, allowing the convolution kernel to pad each region outside the boundary with a zero. This ensures that the output size is the same as the input tensor while simultaneously avoiding the loss of edge information.

[0067] Furthermore, in this embodiment, the step of training the preset multivariate convolutional long short-term memory neural network to learn the spatiotemporal correlations in the historical second ocean dataset and the intrinsic correlations between the various historical ocean environmental features further includes: Step S2061: Using a random deactivation technique, neurons in the preset multivariate convolutional long short-term memory neural network are randomly deactivated with a preset probability.

[0068] It should be noted that, in this embodiment, random dropout is a regularization method commonly used in deep learning training. In each training iteration, a portion of neurons in the network are temporarily "discarded" according to a preset probability, that is, the output of these neurons is set to zero and they do not participate in forward and backward propagation. The preset probability is the Dropout rate, which is usually set between 0.2 and 0.5.

[0069] It is easy to understand that MVC-LSTM networks have a large number of parameters, which may lead to overfitting with limited training data, meaning they perform well on the training set but poorly on the test set. In this embodiment, by introducing Dropout, the network is forced to learn redundant and robust feature representations, thereby reducing the risk of overfitting.

[0070] Further, in this embodiment, the spatiotemporal coordinates include time coordinates, horizontal position coordinates, and depth position coordinates; the generation module has a preset multilayer perceptron neural network; the prediction module has a preset multivariate convolutional long short-term memory neural network; and the step of inputting the first marine dataset into the preset marine environment prediction model to determine the marine environment state prediction result specifically includes: Step S211: The generation module determines each depth to be predicted according to preset application requirements; It should be understood that, in this embodiment, please refer to Figure 6 and Figure 7 , Figure 6 This describes a flowchart of the process for predicting marine environmental processes provided in this embodiment. Figure 7 This describes a basic framework diagram of the process for predicting marine environmental conditions provided in this embodiment.

[0071] It should be noted that, in this embodiment, the preset application requirements refer to the prediction accuracy or resolution requirements that the user hopes to achieve; the depth to be predicted refers to each depth that needs to generate environmental feature values ​​according to the preset application requirements, and each depth is usually not in the sampling depth of the original first ocean dataset.

[0072] It is easy to understand that, in this embodiment, all the depths to be predicted that need to be predicted and interpolated can be listed first according to the vertical resolution requirements in the preset application requirements.

[0073] Step S212: Extract the horizontal position coordinates and the time coordinates from the first ocean dataset using the generation module; Step S213: Through the generation module, the horizontal position coordinates and the time coordinates are combined with each of the depths to be predicted to generate corresponding new spatiotemporal coordinates. It should be noted that in this embodiment, the horizontal position coordinates refer to the longitude and latitude of each preset sampling point in the first ocean dataset; the time coordinates refer to the sampling time corresponding to each preset sampling point; the new spatiotemporal coordinates refer to the four-dimensional coordinate vector (longitude, latitude, depth, time) obtained by combining each set (longitude, latitude, time) with each depth to be predicted, and each new spatiotemporal coordinate corresponds to a specific spatiotemporal point of the marine environmental feature to be predicted.

[0074] It is easy to understand that in this embodiment, the generation module can first extract all combinations of actually observed horizontal position coordinates and time coordinates from the original first ocean dataset, and pair these horizontal position coordinates and time coordinates as invariant coordinates with the predicted depth. By pairing each known horizontal position coordinate and time coordinate with each depth to be predicted, a complete set of four-dimensional coordinates can be generated, namely the aforementioned new spatiotemporal coordinates.

[0075] Step S214: The new spatiotemporal coordinates are converted into prediction environment features corresponding to each of the depths to be predicted by the preset multilayer perceptron neural network of the generation module. Step S215: The generation module merges the first ocean dataset with each of the predicted environmental features, and retains the original depth location coordinates and corresponding ocean environmental features of the first ocean dataset to form a second ocean dataset with higher resolution in the vertical depth direction.

[0076] It should be noted that, in this embodiment, the predicted environmental features refer to the environmental feature values ​​(such as temperature and salinity) output by the MLP for each new spatiotemporal coordinate. The feature values ​​are inferred by the model based on the learned nonlinear mapping relationship. The depth location coordinates refer to the actual depth of each preset sampling point observed in the original first ocean dataset.

[0077] As is readily understood, in this embodiment, the core function of the generation module is to complete the originally missing depth data. On one hand, it can use a trained MLP to fit and calculate the corresponding environmental feature values ​​for each depth to be predicted. On the other hand, it can also merge the fitted and calculated environmental feature values ​​with the original actual observation values. In this way, the originally sparse first ocean dataset can be transformed into a second ocean dataset with higher resolution.

[0078] Step S216: Extract the spatiotemporal correlation and the intrinsic correlation from the second ocean dataset using the preset multivariate convolutional long short-term memory neural network of the prediction module. Step S217: The prediction module outputs the marine environmental state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

[0079] In this embodiment, the core function of the prediction module is to discover patterns and use them for prediction. First, it converts the high-resolution second ocean dataset output by the generation module into a multivariate temporal matrix suitable for MVC-LSTM. Then, the ConvLSTM layer in MVC-LSTM scans the space (different depths and locations) through convolutional operations while simultaneously tracking temporal changes through LSTM. The Conv3D layer in MVC-LSTM further fuses time, space, and multivariates into a three-dimensional cube for analysis. Ultimately, MVC-LSTM can extract three patterns: the trend of ocean environmental features changing over time (i.e., temporal correlation), the trend of ocean environmental features changing with location (i.e., spatial correlation), and the coupling relationship between different types of ocean environmental features (i.e., intrinsic correlation). These patterns are encoded in the network parameters. Subsequently, based on all the extracted patterns, high-dimensional features can be mapped to the prediction target space to form a corresponding high-resolution prediction matrix, thus directly outputting the environmental feature values ​​for future times, which is the aforementioned ocean environmental state prediction result.

[0080] This application also provides a marine environmental state prediction device. Please refer to... Figure 8 The marine environment state prediction device includes: The raw data acquisition module 10 is used to acquire a first ocean dataset, which includes the spatiotemporal coordinates of preset sampling points and the corresponding marine environmental features. The marine environment prediction module 20 is used to input the first marine dataset into a preset marine environment prediction model to determine the marine environment state prediction result. The preset marine environment prediction model includes a generation module and a prediction module. The generation module is used to fit each of the marine environment features based on the spatiotemporal coordinates to generate a second marine dataset with higher resolution in the spatial dimension. The prediction module is used to extract the spatiotemporal correlation of the second marine dataset and the intrinsic correlation between each of the marine environment features, and output the corresponding marine environment state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

[0081] The marine environmental state prediction device provided in this application embodiment, employing the marine environmental state prediction method described in the above embodiments, can solve the technical problem of how to improve the accuracy and stability of marine environmental state prediction. Compared with the prior art, the beneficial effects of the marine environmental state prediction device provided in this application embodiment are the same as those of the marine environmental state prediction method provided in the above embodiments, and other technical features in the marine environmental state prediction device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0082] This application also provides a marine environmental state prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the marine environmental state prediction method in the above embodiment 1.

[0083] The following is for reference. Figure 9 The diagram illustrates a structural schematic suitable for implementing the marine environmental state prediction device of the embodiments of this application. The marine environmental state prediction device in the embodiments of this application may include, but is not limited to, fixed terminals such as vehicle-mounted terminals. Figure 9 The marine environmental state prediction device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0084] like Figure 9 As shown, the marine environmental state prediction device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the marine environmental state prediction device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following devices may be connected to the I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the marine environmental state prediction device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a marine environmental state prediction device with various devices, it should be understood that implementation or possession of all shown devices is not required. More or fewer devices may be implemented alternatively.

[0085] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0086] The marine environmental state prediction device provided in this application, employing the marine environmental state prediction method described in the above embodiments, can solve the technical problem of how to improve the accuracy and stability of marine environmental state prediction. Compared with the prior art, the beneficial effects of the marine environmental state prediction device provided in this application are the same as those of the marine environmental state prediction method provided in the above embodiments, and other technical features of this marine environmental state prediction device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0087] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0089] This application also provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the marine environmental state prediction method in the above embodiments.

[0090] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, apparatuses, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution apparatus, device, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0091] The aforementioned computer-readable storage medium may be included in the marine environmental state prediction device; or it may exist independently and not be assembled into the marine environmental state prediction device.

[0092] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the marine environment state prediction device, the marine environment state prediction device: acquires a first marine dataset through an underwater sensor array, the first marine dataset including the spatiotemporal coordinates of preset sampling points and corresponding marine environmental features; inputs the first marine dataset into a preset marine environment prediction model to determine the marine environment state prediction result, the preset marine environment prediction model including a generation module and a prediction module, the generation module being used to fit each of the marine environmental features based on the spatiotemporal coordinates to generate a second marine dataset with higher resolution in the spatial dimension, and the prediction module being used to extract the spatiotemporal correlation of the second marine dataset and the intrinsic correlation between each of the marine environmental features and output the corresponding marine environment state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

[0093] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0094] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using dedicated hardware-based apparatus to perform the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0095] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0096] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described marine environmental state prediction method, thereby solving the technical problem of how to improve the accuracy and stability of marine environmental state prediction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the marine environmental state prediction method provided in the above embodiments, and will not be repeated here.

[0097] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the marine environmental state prediction method described above.

[0098] The computer program product provided in this application can solve the technical problem of how to improve the accuracy and stability of marine environmental state prediction. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the marine environmental state prediction method provided in the above embodiments, and will not be repeated here.

[0099] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A method for predicting the state of the marine environment, characterized in that, The steps of the marine environment state prediction method include: The first ocean dataset is acquired by an underwater sensor array. The first ocean dataset includes the spatiotemporal coordinates of preset sampling points and the corresponding marine environmental features. The first marine dataset is input into a preset marine environment prediction model to determine the marine environment state prediction result. The preset marine environment prediction model includes a generation module and a prediction module. The generation module is used to fit each of the marine environment features based on the spatiotemporal coordinates to generate a second marine dataset with higher resolution in the spatial dimension. The prediction module is used to extract the spatiotemporal correlation of the second marine dataset and the intrinsic correlation between each of the marine environment features, and output the corresponding marine environment state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

2. The marine environmental state prediction method as described in claim 1, characterized in that, The steps of the marine environment state prediction method also include: Based on historical spatiotemporal coordinates and corresponding historical marine environmental characteristics, spatiotemporal feature submatrices and environmental variable submatrices are constructed respectively. The spatiotemporal feature submatrix is ​​input into a preset multilayer perceptron neural network, and the environmental variable submatrix is ​​used as the prediction target to train the preset multilayer perceptron neural network to learn the nonlinear mapping relationship between the historical spatiotemporal coordinates and each of the historical marine environmental features. The generation module is constructed based on a pre-trained multilayer perceptron neural network.

3. The marine environmental state prediction method as described in claim 2, characterized in that, The step of training the preset multilayer perceptron neural network to learn the nonlinear mapping relationship between the historical spatiotemporal coordinates and each of the historical marine environmental features further includes: In the loss function of the preset multilayer perceptron neural network, the sum of squares of all weight parameters of the network is added as a penalty term to constrain the value of each weight parameter.

4. The marine environmental state prediction method as described in claim 1, characterized in that, The steps of the marine environment state prediction method also include: The historical second ocean dataset was determined by using historical spatiotemporal coordinates and corresponding historical marine environmental characteristics. Based on the aforementioned historical second ocean dataset, a multivariate time series matrix is ​​constructed. The multivariate time series matrix contains multiple time steps in the time dimension, and each time step contains the values ​​of the aforementioned ocean environmental features at different depths. The multivariate time series matrix is ​​input into a preset multivariate convolutional long short-term memory neural network, and the historical second ocean environmental features at the corresponding time are used as prediction targets to train the preset multivariate convolutional long short-term memory neural network to learn the spatiotemporal correlation in the historical second ocean dataset and the intrinsic correlation between the historical ocean environmental features. The prediction module is constructed based on a pre-trained multivariate convolutional long short-term memory neural network.

5. The marine environmental state prediction method as described in claim 4, characterized in that, The step of training the preset multivariate convolutional long short-term memory neural network to learn the spatiotemporal correlations in the historical second ocean dataset and the intrinsic correlations among the various historical ocean environmental features further includes: A random deactivation technique is used to randomly deactivate neurons in the preset multivariate convolutional long short-term memory neural network with a preset probability.

6. The marine environmental state prediction method as described in claim 1, characterized in that, The spatiotemporal coordinates include time coordinates, horizontal position coordinates, and depth position coordinates. The generation module has a preset multilayer perceptron neural network, and the prediction module has a preset multivariate convolutional long short-term memory neural network. The step of inputting the first marine dataset into the preset marine environment prediction model to determine the marine environment state prediction result specifically includes: The generation module determines the various depths to be predicted based on preset application requirements. The generation module extracts the horizontal position coordinates and the time coordinates from the first ocean dataset. The generation module combines the horizontal position coordinates and the time coordinates with each of the depths to be predicted to generate corresponding new spatiotemporal coordinates. The new spatiotemporal coordinates are converted into prediction environment features corresponding to each prediction depth through the preset multilayer perceptron neural network of the generation module. The generation module merges the first ocean dataset with each of the predicted environmental features, and retains the original depth coordinates and corresponding ocean environmental features of the first ocean dataset to form a second ocean dataset with higher resolution in the vertical depth direction. The spatiotemporal correlation and the intrinsic correlation are extracted from the second ocean dataset through the preset multivariate convolutional long short-term memory neural network of the prediction module. The prediction module outputs the marine environmental state prediction results based on the spatiotemporal correlation and the intrinsic correlation.

7. A marine environmental state prediction device, characterized in that, The marine environment state prediction device includes: The raw data acquisition module is used to acquire a first ocean dataset, which includes the spatiotemporal coordinates of preset sampling points and the corresponding marine environmental features. The marine environment prediction module is used to input the first marine dataset into a preset marine environment prediction model to determine the marine environment state prediction result. The preset marine environment prediction model includes a generation module and a prediction module. The generation module is used to fit each of the marine environment features based on the spatiotemporal coordinates to generate a second marine dataset with higher resolution in the spatial dimension. The prediction module is used to extract the spatiotemporal correlation of the second marine dataset and the intrinsic correlation between each of the marine environment features, and output the corresponding marine environment state prediction result based on the spatiotemporal correlation and the intrinsic correlation.

8. A marine environmental state prediction device, characterized in that, The marine environment state prediction device includes: a memory, a processor, and a marine environment state prediction program stored in the memory and executable on the processor, the marine environment state prediction program being configured to implement the steps of the marine environment state prediction method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and the computer-readable storage medium stores a marine environment state prediction program. When the marine environment state prediction program is executed by a processor, it implements the steps of the marine environment state prediction method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the marine environmental state prediction method as described in any one of claims 1 to 6.

Images