Residual ring current monitoring method, system, device and storage medium

Abstract

This invention discloses a method, system, device, and storage medium for residual circulation monitoring, comprising: acquiring basic marine area data and multi-source environmental data of a target sea area; establishing an FVCOM model based on the basic marine area data; using the multi-source environmental data as boundary conditions, performing marine numerical simulations of the target sea area through the FVCOM model to obtain marine dynamic simulation data of the target sea area; inputting the marine dynamic simulation data into a pre-trained residual circulation prediction model of the target sea area to obtain residual circulation distribution prediction data of the target sea area; updating the boundary conditions based on the residual circulation distribution prediction data, and returning to perform marine numerical simulations of the target sea area through the FVCOM model until a preset number of simulations is reached to obtain the residual circulation distribution prediction of the target sea area. This invention improves the accuracy and efficiency of residual circulation distribution prediction, and also improves the accuracy and efficiency of wind field distribution prediction, and can be widely applied in the field of marine engineering technology.

Description

Technical Field

[0001] This invention relates to the field of marine engineering technology, and in particular to a residual circulation monitoring method, system, device, and storage medium. Background Technology

[0002] As one of the main driving forces of ocean circulation, wind fields have a particularly significant impact on residual circulation. By altering the velocity and direction of flow at the ocean surface, wind fields influence the distribution of ocean temperature and salinity, thereby driving the generation and changes in residual circulation. Therefore, in-depth research into the response mechanism of residual circulation to wind fields is of significant theoretical and practical importance for understanding the evolution of ocean circulation, predicting trends in marine environmental changes, and formulating scientifically sound ocean management strategies.

[0003] Domestic scholars have made some progress in studying the impact of wind fields on ocean circulation. For example, some scholars have used numerical simulations to study the impact of different types of wind fields, such as monsoons and typhoons, on the South China Sea circulation, finding that wind fields can significantly alter the circulation structure of the South China Sea; the response mechanism of residual circulation to wind fields has also been studied. For example, some scholars have used observational data and numerical simulations to study the response mechanism of the Yellow Sea residual circulation to wind fields, finding that the residual circulation changes under the influence of wind fields, and that these changes also affect the wind fields in surrounding sea areas. International scholars have also made some progress in studying the impact of wind fields on ocean circulation. For example, some scholars have analyzed the response of the North Atlantic circulation to wind fields using long-term observational data, finding that wind fields can change the intensity and direction of the North Atlantic circulation; the response mechanism of residual circulation to wind fields has also been studied. For example, some scholars have used numerical simulations to study the response mechanism of the Pacific residual circulation to wind fields.

[0004] Currently, some studies have attempted to explore the response mechanism of residual circulation to wind fields through field observation and data analysis. The most typical approach involves using high-frequency ground-wave radar to observe ocean surface current velocity and direction, and combining this with meteorological data to analyze the impact of wind fields on ocean circulation. These approaches typically include the following steps: First, deploying high-frequency ground-wave radar stations in the target sea area to acquire real-time data on ocean surface current velocity and direction; second, collecting concurrent meteorological data, including wind speed and direction; third, extracting key characteristic parameters of residual circulation and wind fields through data processing and analysis; and finally, using statistical analysis and physical models to study the response patterns of residual circulation to changes in wind fields.

[0005] These implementation schemes have revealed the mechanism of residual circulation's response to wind fields to some extent, but they still have some limitations. For example, the observation range of high-frequency ground wave radar is limited and may not be able to cover the entire target sea area; at the same time, the acquisition of meteorological data may also be affected by factors such as weather conditions and the distribution of observation stations. In addition, these schemes mainly focus on the statistical relationship between residual circulation and wind fields, lacking in-depth exploration of the formation and evolution mechanism of residual circulation. Furthermore, existing schemes also suffer from low efficiency and insufficient stability, and may not be able to accurately capture circulation changes caused by complex physical processes. Summary of the Invention

[0006] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0007] Therefore, one objective of this invention is to provide a residual circulation monitoring method that improves the accuracy and efficiency of residual circulation distribution prediction, as well as the accuracy and efficiency of wind field distribution prediction.

[0008] Another objective of this invention is to provide a residual circulation monitoring system.

[0009] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of the present invention include:

[0010] In a first aspect, embodiments of the present invention provide a residual circulation monitoring method, comprising the following steps:

[0011] Acquire basic marine area data and multi-source environmental data for the target sea area, and establish an FVCOM model based on the basic marine area data;

[0012] Using the multi-source environmental data as boundary conditions, the target sea area is subjected to marine numerical simulation through the FVCOM model to obtain marine dynamic simulation data of the target sea area;

[0013] The ocean dynamics simulation data is input into the pre-trained residual circulation prediction model of the target sea area to obtain the residual circulation distribution prediction data of the target sea area.

[0014] The boundary conditions are updated based on the residual circulation distribution prediction data, and the target sea area is simulated using the FVCOM model until the preset number of simulations is reached to obtain the residual circulation distribution prediction of the target sea area.

[0015] Furthermore, in one embodiment of the present invention, the basic marine area data includes seabed topography data, coastline shape data, and seawater level data of the target marine area, and the multi-source environmental data includes meteorological data, ocean current data, seawater temperature data, and seawater salinity data.

[0016] Furthermore, in one embodiment of the present invention, the residual circulation prediction model is trained through the following steps:

[0017] Acquire first ocean dynamic sample data for multiple historical time periods of the target sea area and corresponding residual circulation distribution labels. The residual circulation distribution labels include the distribution of residual circulation velocity, residual circulation direction and residual circulation intensity of the target sea area.

[0018] A labeled dataset was constructed based on the first ocean dynamics sample data and the residual circulation distribution labels;

[0019] Acquire second ocean dynamics sample data for multiple historical periods of the target sea area, and construct an unlabeled dataset based on the second ocean dynamics sample data;

[0020] The pre-built convolutional neural network is semi-supervised learning based on the labeled dataset and the unlabeled dataset to obtain the trained residual circulation prediction model.

[0021] Furthermore, in one embodiment of the present invention, the step of performing semi-supervised learning on the convolutional neural network based on the labeled dataset and the unlabeled dataset to obtain the trained co-circulation prediction model specifically includes:

[0022] The labeled dataset is input into the convolutional neural network to obtain a trained initial model.

[0023] The initial model is used to predict the unlabeled dataset, and the prediction results with high confidence are selected as pseudo-labels based on the confidence threshold.

[0024] The unlabeled data is labeled using the pseudo-labels to obtain a pseudo-label dataset;

[0025] The initial model is fine-tuned based on the pseudo-labeled dataset and the labeled dataset, and the prediction of the unlabeled dataset is performed using the initial model until a preset number of model fine-tunings is reached, resulting in the trained residual circulation prediction model.

[0026] Furthermore, in one embodiment of the present invention, the step of performing semi-supervised learning on the convolutional neural network based on the labeled dataset and the unlabeled dataset to obtain the trained co-circulation prediction model specifically includes:

[0027] The labeled dataset is divided into a first data subset and a second data subset;

[0028] The first data subset and the second data subset are respectively input into the convolutional neural network to obtain the trained first initial model and the second initial model;

[0029] The first initial model and the second initial model are used to predict the unlabeled dataset respectively. Based on the confidence threshold, the prediction result with high confidence is selected as the pseudo label to obtain the first pseudo label and the second pseudo label.

[0030] The unlabeled data is labeled according to the first pseudo-label and the second pseudo-label respectively, to obtain the first pseudo-label dataset and the second pseudo-label dataset;

[0031] The first initial model is fine-tuned based on the second pseudo-label dataset and the first data subset. The second initial model is fine-tuned based on the first pseudo-label dataset and the second data subset. The first initial model and the second initial model are then used to predict the unlabeled dataset. This process is repeated until a preset number of model fine-tuning steps are reached, resulting in a trained first co-circulation prediction model and a second co-circulation prediction model.

[0032] The first residual circulation prediction model and the second residual circulation prediction model are integrated for learning to obtain the trained residual circulation prediction model.

[0033] Furthermore, in one embodiment of the present invention, the step of inputting the labeled dataset into the convolutional neural network to obtain a trained initial model specifically includes:

[0034] The labeled dataset is divided into a training set, a validation set, and a test set;

[0035] The training set is input into the convolutional neural network to obtain the prediction result of the residual circulation distribution;

[0036] The loss value is determined based on the residual circulation distribution prediction results and the residual circulation distribution labels;

[0037] The internal parameters of the convolutional neural network are updated based on the loss value, and the validation set is input into the updated convolutional neural network to obtain the model performance of the convolutional neural network on the validation set.

[0038] The hyperparameters of the convolutional neural network are updated based on the model performance, and the training set is returned to be input into the convolutional neural network until the model performance no longer improves in multiple consecutive iterations, thus obtaining the first initial model.

[0039] The test set is input into the first initial model to obtain the model evaluation index of the first initial model;

[0040] When the model evaluation metric meets the preset requirements, the first initial model is used as the trained initial model.

[0041] Furthermore, in one embodiment of the present invention, the residual circulation monitoring method further includes the following steps:

[0042] Acquire historical residual circulation distribution data and historical wind field observation data for the target area;

[0043] Regression analysis was performed on the historical residual circulation distribution data and the historical wind field observation data to obtain the data relationship between the residual circulation distribution and the wind field parameters.

[0044] The wind field distribution prediction for the target area is determined based on the predicted residual circulation distribution and the data relationship.

[0045] Secondly, embodiments of the present invention provide a residual circulation monitoring system, comprising:

[0046] The data acquisition module is used to acquire basic marine area data and multi-source environmental data of the target sea area, and to establish an FVCOM model based on the basic marine area data;

[0047] The marine numerical simulation module is used to perform marine numerical simulation on the target sea area using the multi-source environmental data as boundary conditions and the FVCOM model to obtain marine dynamic simulation data of the target sea area.

[0048] The residual circulation distribution prediction module is used to input the ocean dynamics simulation data into the pre-trained residual circulation prediction model of the target sea area to obtain the residual circulation distribution prediction data of the target sea area.

[0049] The boundary condition update module is used to update the boundary conditions based on the residual circulation distribution prediction data, and return the ocean numerical simulation of the target sea area using the FVCOM model until a preset number of simulations is reached to obtain the residual circulation distribution prediction of the target sea area.

[0050] Thirdly, embodiments of the present invention provide a residual circulation monitoring device, comprising:

[0051] At least one processor;

[0052] At least one memory for storing at least one program;

[0053] When the at least one program is executed by the at least one processor, the at least one processor implements the above-described residual current monitoring method.

[0054] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the aforementioned residual current monitoring method.

[0055] The advantages and beneficial effects of the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention:

[0056] This invention acquires basic marine area data and multi-source environmental data for a target sea area. An FVCOM model is established based on the basic marine area data, using the multi-source environmental data as boundary conditions. The FVCOM model is then used to perform marine numerical simulations of the target sea area, yielding marine dynamic simulation data. This data is then input into a pre-trained residual circulation prediction model for the target sea area, resulting in residual circulation distribution prediction data. Boundary conditions are updated based on the residual circulation distribution prediction data, and the FVCOM model is used to perform marine numerical simulations of the target sea area until a preset number of simulations is reached, thus obtaining the residual circulation distribution prediction for the target sea area. This invention combines the FVCOM model with a deep neural network. The FVCOM model generates ocean dynamics simulation data, which is then used as input to the residual circulation prediction model. The boundary conditions of the FVCOM model are updated based on the residual circulation distribution prediction data output by the residual circulation prediction model. This data-driven approach improves the model's ability to predict the residual circulation distribution field, thereby enhancing the accuracy and efficiency of residual circulation distribution prediction. Furthermore, the wind field distribution prediction can be determined based on the obtained residual circulation distribution prediction, further improving the accuracy and efficiency of wind field distribution prediction. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments of the present invention are described below. It should be understood that the drawings described below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 A flowchart illustrating the steps of a residual circulation monitoring method provided in an embodiment of the present invention;

[0059] Figure 2 A structural block diagram of a residual circulation monitoring system provided in an embodiment of the present invention;

[0060] Figure 3 This is a structural block diagram of a residual circulation monitoring device provided in an embodiment of the present invention. Detailed Implementation

[0061] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0062] In the description of this invention, "multiple" means two or more. The use of "first" and "second" is for distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or the order of the indicated technical features. Furthermore, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.

[0063] With the deepening development of marine engineering, the study of residual circulation is becoming increasingly important in fields such as marine engineering and wind power generation. The FVCOM model has wide applications in marine science and marine engineering, accurately simulating ocean circulation characteristics, including ocean currents, eddies, and boundary flows. This invention innovatively proposes a novel residual circulation monitoring method by combining the FVCOM model with a deep neural network model, achieving accurate prediction of the spatial residual flow field.

[0064] Reference Figure 1 This invention provides a residual circulation monitoring method, which specifically includes the following steps:

[0065] S101. Obtain basic marine area data and multi-source environmental data for the target sea area, and establish an FVCOM model based on the basic marine area data;

[0066] S102. Using multi-source environmental data as boundary conditions, the target sea area is subjected to marine numerical simulation using the FVCOM model to obtain marine dynamic simulation data of the target sea area.

[0067] S103. Input the ocean dynamics simulation data into the pre-trained residual circulation prediction model of the target sea area to obtain the residual circulation distribution prediction data of the target sea area.

[0068] S104. Update the boundary conditions based on the residual circulation distribution prediction data, and return to perform marine numerical simulation of the target sea area using the FVCOM model until the preset number of simulations is reached to obtain the residual circulation distribution prediction of the target sea area.

[0069] Specifically, FVCOM is an unstructured triangular mesh architecture, finite volume, free surface, three-dimensional primitive equations-based ocean numerical model. The primitive equations of FVCOM mainly include momentum equations, mass continuity equations, and temperature, salinity, and density equations. Physically and mathematically, the equations are closed using the Mellor-Yamada 2.5th order vertical turbulence closure model and the Smagorinsky horizontal turbulence closure model. Vertically, a universal vertical coordinate system is used to fit the irregular bottom topography; horizontally, an unstructured triangular mesh is used to spatially discretize the horizontal computational domain. Numerically, the governing equations are solved discretically using flux finite volume integration over the horizontal triangular control volume. This finite volume integration method combines the free geometric fitting characteristics of the finite element method with the simple discretization structure and computational efficiency of the finite difference method, thus integrating the advantages of both methods. Using the finite volume integration scheme better ensures the conservation of mass, momentum, salinity, temperature, and heat in calculations involving complex geometries such as estuaries and bays. The advantages of the FVCOM model in numerical processing methods and shoreline topography fitting have led to its widespread application in estuarine and coastal areas.

[0070] FVCOM includes various physical, water quality, and ecological computational modules. The FVCOM model's input and output are based on the standardized NETCDF format, ensuring platform compatibility. The input and output structures can be quickly visualized in 2D / 3D using the VISIT visualization software. Based on the Fortran 90 / 95 standard and implemented in parallel computing within the MPI (Message Passing Interface) framework, the model can achieve fast parallel simulations on high-performance computers with multiple computing nodes, including those with shared and distributed memory.

[0071] The overall concept of this invention is as follows:

[0072] 1) Use the FVCOM model to generate simulated ocean dynamics data, such as temperature, salinity, and current velocity; standardize or normalize these data to adapt them to the input requirements of deep neural network models.

[0073] 2) Select a suitable deep neural network architecture, such as AlexNet, VGGNet, or a custom architecture; design network layers, including convolutional layers, pooling layers, and fully connected layers. Train the deep neural network using historical datasets to learn patterns in ocean data, and optimize the network weights using the backpropagation algorithm to obtain the residual circulation prediction model.

[0074] 3) Using the output data of the FVCOM model, spatial features are extracted through the residual circulation prediction model to capture local features in ocean data, such as eddies and fronts.

[0075] 4) Use the output of the residual circulation prediction model as the input to the FVCOM model or as the basis for adjusting the boundary conditions. For example, the residual circulation prediction model can predict velocity changes under specific conditions, and these prediction results can be fed back into the FVCOM model to adjust its boundary conditions.

[0076] 5) Repeat the process of using the FVCOM model to generate ocean dynamics simulation data as input to the residual circulation prediction model, and updating the boundary conditions of the FVCOM model based on the residual circulation distribution prediction data output by the residual circulation prediction model, until the preset number of simulations is reached, and the residual circulation distribution prediction can be obtained.

[0077] This invention combines the FVCOM model with a deep neural network. The FVCOM model generates ocean dynamics simulation data, which is then used as input to the residual circulation prediction model. The boundary conditions of the FVCOM model are updated based on the residual circulation distribution prediction data output by the residual circulation prediction model. This data-driven approach improves the model's ability to predict the residual circulation distribution field, thereby enhancing the accuracy and efficiency of residual circulation distribution prediction. Furthermore, the wind field distribution prediction can be determined based on the obtained residual circulation distribution prediction, further improving the accuracy and efficiency of wind field distribution prediction.

[0078] It should be noted that the FVCOM model is a well-known model in the prior art. The focus of this embodiment is to use the model to obtain ocean dynamics simulation data of the target sea area. The specific construction of the FVCOM model and the process of ocean numerical simulation are not described in detail. Only the updating of its boundary conditions (input data) and the output ocean dynamics simulation data are considered.

[0079] As an optional implementation method, the basic marine data includes seabed topography data, coastline shape data, and seawater level data of the target marine area, and the multi-source environmental data includes meteorological data, ocean current data, seawater temperature data, and seawater salinity data.

[0080] The training process of the residual circulation prediction model in this embodiment of the invention will be described in detail below.

[0081] First, we will introduce the design of the residual circulation prediction model in this embodiment of the invention.

[0082] 1) Data preprocessing and augmentation

[0083] Standardization / Normalization: Scales the input data to a uniform range (such as 0 to 1 or -1 to 1) to speed up convergence.

[0084] Data augmentation: Increase data diversity and reduce overfitting by using methods such as rotation, scaling, cropping, and flipping.

[0085] The embodiments of the present invention use automated data augmentation techniques, such as using neural networks to predict the optimal augmentation parameters; and customize data augmentation strategies for the data characteristics of specific sea areas to better simulate changes in the actual marine environment.

[0086] 2) Network architecture design

[0087] Choose the right infrastructure: Select a suitable CNN architecture (such as AlexNet, VGGNet, ResNet, etc.).

[0088] Custom layer design: Custom layers are designed based on the characteristics of ocean dynamics data. An attention mechanism is introduced to enhance the model's ability to identify key features. Residual connections are designed to solve the gradient vanishing problem in deep network training, thereby improving the model's ability to identify key ocean features.

[0089] 3) Hyperparameter selection

[0090] Learning rate: Choosing an appropriate learning rate can be determined through cross-validation to ensure the speed and stability of the loss function's decrease during model training.

[0091] Batch size: Determines the number of samples in each batch, which affects the convergence speed and stability of the model.

[0092] Optimizer: Select an appropriate optimization algorithm, such as SGD, Adam, RMSprop, etc.

[0093] The embodiments of the present invention use Bayesian optimization or genetic algorithms to automatically find the optimal hyperparameters and dynamically adjust the learning rate and other hyperparameters during training to adapt to the learning progress of the model.

[0094] 4) Model Training

[0095] Training strategy: Use early stopping (stop training when performance on the validation set no longer improves) to avoid overfitting.

[0096] Model evaluation: Evaluate model performance on independent test sets, using metrics such as accuracy and recall to measure the model's predictive ability.

[0097] This invention trains the model on multiple related tasks simultaneously to improve its generalization ability; uncertainty estimation is incorporated into the model training process to better evaluate the reliability of the model's predictions.

[0098] 5) Model optimization

[0099] Transfer learning: Using pre-trained models for transfer learning, leveraging features learned from similar tasks to improve the initial performance of the model.

[0100] Hyperparameter tuning: Based on the initial training results, further adjust the hyperparameters using methods such as grid search, random search, or Bayesian optimization.

[0101] The embodiments of this invention use meta-learning methods to quickly adapt to new tasks and reduce reliance on large amounts of labeled data; and improve the stability and accuracy of the model through model fusion techniques, such as Bagging or Boosting.

[0102] 6) Model Validation and Optimization

[0103] Cross-validation: K-fold cross-validation is used to evaluate the stability and generalization ability of the model.

[0104] Performance evaluation: Conduct a detailed analysis of the model's performance on different datasets to identify the model's strengths and weaknesses.

[0105] 7) Model Deployment and Application

[0106] Model deployment: Deploy the trained model into the actual marine monitoring system to ensure that the model can receive real-time data and make predictions.

[0107] Real-time monitoring: Utilize models for real-time residual circulation monitoring and prediction, and update models in a timely manner to adapt to environmental changes.

[0108] Association response mechanism: Establish an association response mechanism for monitoring points to identify abnormal monitoring points and classify them into different abnormal monitoring point groups.

[0109] 8) Results Analysis and Evaluation

[0110] Characteristic analysis: Based on the model prediction results, analyze the characteristics and motion laws of the residual circulation field, such as velocity, direction, temperature and salinity distribution.

[0111] Motion pattern assessment: Assess the motion pattern of the residual circulation and its impact on the wind field.

[0112] Analysis of influencing factors: This study explores factors that affect the characteristics of residual circulation, such as marine environment, topography, and climate change.

[0113] Risk assessment: Based on the model prediction results, assess the potential risks posed by residual circulation and propose corresponding countermeasures.

[0114] 9) Multi-source data fusion

[0115] Ensemble learning: Ensemble learning methods combine information from different sensors and data sources to improve data accuracy.

[0116] Data fusion algorithms: Develop new data fusion algorithms that can process and integrate multimodal data from high-frequency ground wave radar, satellite remote sensing, buoys and weather stations.

[0117] 10) Adaptive data preprocessing

[0118] Intelligent denoising: Develop intelligent denoising algorithms that can adaptively identify and remove noise from data without losing important signal features.

[0119] Dynamic data cleaning: Implement dynamic data cleaning technology, which can identify outliers and missing data based on the real-time characteristics and historical patterns of the data.

[0120] 11) High-efficiency data compression

[0121] Lossless and lossy compression techniques: Researching and applying efficient data compression techniques to reduce storage requirements and increase data transmission speed, while maximizing the preservation of data integrity and availability.

[0122] Feature selection: Advanced feature selection techniques, such as model-based feature selection, automatically determine which data features are most critical for model training, thereby reducing unnecessary data storage and processing.

[0123] 12) Applications of Augmented Reality (AR) and Virtual Reality (VR) technologies

[0124] Interactive data visualization: Utilizing AR and VR technologies to provide more intuitive and interactive data visualization tools, helping researchers gain a deeper understanding of complex ocean data.

[0125] Virtual environment simulation: Simulating the marine environment in a VR environment allows researchers to explore and analyze data in a virtual three-dimensional space to discover potential patterns and correlations.

[0126] 13) Utilize blockchain technology to ensure data integrity

[0127] Data immutability: Blockchain technology ensures the integrity and immutability of data during collection, storage, and processing, increasing data credibility.

[0128] Decentralized data management: By leveraging the decentralized nature of blockchain, distributed storage and management of data are achieved, improving data security and access efficiency.

[0129] As an optional implementation, the residual circulation prediction model is trained through the following steps:

[0130] S201. Obtain first ocean dynamic sample data for multiple historical periods of the target sea area and corresponding residual circulation distribution labels. The residual circulation distribution labels include the distribution of residual circulation velocity, residual circulation direction and residual circulation intensity in the target sea area.

[0131] S202. A labeled dataset was constructed based on the first ocean dynamics sample data and the residual circulation distribution labels;

[0132] S203. Obtain second ocean dynamics sample data for multiple historical periods in the target sea area, and construct an unlabeled dataset based on the second ocean dynamics sample data;

[0133] S204. Perform semi-supervised learning on the pre-built convolutional neural network based on the labeled and unlabeled datasets to obtain a trained residual circulation prediction model.

[0134] Specifically, embodiments of the present invention use a semi-supervised learning method to reduce the workload of data labeling. A large amount of unlabeled data is automatically labeled through an algorithm, and then reviewed and corrected by experts.

[0135] Semi-supervised learning can leverage large amounts of unlabeled data to improve model performance and reduce the workload of manual annotation. In the field of marine engineering, obtaining large amounts of accurately labeled data is often expensive and time-consuming, while unlabeled data, such as sensor data, is relatively easy to obtain. Self-training methods are suitable for datasets where the relationship between features and labels is relatively clear, while co-training is suitable for datasets that can be segmented by two different sets of features, each with a certain degree of discriminative power.

[0136] As a further optional implementation, a semi-supervised learning process is performed on the convolutional neural network based on labeled and unlabeled datasets to obtain a trained residual circulation prediction model, which specifically includes:

[0137] S2040. Input the labeled dataset into the convolutional neural network to obtain the trained initial model;

[0138] S2041. Predict the unlabeled dataset using the initial model, and select the prediction results with high confidence as pseudo-labels based on the confidence threshold.

[0139] S2042. Label the unlabeled data based on the pseudo-labels to obtain the pseudo-labeled dataset;

[0140] S2043. Fine-tune the initial model based on the pseudo-labeled dataset and the labeled dataset, and return the model to predict the unlabeled dataset using the initial model until the preset number of model fine-tuning times is reached, thus obtaining the trained residual circulation prediction model.

[0141] As a further optional implementation, a semi-supervised learning process is performed on the convolutional neural network based on labeled and unlabeled datasets to obtain a trained residual circulation prediction model, which specifically includes:

[0142] S2044. Divide the labeled dataset into a first data subset and a second data subset;

[0143] S2045. Input the first data subset and the second data subset into the convolutional neural network respectively to obtain the trained first initial model and the second initial model.

[0144] S2046. The unlabeled dataset is predicted using the first initial model and the second initial model respectively. Based on the confidence threshold, the prediction result with high confidence is selected as the pseudo label to obtain the first pseudo label and the second pseudo label.

[0145] S2047. Label the unlabeled data according to the first pseudo-label and the second pseudo-label respectively to obtain the first pseudo-label dataset and the second pseudo-label dataset.

[0146] S2048. Fine-tune the first initial model based on the second pseudo-label dataset and the first data subset, fine-tune the second initial model based on the first pseudo-label dataset and the second data subset, and return to predict the unlabeled dataset using the first initial model and the second initial model respectively, until the preset number of model fine-tuning times is reached, to obtain the trained first co-circulation prediction model and the second co-circulation prediction model.

[0147] S2049. Integrate the first residual circulation prediction model and the second residual circulation prediction model to obtain a trained residual circulation prediction model.

[0148] Specifically, the implementation steps of semi-supervised learning in this embodiment of the invention are as follows:

[0149] 1) Data preprocessing: Cleaning, standardizing and normalizing the collected marine data.

[0150] 2) Initial model training: Train the initial model using labeled data until it converges.

[0151] 3) Pseudo-tag generation:

[0152] Self-training: The initial model is used to predict unlabeled data, and high-confidence predictions are selected as pseudo-labels based on a confidence threshold.

[0153] Co-training: If co-training is used, the dataset is divided into two sets of features, two models are trained separately, and pseudo-labels are provided to each other.

[0154] 4) Model fine-tuning: Add pseudo-labeled data to the labeled data and continue training the model.

[0155] 5) Iterative optimization: Repeat the pseudo-label generation step and the model fine-tuning step until the model performance no longer improves or the predetermined number of iterations is reached.

[0156] 6) Model evaluation: Evaluate the model performance on the validation set to ensure that the newly added pseudo-label data does indeed improve the model performance.

[0157] 7) Manual review: A manual review process can be set up for the automatically labeled results to further improve the accuracy of data labeling.

[0158] In addition, semi-supervised learning can be achieved by combining self-training and collaborative training methods, as follows:

[0159] 1) Feature selection: First, analyze the features in the dataset to determine whether the features can be divided into two groups, each with good discriminative ability.

[0160] 2) Co-training as the primary method: If the dataset meets the conditions for co-training, the co-training method can be used first, allowing the two models to provide pseudo-labels independently.

[0161] 3) Self-training as a supplement: Based on collaborative training, for those samples with high confidence predicted by each model, self-training methods can be used to further generate pseudo-labels.

[0162] 4) Dynamic adjustment: Based on the model's performance on the validation set, dynamically adjust the confidence threshold and other hyperparameters during self-training and co-training processes.

[0163] 5) Ensemble learning: Integrating models obtained from self-training and co-training to improve the robustness and accuracy of the final model.

[0164] As a further, optional implementation, a labeled dataset is input into a convolutional neural network to obtain a trained initial model, which specifically includes:

[0165] S20401. Divide the labeled dataset into a training set, a validation set, and a test set;

[0166] S20402. Input the training set into the convolutional neural network to obtain the prediction results of the residual circulation distribution;

[0167] S20403. Determine the loss value based on the residual circulation distribution prediction results and residual circulation distribution labels;

[0168] S20404. Update the internal parameters of the convolutional neural network based on the loss value, and input the validation set into the updated convolutional neural network to obtain the model performance of the convolutional neural network on the validation set.

[0169] S20405. Update the hyperparameters of the convolutional neural network based on the model performance, and return to input the training set into the convolutional neural network until the model performance no longer improves in multiple consecutive iterations, thus obtaining the first initial model.

[0170] S20406. Input the test set into the first initial model to obtain the model evaluation index of the first initial model;

[0171] S20407. When the model evaluation metrics meet the preset requirements, the first initial model is used as the trained initial model.

[0172] Specifically, the steps and details for training the initial model using a labeled dataset are as follows:

[0173] 1. Selection and preparation of datasets

[0174] 1) Data collection: Select multi-source datasets including high-frequency ground wave radar data, meteorological station data, satellite remote sensing data, etc.

[0175] 2) Data labeling: Accurately label the data, especially in the identification of residual circulation features, to ensure the accuracy of the training data.

[0176] 3) Data splitting: The dataset is split into training set, validation set and test set, with a ratio of 70% training set, 15% validation set and 15% test set.

[0177] 4) Data balancing: If the number of samples in some categories in the dataset is unbalanced, the data balancing method is to oversample the minority class or undersample the majority class.

[0178] 2. Considerations for influencing variables

[0179] 1) Environmental factors: including wind speed, wind direction, temperature, salinity, tides, etc.

[0180] 2) Time series: Consider the time series characteristics of the data to capture seasonal and periodic changes.

[0181] 3) Spatial characteristics: Analyze the spatial distribution characteristics of the sea area, such as seabed topography and coastline shape.

[0182] 3. Specific steps of model training

[0183] 1) Determine the input parameters: Input parameters typically include ocean dynamic characteristics such as temperature, salinity, current velocity, and current direction, as well as possible temporal and spatial characteristics.

[0184] 2) Determine the output parameters: The output parameters are the targets predicted by the model, such as the velocity, direction or intensity changes of the residual circulation.

[0185] 3) Select a CNN architecture: Choose a CNN architecture that is suitable for the characteristics of ocean data, such as AlexNet, VGGNet or ResNet.

[0186] 4) Define the loss function: Select an appropriate loss function based on the task type (such as classification or regression), such as mean squared error (MSE) or cross-entropy loss.

[0187] 5) Select an optimizer: Choose an optimization algorithm to update the network weights, such as SGD, Adam, or RMSprop.

[0188] 6) Set learning rate and hyperparameters: Determine the learning rate, batch size, number of training epochs, and other hyperparameters.

[0189] 7) Training process: The CNN model is trained using the training dataset. In each epoch, data is input into the model in batches, and forward and back propagation are performed to update the network weights.

[0190] 8) Validation and early stopping: After each epoch, evaluate the model performance using a validation set. If the performance on the validation set no longer improves over several consecutive epochs, stop training to avoid overfitting.

[0191] 9) Model evaluation: Evaluate model performance on an independent test set using accuracy, recall, F1 score or other relevant metrics.

[0192] 10) Model tuning: Adjust the network architecture, hyperparameters, or training process based on the model's performance on the test set to optimize model performance.

[0193] 11) Feature importance analysis: Analyze the importance of each feature in the model to understand which input parameters contribute the most to the model's prediction.

[0194] 4. Input and Output Parameters

[0195] Input parameters include pre-processed multi-dimensional data such as ocean current velocity, direction, temperature, and salinity.

[0196] Output parameters: The distribution characteristics of the residual circulation predicted by the model, such as flow velocity, flow direction, vortex intensity, etc.

[0197] 5. Improvements and Optimizations

[0198] 1) Transfer learning and fine-tuning

[0199] Pre-trained models: Use models pre-trained on similar tasks as a starting point. These models have been trained on large amounts of data and are able to capture general features.

[0200] Fine-tuning: Based on the pre-trained model, fine-tuning is performed according to the characteristics of the current task. This may include replacing the last few layers of the network or adding new layers to the existing network.

[0201] Optimization measures: Make detailed adjustments to the structure of the CNN model, such as increasing or decreasing the number of layers, adjusting the size of the convolutional kernels, and introducing or optimizing residual connections.

[0202] 2) Hyperparameter optimization

[0203] Automated search: Using methods such as Bayesian optimization and genetic algorithms to automatically search for the optimal combination of hyperparameters.

[0204] Hyperparameter space definition: Defines the search space for hyperparameters such as learning rate, batch size, number of network layers, and number of neurons.

[0205] Optimization measures: Use automated hyperparameter optimization techniques, such as grid search, random search, or Bayesian optimization, to find the optimal model parameters.

[0206] 3) Network architecture innovation

[0207] Custom layers: Design custom layers or modules, such as attention mechanisms and residual connections, to enhance the model's ability to capture key information.

[0208] Multi-scale feature fusion: Combining features at different scales to capture information from local to global perspectives.

[0209] 4) Loss function optimization

[0210] Custom loss function: Design a loss function according to the task requirements. For example, for imbalanced datasets, a weighted loss function can be used to improve the model's ability to identify the minority class.

[0211] Optimization measures: Design or select a loss function suitable for predicting residual circulation characteristics, which may include weighting the loss function to address imbalances in the data.

[0212] 5) Regularization techniques

[0213] L1 / L2 regularization: Adding L1 or L2 regularization during model training reduces model complexity and overfitting.

[0214] Dropout: Randomly discards some network connections during training to prevent the network from overfitting to the training data.

[0215] Optimization measures: Apply techniques such as Dropout and L1 / L2 regularization to reduce overfitting and improve the model's generalization ability.

[0216] 6) Integrated learning strategy

[0217] Optimization measures: Employ ensemble learning methods, such as Bagging or Boosting, to combine the prediction results of multiple models to improve the overall stability and accuracy of predictions.

[0218] 7) Real-time performance optimization

[0219] Optimization measures: Optimize the model to meet real-time or near-real-time prediction requirements, including techniques such as model pruning, quantization, and hardware acceleration.

[0220] In some alternative embodiments, the trained model can be deployed on edge computing devices, such as offshore buoys or near-shore platforms, to enable real-time data processing and analysis and reduce data transmission latency.

[0221] Specifically, the model is optimized to adapt to the computing power of edge devices; data preprocessing and model inference are performed on edge devices; and key prediction results are transmitted in real time via wireless communication technology.

[0222] In some alternative embodiments, an interactive marine environment simulation and visualization platform can be developed, enabling users to intuitively understand residual circulation characteristics and their impact on wind field response.

[0223] Specifically, it integrates model prediction results into marine environmental simulation software; provides a 3D visualization interface to display the dynamic changes of residual circulation; and allows users to explore the impact of different environmental factors on residual circulation through interactive tools.

[0224] In some alternative embodiments, an adaptive data collection system can be developed that dynamically adjusts the frequency and scope of data collection based on the model’s predictive needs.

[0225] Specifically, the integrated model is fed back into the data collection strategy, enabling the system to automatically adjust the data collection plan based on the uncertainty of the forecast; enabling remote control and scheduling of data collection equipment; and optimizing data storage and processing flow to support adaptive data collection.

[0226] In some alternative embodiments, an interdisciplinary decision support system can be built, integrating data and models from multiple fields such as oceanography, meteorology, and environmental science to support ocean management and policy making.

[0227] Specifically, it integrates models and data sources from different fields to establish a comprehensive decision-making platform; develops decision support tools such as scenario analysis and risk assessment; and provides decision-makers with a user-friendly interface and report generation tools.

[0228] As an optional implementation, the residual circulation monitoring method further includes the following steps:

[0229] S105. Obtain historical residual circulation distribution data and historical wind field observation data for the target area;

[0230] S106. Perform regression analysis on historical residual circulation distribution data and historical wind field observation data to obtain the data relationship between residual circulation distribution and wind field parameters.

[0231] S107. Determine the wind field distribution prediction of the target area based on the residual circulation distribution prediction and data relationships.

[0232] Specifically, based on the prediction results, the characteristics, motion patterns, and impact on the wind field of the residual circulation field are analyzed and evaluated, as follows:

[0233] 1) Determine the evaluation objectives and indicators

[0234] Characteristic evaluation: Evaluate the physical properties of the residual circulation, such as velocity, direction, temperature, and salinity.

[0235] Motion pattern assessment: Analysis of the periodicity, stability, and spatial distribution pattern of the residual circulation.

[0236] Wind field impact assessment: quantifying the impact of residual circulation on wind field parameters such as wind speed, wind direction, and wind energy density.

[0237] 2) Data collection and processing

[0238] Historical observation data of residual circulation and wind field, as well as model prediction results, were collected.

[0239] Perform data cleaning, standardization, and integration to ensure data quality.

[0240] 3) Statistical analysis

[0241] Descriptive statistics: Calculate the mean, median, standard deviation, etc., to describe the basic characteristics of residual circulation and wind field.

[0242] Correlation analysis: using correlation coefficients (such as Pearson correlation coefficient) to assess the linear relationship between variables.

[0243] Regression analysis: Using linear or nonlinear regression models to evaluate the relationship between residual circulation characteristics and wind field parameters.

[0244] 4) Model Validation

[0245] Cross-validation: Use methods such as K-fold cross-validation to evaluate the model's prediction accuracy and generalization ability.

[0246] Error analysis: Calculate indicators such as mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) to assess prediction error.

[0247] 5) Dynamic analysis

[0248] Flow field visualization: Visualize the flow characteristics of the residual circulation using tools such as streamline diagrams and velocity vector fields.

[0249] Vortex analysis: Calculate vorticity and vortexes to analyze the dynamic structure of the residual circulation.

[0250] 6) Analysis of influencing factors

[0251] Identify and quantify key factors that influence the characteristics and motion patterns of residual circulation, such as topography, climate conditions, and seasonal variations.

[0252] 7) Climate impact assessment

[0253] Climate model integration: Combining residual circulation models with climate models to assess their impact on regional climate patterns.

[0254] Extreme event analysis: assessing the residual circulation response to extreme weather events (such as typhoons and storm surges).

[0255] 8) Model prediction and risk assessment

[0256] Scenario analysis: Simulate different wind field scenarios to predict changes in residual circulation and its potential impacts.

[0257] Risk Matrix: Construct a risk matrix to assess the potential risks and impacts under different scenarios.

[0258] 9) Comprehensive Assessment Report

[0259] Prepare an assessment report, summarizing the analysis results of residual circulation characteristics, motion patterns, and wind field influence.

[0260] Provides charts, data analysis, and professional explanations to make the report easy to understand.

[0261] 10) Application and Decision Support

[0262] The assessment results will be applied to fields such as marine engineering and wind power generation to provide a scientific basis for decision-making.

[0263] The relevant optimization measures are as follows:

[0264] 1) Multi-dimensional feature analysis:

[0265] Develop a multi-dimensional feature analysis method that considers not only flow velocity and direction, but also ocean physical parameters such as temperature, salinity, and pressure, to comprehensively evaluate the characteristics of residual circulation.

[0266] 2) Dynamic time series analysis:

[0267] Dynamic time series analysis techniques, such as Long Short-Term Memory (LSTM) networks or Temporal Convolutional Networks (TCN), are introduced to capture the temporal dynamics of the residual circulation.

[0268] 3) Weighting of influencing factors:

[0269] Machine learning algorithms, such as random forests or gradient boosting decision trees, are used to evaluate the weights of different environmental factors on the residual circulation characteristics.

[0270] 4) Interactive visualization tools:

[0271] Develop interactive visualization tools that allow users to dynamically adjust parameters and view changes in residual circulation characteristics in real time according to different needs and scenarios.

[0272] 5) Quantification of prediction uncertainty:

[0273] Introduce techniques for quantifying prediction uncertainty, such as Bayesian methods or ensemble learning, to provide an assessment of the reliability of prediction results.

[0274] 6) Interdisciplinary impact assessment:

[0275] By combining research findings from multiple disciplines such as oceanography, meteorology, and ecology, the impact of residual circulation on wind fields and the wider environment is assessed.

[0276] 7) Real-time data feedback mechanism:

[0277] Establish a real-time data feedback mechanism to feed the latest observation data into the model in order to dynamically adjust and optimize the prediction results.

[0278] The method steps of the embodiments of the present invention have been described above. It can be understood that the embodiments of the present invention combine the FVCOM model with a deep neural network, using the FVCOM model to generate ocean dynamics simulation data as input to the residual circulation prediction model. The boundary conditions of the FVCOM model are updated based on the residual circulation distribution prediction data output by the residual circulation prediction model. This data-driven approach improves the model's predictive ability for the residual circulation distribution field, thereby improving the accuracy and efficiency of residual circulation distribution prediction. Furthermore, the wind field distribution prediction can be determined based on the obtained residual circulation distribution prediction, improving the accuracy and efficiency of wind field distribution prediction. The embodiments of the present invention have broad application prospects in marine engineering, wind power generation, and other fields, and are expected to provide strong technical support for the development of related fields.

[0279] Reference Figure 2 This invention provides a residual circulation monitoring system, comprising:

[0280] The data acquisition module is used to acquire basic marine area data and multi-source environmental data of the target sea area, and to build an FVCOM model based on the basic marine area data;

[0281] The marine numerical simulation module is used to perform marine numerical simulations of the target sea area using multi-source environmental data as boundary conditions and the FVCOM model, thereby obtaining marine dynamic simulation data of the target sea area.

[0282] The residual circulation distribution prediction module is used to input ocean dynamics simulation data into a pre-trained residual circulation prediction model for the target sea area to obtain residual circulation distribution prediction data for the target sea area.

[0283] The boundary condition update module is used to update the boundary conditions based on the residual circulation distribution prediction data, and return the results of ocean numerical simulation of the target sea area using the FVCOM model until the preset number of simulations is reached, so as to obtain the residual circulation distribution prediction of the target sea area.

[0284] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0285] Reference Figure 3 This invention provides a residual circulation monitoring device, comprising:

[0286] At least one processor;

[0287] At least one memory for storing at least one program;

[0288] When the above-mentioned at least one program is executed by the above-mentioned at least one processor, the above-mentioned at least one processor implements the above-mentioned residual current monitoring method.

[0289] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0290] This invention also provides a computer-readable storage medium storing a processor-executable program that, when executed by a processor, performs the aforementioned residual current monitoring method.

[0291] This invention provides a computer-readable storage medium that can execute a residual current monitoring method provided in the method embodiments of this invention. It can execute any combination of the implementation steps of the method embodiments and has the corresponding functions and beneficial effects of the method.

[0292] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform... Figure 1 The method shown.

[0293] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the aforementioned blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0294] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0295] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0296] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0297] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the aforementioned program can be printed, because the aforementioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or, if necessary, processing in other suitable ways, and then stored in computer memory.

[0298] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0299] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0300] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0301] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for monitoring residual circulation, characterized in that, Includes the following steps: Acquire basic marine area data and multi-source environmental data for the target sea area, and establish an FVCOM model based on the basic marine area data; Using the multi-source environmental data as boundary conditions, the target sea area is subjected to marine numerical simulation through the FVCOM model to obtain marine dynamic simulation data of the target sea area; The ocean dynamics simulation data is input into the pre-trained residual circulation prediction model of the target sea area to obtain the residual circulation distribution prediction data of the target sea area. The boundary conditions are updated based on the residual circulation distribution prediction data, and the target sea area is simulated using the FVCOM model until the preset number of simulations is reached to obtain the residual circulation distribution prediction of the target sea area.

2. The residual circulation monitoring method according to claim 1, characterized in that, The basic marine data includes seabed topography data, coastline shape data, and seawater level data of the target marine area. The multi-source environmental data includes meteorological data, ocean current data, seawater temperature data, and seawater salinity data.

3. The residual circulation monitoring method according to claim 1, characterized in that, The residual circulation prediction model is trained through the following steps: Acquire first ocean dynamic sample data for multiple historical time periods of the target sea area and corresponding residual circulation distribution labels. The residual circulation distribution labels include the distribution of residual circulation velocity, residual circulation direction and residual circulation intensity of the target sea area. A labeled dataset was constructed based on the first ocean dynamics sample data and the residual circulation distribution labels; Acquire second ocean dynamics sample data for multiple historical periods of the target sea area, and construct an unlabeled dataset based on the second ocean dynamics sample data; The pre-built convolutional neural network is semi-supervised learning based on the labeled dataset and the unlabeled dataset to obtain the trained residual circulation prediction model.

4. The residual circulation monitoring method according to claim 3, characterized in that, The step of performing semi-supervised learning on the convolutional neural network based on the labeled dataset and the unlabeled dataset to obtain the trained co-circulation prediction model specifically includes: The labeled dataset is input into the convolutional neural network to obtain a trained initial model. The initial model is used to predict the unlabeled dataset, and the prediction results with high confidence are selected as pseudo-labels based on the confidence threshold. The unlabeled dataset is labeled according to the pseudo-labels to obtain a pseudo-labeled dataset; The initial model is fine-tuned based on the pseudo-labeled dataset and the labeled dataset, and the prediction of the unlabeled dataset is performed using the initial model until a preset number of model fine-tunings is reached, resulting in the trained residual circulation prediction model.

5. The residual circulation monitoring method according to claim 3, characterized in that, The step of performing semi-supervised learning on the convolutional neural network based on the labeled dataset and the unlabeled dataset to obtain the trained co-circulation prediction model specifically includes: The labeled dataset is divided into a first data subset and a second data subset; The first data subset and the second data subset are respectively input into the convolutional neural network to obtain the trained first initial model and the second initial model; The first initial model and the second initial model are used to predict the unlabeled dataset respectively. Based on the confidence threshold, the prediction result with high confidence is selected as the pseudo label to obtain the first pseudo label and the second pseudo label. The unlabeled dataset is labeled according to the first pseudo-label and the second pseudo-label respectively to obtain the first pseudo-labeled dataset and the second pseudo-labeled dataset; The first initial model is fine-tuned based on the second pseudo-label dataset and the first data subset, and the second initial model is fine-tuned based on the first pseudo-label dataset and the second data subset. The model is then returned to predict the unlabeled dataset using the first initial model and the second initial model respectively, until the preset number of model fine-tuning times is reached, resulting in a trained first co-circulation prediction model and a second co-circulation prediction model. The first residual circulation prediction model and the second residual circulation prediction model are integrated for learning to obtain the trained residual circulation prediction model.

6. The residual circulation monitoring method according to claim 4, characterized in that, The step of inputting the labeled dataset into the convolutional neural network to obtain a trained initial model specifically includes: The labeled dataset is divided into a training set, a validation set, and a test set; The training set is input into the convolutional neural network to obtain the prediction result of the residual circulation distribution; The loss value is determined based on the residual circulation distribution prediction results and the residual circulation distribution labels; The internal parameters of the convolutional neural network are updated based on the loss value, and the validation set is input into the updated convolutional neural network to obtain the model performance of the convolutional neural network on the validation set. The hyperparameters of the convolutional neural network are updated based on the model performance, and the training set is returned to be input into the convolutional neural network until the model performance no longer improves in multiple consecutive iterations, thus obtaining the first initial model. The test set is input into the first initial model to obtain the model evaluation index of the first initial model; When the model evaluation metric meets the preset requirements, the first initial model is used as the trained initial model.

7. A residual circulation monitoring method according to any one of claims 1 to 6, characterized in that, The residual circulation monitoring method further includes the following steps: Acquire historical residual circulation distribution data and historical wind field observation data for the target sea area; Regression analysis was performed on the historical residual circulation distribution data and the historical wind field observation data to obtain the data relationship between the residual circulation distribution and the wind field parameters. The wind field distribution prediction for the target sea area is determined based on the predicted residual circulation distribution and the data relationship.

8. A residual circulation monitoring system, characterized in that, include: The data acquisition module is used to acquire basic marine area data and multi-source environmental data of the target sea area, and to establish an FVCOM model based on the basic marine area data; The marine numerical simulation module is used to perform marine numerical simulation on the target sea area using the multi-source environmental data as boundary conditions and the FVCOM model to obtain marine dynamic simulation data of the target sea area. The residual circulation distribution prediction module is used to input the ocean dynamics simulation data into the pre-trained residual circulation prediction model of the target sea area to obtain the residual circulation distribution prediction data of the target sea area. The boundary condition update module is used to update the boundary conditions based on the residual circulation distribution prediction data, and return the ocean numerical simulation of the target sea area using the FVCOM model until a preset number of simulations is reached to obtain the residual circulation distribution prediction of the target sea area.

9. A residual circulation monitoring device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a residual circulation monitoring method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform a residual circulation monitoring method as described in any one of claims 1 to 7.

Images