A lightweight shipborne marine environment forecasting method, system and shipborne terminal

Abstract

This application discloses a lightweight shipborne marine environment forecasting method, system, and shipborne terminal. The method includes: the shipborne terminal sending a data request command to a shore-based cloud server; receiving an initial wind field forecast data packet from the cloud server; decoding the initial wind field forecast data packet and transcoding it into a preset format to obtain transcoded initial wind field forecast data; inputting the initial wind field forecast data into a pre-trained wind field inference model to output optimized target wind field forecast data; generating corresponding initial wave field data based on the optimized target wind field forecast data using empirical formulas for wind-wave relationships; and inputting the initial wave field data into a pre-trained wave field inference model to output target wave field forecast data. This application offers high timeliness, small data download size, flexible forecast area selection, and fast processing speed, providing timely and accurate refined support data and products for ships.

Description

Technical Field

[0001] The present application relates to the technical field of environmental forecasting, and in particular to a lightweight shipborne marine environment forecasting method, system and shipborne terminal. Background Art

[0002] Timely and accurate forecasts of marine environmental factors are the basis for ensuring safe navigation of ships at sea. At present, there are two main ways for domestic ships to obtain marine environmental forecasts. Commercial transport ships mainly purchase commercial meteorological navigation services, while official and scientific research ships obtain forecasts through the forecast guarantee services provided by relevant national marine meteorological forecasting departments.

[0003] Commercial weather navigation service is a relatively mature form of support. Its marine environment forecast products are mainly produced for pre-set routes. The content and form of the forecast products are relatively fixed and are sent to the support ship by email via maritime satellite. The National Marine Meteorological Forecast Department generally sends marine environment forecast products produced on land to official and scientific research ships. The forecast products are mainly for the sea area where the ship is located or the pre-set route. The product content may change due to changes in the ship's navigation plan. Generally, regular communication between the ship and the land is required. The forecast products sent to the ship via maritime satellite will be used and interpreted by relevant responsible personnel to achieve forecast support services.

[0004] The current mainstream ocean and atmosphere forecasting method is still the numerical forecasting method. The forecasting guarantee products provided by the National Oceanic and Meteorological Forecasting Department to official and scientific research ships are also processed and produced based on the forecast results of numerical models. The business process is to first obtain the initial forecast field of the numerical model from the Internet, and then read and process the initial field data and other observation data and input them into the numerical model. After the integral calculation of the model, the forecast result of a certain area is obtained, and then the forecast result is post-processed by graphical methods to obtain a usable forecast product. The time required for this process varies greatly depending on the computing power of large computers (machine clusters). Generally, it takes 1-2 hours. If the spatial resolution of the forecast results is required to be high, the numerical model also needs to be nested in regional grids, and the entire calculation time will increase to 2-3 hours. In addition, the operation of the wind and wave intelligent forecasting system currently used in business scientific research development requires not only the download of wind field data, but also the download of the initial field data of the waves. The large amount of data download affects the computing efficiency.

[0005] In addition, the forecast products produced are generally in the form of pictures showing the marine environmental elements of the sea area where the ship is located or on the planned route. The product form is relatively simple and the content is relatively fixed. Although the regional forecast data results calculated by the numerical model can now be sent to the ship through satellite transmission, and then the display system on the ship completes the data analysis and display, the forecast data is generally streamlined and compressed, and the data resolution and accuracy will be significantly reduced. In addition, the forecast area is fixed and cannot be customized flexibly. Summary of the invention

[0006] The present application provides a lightweight shipborne marine environment forecasting method, system and shipborne terminal, aiming to solve the technical problems that the existing methods for obtaining marine environment forecasts by marine ships have poor timeliness, large data download volume and cannot be customized independently.

[0007] In a first aspect, a lightweight shipborne marine environment forecasting method is applied to a shipborne terminal, wherein the shipborne terminal is deployed on a public service or scientific research vessel, and the method comprises:

[0008] Send data request instructions to the cloud server deployed on the shore;

[0009] After the cloud server divides, compresses, and encapsulates the latest initial wind farm forecast data into data packets according to the data request instruction, receiving the initial wind farm forecast data packet sent by the cloud server;

[0010] Decoding the initial wind farm forecast data packet, and transcoding the initial wind farm forecast data packet into a preset format to obtain transcoded initial wind farm forecast data;

[0011] Inputting the initial wind field forecast data into a pre-trained wind field inference model, and outputting optimized target wind field forecast data;

[0012] Using the empirical formula of wind-wave relationship, the corresponding initial wave field data is generated according to the optimized target wind field forecast data;

[0013] The initial ocean wave field data is input into a pre-trained ocean wave field inference model, and target ocean wave field forecast data is output.

[0014] Optionally, the data request instruction includes area information, element information, forecast duration information, and spatiotemporal resolution information.

[0015] Optionally, the wind field inference model is trained using a residual network model, and the wave field inference model is trained using a long-term and short-term convolutional network model.

[0016] Further optionally, the training process of the wind field inference model includes:

[0017] Collect historical wind field observation data and corresponding wind field forecast data for the world or sea area;

[0018] Performing spatiotemporal fusion processing on the historical wind field observation data and the corresponding wind field forecast data to form a real-time analysis set;

[0019] Dividing the actual analysis set into a training set, a validation set and a test set according to a preset ratio;

[0020] The residual network model is trained using the training set to obtain a wind field inference model; during the training process, the model parameters of the residual network model are continuously adjusted through an optimization algorithm;

[0021] The wind farm inference model is evaluated using the validation set, and the wind farm inference model is tuned according to the evaluation result.

[0022] Further optionally, the training process of the wave field inference model includes:

[0023] Collect historical wave field observation data and corresponding wave field forecast data for the world or sea area;

[0024] Performing spatiotemporal fusion processing on the historical wave field observation data and the corresponding wave field forecast data to form a real-time analysis set;

[0025] Dividing the actual analysis set into a training set, a validation set and a test set according to a preset ratio;

[0026] The long-term and short-term convolutional network model is trained using the training set to obtain a wave field inference model; during the training process, the model parameters of the long-term and short-term convolutional network model are continuously adjusted through an optimization algorithm;

[0027] The wave field reasoning model is evaluated using the validation set, and the wave field reasoning model is tuned according to the evaluation result.

[0028] Optionally, the wind-wave relationship empirical formula is obtained by fitting wind field observation data and sea wave field observation data.

[0029] Optionally, the method further comprises:

[0030] Based on the target wind field forecast data and the target wave field forecast data, wind and wave forecast products in various formats are drawn and output according to pre-set drawing parameters.

[0031] In a second aspect, a lightweight shipborne marine environment forecasting system includes a forecast data cloud service system and a shipborne intelligent forecasting system, wherein the forecast data cloud service system runs on a cloud server deployed on a shore, and the shipborne intelligent forecasting system runs on a shipborne terminal deployed on a public service or scientific research vessel;

[0032] The forecast data cloud service system includes a data acquisition and archiving module, a data request and response module, and a data segmentation and packaging module; the shipborne intelligent forecast system includes a data acquisition subsystem and a wind and wave intelligent forecast subsystem, the data acquisition subsystem includes a data customization request module and a data decoding and transcoding module, and the wind and wave intelligent forecast subsystem includes a wind field intelligent reasoning and calculation module, a sea wave initial field preprocessing module, and a wave field intelligent reasoning and calculation module;

[0033] The data acquisition and archiving module is used to regularly monitor and scan the data servers of mainstream forecasting agencies at home and abroad, and obtain the latest initial wind field forecast data from the data servers of the mainstream forecasting agencies at home and abroad and archive them; the data request response module is used to receive the data request instruction sent by the shipboard terminal; the data segmentation and packaging module is used to segment, compress, and package the latest initial wind field forecast data into data packets according to the data request instruction, and send the initial wind field forecast data packets to the shipboard terminal;

[0034] The data customization request module is used to send a data request instruction to the cloud server; the data decoding and transcoding module is used to receive the initial wind field forecast data packet sent by the cloud server, decode the initial wind field forecast data packet, and transcode the initial wind field forecast data packet into a preset format to obtain the transcoded initial wind field forecast data; the wind field intelligent reasoning and calculation module is used to input the initial wind field forecast data into a pre-trained wind field reasoning model, and output the optimized target wind field forecast data; the wave initial field preprocessing module is used to generate corresponding initial wave field data according to the optimized target wind field forecast data by using the wind-wave relationship empirical formula; the wave field intelligent reasoning and calculation module is used to input the initial wave field data into a pre-trained wave field reasoning model, and output the target wave field forecast data.

[0035] Optionally, the wind and wave intelligent forecasting subsystem further includes a forecast element drawing module, which is used to draw and output wind and wave forecast products in various formats based on the target wind field forecast data and the target wave field forecast data according to preset drawing parameters.

[0036] In a third aspect, a shipborne terminal comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0037] Compared with the prior art, this application has at least the following beneficial effects:

[0038] The most concerned marine environmental elements in ensuring the safe navigation of ships at sea are wind and waves. The current application of deep learning networks based on artificial intelligence in marine meteorology has become increasingly mature, and the accuracy of intelligent forecasting results is very close to that of numerical model forecasts, and the computing speed is hundreds of times faster than that of numerical models. The embodiment of the present application has developed a lightweight wind and wave intelligent forecasting method and system suitable for shipboard based on a deep learning network model. Compared with the wind and wave intelligent forecasting model currently used in business, it does not need to download the initial wave field data, but only needs to download the wind field data, which saves the amount of data download for the shipboard system and improves the computing efficiency. This application not only The wind field data is corrected and optimized to improve the quality of the wind field. The initial wave field is calculated based on the wind-wave relationship formula fitted based on the observation data of different sea areas, which improves the accuracy of the initial wave field. In addition, the embodiment of the present application starts from the actual marine security business and can independently customize factors such as forecast area and forecast duration according to data request instructions. The present application improves the mature marine environment intelligent forecasting model to meet the needs of marine business applications. It has higher timeliness than traditional numerical forecasting systems, and has a small data download volume, flexible and optional forecast areas, and fast computing speed, which can provide ships with timely and accurate refined security data and products. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of an application environment of a lightweight shipborne marine environment forecasting method provided by an embodiment of the present application;

[0040] Figure 2 A schematic flow chart of a lightweight shipborne marine environment forecasting method provided in one embodiment of the present application;

[0041] Figure 3 A module architecture block diagram of a lightweight shipborne marine environment forecasting system provided in one embodiment of the present application;

[0042] Figure 4 This is an example diagram of a wind-wave relationship formula based on the fitting of satellite scatterometer and altimeter observation data in one embodiment of the present application;

[0043] Figure 5 An internal structure diagram of a shipborne terminal provided for one embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] In the description of the present application: unless otherwise specified, expressions such as “include”, “comprising”, “having”, etc. also mean “not limited to” (certain units, components, materials, steps, etc.).

[0046] The lightweight shipborne marine environment forecasting method provided in this application can be applied to Figure 1 In the application environment shown, the shipboard terminal 101 communicates with the cloud server 102 through a network. The shipboard terminal 101 can be, but is not limited to, various personal computers, laptops, and tablet computers, and the cloud server 102 can be implemented by an independent server or a server cluster consisting of multiple servers.

[0047] In one embodiment, Figure 2 As shown in the figure, a lightweight shipborne marine environment forecasting method is provided. Figure 1 Taking the shipborne terminal in the example as an example, the shipborne terminal is deployed on a public service or scientific research ship, and the method includes the following steps:

[0048] S1, the shipboard terminal sends a data request instruction to the cloud server deployed on the shore.

[0049] Among them, the data request instruction includes regional information, element information, forecast duration information, spatiotemporal resolution information and other information.

[0050] S2, after the cloud server divides, compresses and encapsulates the latest initial wind field forecast data into data packets according to the data request instruction, the shipboard terminal receives the initial wind field forecast data packet sent by the cloud server.

[0051] When the shipborne terminal sends a data request command to the cloud server, the cloud server will split and compress the latest forecast field data into a data packet based on the area, elements, forecast duration, spatiotemporal resolution and other command information requested for the data, and send the data packet to the specified file directory of the terminal via satellite to complete the data response. Data transmission and reception supports breakpoint resumption to adapt to the special situation where maritime communications are often interrupted.

[0052] S3, the shipborne terminal decodes the initial wind field forecast data packet, and transcodes the initial wind field forecast data packet into a preset format to obtain transcoded initial wind field forecast data.

[0053] The shipborne terminal will decode and transcode the data packets, transcode the compressed data into the required data format, and provide initial field data for subsequent intelligent forecast calculations.

[0054] S4, the shipborne terminal inputs the initial wind field forecast data into the pre-trained wind field inference model, optimizes the initial wind field forecast data, and outputs the optimized target wind field forecast data.

[0055] Among them, the wind farm inference environment is obtained through pre-training of the residual network model. The residual network is a deep learning neural network structure. By introducing a "residual block" to solve the problems of gradient disappearance and gradient explosion in deep network training, a very deep network model can be trained. In the method provided in this embodiment, the residual network model is used to train a wind farm inference model to improve the prediction accuracy of wind farm data. Using the wind farm inference model, it only takes 3-5 minutes to complete the intelligent forecast of the wind farm and output the forecast wind farm data with higher accuracy.

[0056] Furthermore, the training process of the wind farm inference model includes:

[0057] (1) Collect historical wind field observation data and corresponding wind field forecast data for the world or sea area; historical wind field observation data and corresponding wind field forecast data can come from multiple sources, including satellite observations, buoy observations, meteorological station observations, etc. The quality and accuracy of the data must be ensured during the data collection process;

[0058] (2) Processing historical wind farm observation data and corresponding wind farm forecast data; the processing first includes data cleaning, denoising, filling missing values, standardization and other operations, in order to improve the quality and availability of the data so that it can be better applied to model training; then the data is processed by spatiotemporal fusion to form a real-time analysis set;

[0059] (3) Dividing the actual analysis set data into a training set, a validation set, and a test set according to a preset ratio;

[0060] (4) Using the training set to train the residual network model, a wind field inference model is obtained; during the training process, the model parameters of the residual network model are continuously adjusted by an optimization algorithm; the network parameters can be continuously adjusted by a back propagation algorithm so that the model can learn and accurately predict the changes in the wind field; an optimization algorithm such as stochastic gradient descent (SGD) or its variants can be used to minimize the loss function;

[0061] For the residual network model, it is necessary to select an appropriate loss function to measure the difference between the model output and the actual forecast data. Common loss functions include mean square error (MSE) and so on.

[0062] (5) Use the validation set to evaluate the wind farm inference model and tune the wind farm inference model based on the evaluation results; for example, adjust hyperparameters such as learning rate and regularization to improve the model's generalization ability on new data.

[0063] S5, using the empirical formula of wind-wave relationship, generating corresponding initial wave field data according to the optimized target wind field forecast data.

[0064] Among them, the empirical formula of wind-wave relationship is obtained by fitting the wind field observation data and the sea wave field observation data. Using the empirical formula of wind-wave relationship, the initial sea wave field of the forecast area can be generated from the wind field (the data includes elements such as wave height, wave direction and wave period).

[0065] By using the empirical formula of wind-wave relationship, the initial wave field of the forecast area (data including elements such as wave height, wave direction and wave period) can be generated from the optimized wind field, without having to transmit this data separately, thus saving the bandwidth of the shipboard client.

[0066] S6, inputting the initial wave field data into the pre-trained wave field inference model, and outputting the target wave field forecast data.

[0067] Among them, the wave field inference model is trained using the long short-term convolutional network model. The long short-term convolutional network (LSTM-CNN) is a deep learning model that combines the long short-term memory network (LSTM) and the convolutional neural network (CNN) to process time series data and extract spatial features. Using the wave field inference model, high-precision wave forecast field data can be output after 3-5 minutes of inference calculation.

[0068] Furthermore, the training process of the wave field inference model includes:

[0069] (1) Collect historical wave field observation data and corresponding wave field forecast data for the world or sea area; these data may be obtained through observation equipment such as satellites and buoys;

[0070] (2) Processing the historical wave field observation data and the corresponding wave field forecast data, including data cleaning, normalization, time series data segmentation, etc., to ensure the data quality and the form suitable for model input; then the data is processed by spatiotemporal fusion to form a real-time analysis set;

[0071] (3) Dividing the actual analysis set data into a training set, a validation set, and a test set according to a preset ratio;

[0072] (4) Using the training set to train the long-term and short-term convolutional network model, a wave field inference model is obtained; during the training process, the model parameters of the long-term and short-term convolutional network model are continuously adjusted through the optimization algorithm; the prepared ocean wave field data is input into the LSTM-CNN model for training; during the training process, the parameters are updated through the back propagation algorithm and the optimizer, so that the model can gradually learn the spatiotemporal characteristics and change laws in the ocean wave field data;

[0073] (5) The wave field inference model is evaluated using the validation set, and the wave field inference model is tuned according to the evaluation results; during the tuning process, hyperparameters such as model structure, learning rate, and regularization may be adjusted.

[0074] Furthermore, the method further comprises:

[0075] Based on the target wind field forecast data and target wave field forecast data and according to pre-set drawing parameters, the shipborne terminal draws and outputs wind and wave forecast products in various formats.

[0076] After obtaining the target wind field forecast data and the target wave field forecast data, the shipborne terminal automatically identifies the regional scope, temporal and spatial resolution and other elements of the data, and produces wind and wave forecast products according to the pre-set drawing parameters, and outputs forecast products in various formats such as png and pdf.

[0077] In summary, the most concerned marine environmental elements in ensuring the safe navigation of ships at sea are wind and waves. The embodiment of the present application proposes a method for intelligent forecasting of wind and waves based on a deep learning network model. The operation of the intelligent wind and wave forecasting system currently used in business scientific research and development requires not only the download of wind field data, but also the download of the initial field data of the waves. However, through the method provided by the embodiment of the present application, it is only necessary to download the wind field data, and then use the empirical formula of wind and wave relationship to generate the initial field of waves in the forecast area. The wind and wave relationship formula based on the observation data fitting can be calculated separately according to different forecast sea areas, thereby improving the accuracy of the initial field of waves.

[0078] Figure 4 It is a wind-wave relationship equation based on the wind speed observed by satellite scatterometers and wave height data observed by altimeters in the northwest Pacific Ocean. After obtaining the wind field data in this area, the corresponding wave height data can be calculated using this formula. The wave period data can also be obtained based on the wave spectrum data observed by satellites or the Beaufort wind-wave relationship statistical fitting, so there is no need to download and transmit the initial wave field data separately. This part of the research and development work saves data downloads for the shipboard system while ensuring the accuracy of the forecast, and improves the computing efficiency.

[0079] In this method, rapid calculation and forecasting services are realized on high-sea vessels through data segmentation and packaging, breakpoint resumption and other technologies. Compared with the use of traditional numerical forecasting methods for high-sea ships to carry out forecasting guarantees, the method provided in the embodiment of the present application has the advantages of high forecast timeliness, small data download volume, flexible forecast area, and fast calculation speed.

[0080] At present, the forecast support data and products obtained by maritime official and scientific research ships are all based on the acquired model initial field data through numerical model integration calculation, and then the calculated forecast data or further produced forecast products are transmitted to the ship. The timeliness is relatively lagging, and the forecast area is relatively fixed. The form of forecast products is relatively single and cannot be customized flexibly and independently. The most important thing is that the spatial resolution and accuracy of the forecast data cannot meet the increasingly refined and intelligent needs of forecast support.

[0081] The current application of deep learning networks based on artificial intelligence in marine meteorology has become increasingly mature, and the accuracy of intelligent forecast results is very close to that of numerical model forecasts, and the computing speed is hundreds of times faster than that of numerical models. This application is based on improving the timeliness of forecast products obtained by offshore public scientific research ships and meeting the actual needs of refined security, and can complete the complete forecast business chain of initial data download, forecast calculation, and product production on board the ship. This method makes full use of the neural network model, and can output more accurate forecast wind field data or output forecast wave field data in 3-5 minutes; compared with the current forecast data obtained by integrating the numerical model based on the acquired model initial field data, the operation time is greatly reduced and the operation speed is improved.

[0082] The method provided in the embodiment of the present application, starting from the actual marine security business, can independently customize the area, forecast duration and other elements of the forecast data, and improve the mature marine environment intelligent forecast model to adapt to the needs of marine business applications. It has higher timeliness than traditional numerical forecasting methods, and has a small data download volume, flexible and selectable forecast areas, and fast calculation speed. It can provide ships with timely and accurate refined security data and products.

[0083] In one embodiment, Figure 3 As shown, a lightweight shipborne marine environment forecasting system is provided, including a forecast data cloud service system and a shipborne intelligent forecasting system. The forecast data cloud service system runs on a cloud server deployed on shore, and the shipborne intelligent forecasting system runs on a shipborne terminal deployed on a ship.

[0084] The forecast data cloud service system includes a data acquisition and archiving module, a data request and response module, and a data segmentation and packaging module; the shipborne intelligent forecast system includes a data acquisition subsystem and a wind and wave intelligent forecast subsystem, the data acquisition subsystem includes a data customization request module and a data decoding and transcoding module, and the wind and wave intelligent forecast subsystem includes a wind field intelligent reasoning and calculation module, a sea wave initial field preprocessing module, and a wave field intelligent reasoning and calculation module;

[0085] The data acquisition and archiving module is used to monitor and scan the data servers of mainstream forecasting agencies at home and abroad at regular intervals, obtain the latest initial wind field forecast data from the data servers of mainstream forecasting agencies at home and abroad and archive them; the data request response module is used to receive data request instructions sent by the shipboard terminal; the data segmentation and packaging module is used to segment, compress and package the latest initial wind field forecast data into data packets according to the data request instructions, and send the initial wind field forecast data packets to the shipboard terminal;

[0086] The data customization request module is used to send data request instructions to the cloud server; the data decoding and transcoding module is used to receive the initial wind field forecast data packet sent by the cloud server, decode the initial wind field forecast data packet, and transcode the initial wind field forecast data packet into a preset format to obtain the transcoded initial wind field forecast data; the wind field intelligent reasoning and calculation module is used to input the initial wind field forecast data into a pre-trained wind field reasoning model, and output the optimized target wind field forecast data; the wave initial field preprocessing module is used to use the wind-wave relationship empirical formula to generate the corresponding initial wave field data according to the optimized target wind field forecast data; the wave field intelligent reasoning and calculation module is used to input the initial wave field data into a pre-trained wave field reasoning model, and output the target wave field forecast data.

[0087] Furthermore, the wind and wave intelligent forecasting subsystem also includes a forecast element drawing module, which is used to draw and output wind and wave forecast products in various formats based on target wind field forecast data and target wave field forecast data according to pre-set drawing parameters.

[0088] The specific implementation content of each module can be found in the above definition of a lightweight shipborne marine environment forecasting method, which will not be repeated here.

[0089] In other words, the embodiment of the present application provides a lightweight shipborne marine environment forecasting system for use in offshore security, including a forecasting data cloud service system deployed on shore and a shipborne intelligent forecasting system deployed on official and scientific research vessels. The two systems can interact via satellite communications during marine navigation operations.

[0090] The shore-based forecast data cloud service system includes a data acquisition and archiving module, a data request and response module, and a data segmentation and packaging module. The shipborne intelligent forecast system includes a data acquisition subsystem and a wind and wave intelligent forecast subsystem. The data acquisition subsystem includes a data customization request module and a data decoding and transcoding module. The wind and wave intelligent forecast subsystem includes a wind field intelligent reasoning and calculation module, a wave initial field preprocessing module, a wave field intelligent reasoning and calculation module, and a forecast element drawing module.

[0091] The specific forecasting process of the forecasting system is as follows:

[0092] The data acquisition and archiving module of the shore-based forecasting system regularly monitors and scans the data servers of mainstream forecasting agencies at home and abroad, obtains the latest wind field forecast data from the Internet and archives it in the cloud server as soon as possible; when the data customization request module of the ship-borne intelligent forecasting system sends a data request instruction to the shore-based forecast data cloud service system, the cloud service system will divide and compress the latest forecast field data into a data packet according to the area, elements, forecast duration, spatiotemporal resolution and other instruction information of the data request, and send the data packet to the specified file directory of the ship-borne forecasting system via satellite to complete the data response. Data sending and receiving supports breakpoint resumption to adapt to the special situation where maritime communications are often interrupted.

[0093] Subsequently, the ship-borne data acquisition subsystem will decode and transcode the data packets, and transcode the compressed data into the data format required by the forecasting system; the transcoded wind field data will be imported into the wind and wave intelligent forecasting subsystem to provide initial field data for subsequent intelligent forecasting calculations.

[0094] The wind field intelligent reasoning calculation module contains a global or sea area wind field reasoning environment that has been pre-trained using a residual network model. After importing the transcoded wind field data, it only takes 3-5 minutes to complete the intelligent forecast of the wind field and output more accurate forecast wind field data; the forecast wind field is imported into the wave initial field preprocessing module, which contains an empirical formula for the wind-wave relationship based on statistical fitting of satellite, buoy and other observation data. This formula can be used to generate the initial wave field of the forecast area from the wind field (data including elements such as wave height, wave direction and wave period) without having to transmit this part of the data separately, thereby saving the bandwidth of the shipborne client.

[0095] The wave field intelligent reasoning and calculation module contains a global or sea area wave field reasoning environment that has been trained using a long- and short-term convolutional network model. After reading in the wave field matrix data, the module can output highly accurate wave forecast field data after 3-5 minutes of reasoning and calculation; finally, after the forecast element drawing module reads in the wind field and wave field data output by the intelligent forecast, it will automatically identify the regional scope, spatiotemporal resolution and other elements of the data, and produce wind and wave forecast products according to the drawing parameters set in advance, and output forecast products in various formats such as png and pdf.

[0096] In summary, the most concerned marine environmental elements in ensuring the safe navigation of ships at sea are wind and waves. The embodiment of the present application has developed a lightweight wind and wave intelligent forecasting system suitable for shipboard based on a deep learning network model. The operation of the wind and wave intelligent forecasting system currently used in business scientific research and development requires not only the download of wind field data, but also the download of the initial field data of the waves. The wind and wave intelligent forecasting system of the present invention has developed an initial wave field preprocessing module, which only needs to download the wind field data, and use the wind and wave relationship empirical formula in the module to generate the initial wave field of the forecast area. The wind and wave relationship formula based on the observation data fitting can be calculated separately according to different forecast sea areas, thereby improving the accuracy of the initial wave field.

[0097] The shipborne intelligent forecasting system realizes rapid calculation and forecasting services on high-sea ships through data segmentation and packaging, breakpoint resumption and other technologies. Compared with the use of traditional numerical forecasting methods for forecasting support by high-sea ships, the lightweight shipborne marine environment forecasting system provided in the embodiment of the present application has the advantages of high forecast timeliness, small data download volume, flexible forecast area, and fast calculation speed.

[0098] Table 1 Comparison between shipborne forecasting system and numerical forecasting method

[0099]

[0100] Table 1 compares the Northwest Pacific Ocean wave numerical prediction system currently in operation at the National Marine Environmental Forecast Center. Under the same computing power platform, when the duration, resolution and other conditions of the lightweight shipborne forecast system are set to be consistent with those of the numerical forecast system, it can be seen that the lightweight shipborne forecast system has obvious advantages in forecast timeliness, calculation time, data download volume, etc., and the forecast area and spatiotemporal resolution can be flexibly customized according to the needs of the offshore site, which is not available in the numerical forecast system.

[0101] At present, the forecast support data and products obtained by maritime official and scientific research ships are all based on the acquired model initial field data through numerical model integration calculation, and then the calculated forecast data or further produced forecast products are transmitted to the ship. The timeliness is relatively lagging, and the forecast area is relatively fixed. The form of forecast products is relatively single and cannot be customized flexibly and independently. The most important thing is that the spatial resolution and accuracy of the forecast data cannot meet the increasingly refined and intelligent needs of forecast support.

[0102] At present, the application of deep learning networks based on artificial intelligence in marine meteorology has become increasingly mature, and the accuracy of intelligent forecast results is very close to that of numerical model forecasts, and the computing speed is hundreds of times faster than that of numerical models. Starting from improving the timeliness of forecast products obtained by offshore public service and scientific research ships and meeting the actual needs of refined security, the present invention uses intelligent forecast models to develop a set of lightweight marine environment forecast systems that can be quickly calculated on ordinary desktop computers or portable computers. The system can be deployed on various types of public service, scientific research and military ships performing offshore operations, and can complete the complete forecast business chain of initial data download, forecast calculation, and product production on the ship. The system has the characteristics of small data download volume, flexible and optional forecast areas, and fast computing speed. It can provide timely and accurate refined forecast security data and products for the above-mentioned ships, and significantly improve the on-site forecast security capabilities of my country's offshore ships.

[0103] For the first time in China, a lightweight intelligent marine environment forecasting system has been put into commercial use on offshore vessels. Starting from the actual marine security business, the present invention can independently customize the forecast data area, forecast duration and other elements, and improve the mature marine environment intelligent forecasting model to meet the needs of marine business applications. Compared with the traditional numerical forecasting system, it has higher timeliness, small data download volume, flexible forecast area selection, and fast calculation speed, and can provide ships with timely and accurate refined security data and products.

[0104] In one embodiment, a shipborne terminal is provided, and its internal structure diagram can be as follows: Figure 5 As shown. The shipborne terminal includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the shipborne terminal is used to provide computing and control capabilities, and the communication interface is used to communicate with an external terminal in a wired or wireless manner. The wireless method can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The shipborne terminal loads and runs a computer program to implement the above-mentioned lightweight shipborne marine environment forecasting method. The display screen of the shipborne terminal can be a liquid crystal display screen or an electronic ink display screen, and the input device of the shipborne terminal can be a touch layer covering the display screen, or a button, trackball or touchpad set on the shell of the shipborne terminal, or an external keyboard, touchpad or mouse, etc.

[0105] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the shipborne terminal to which the scheme of the present application is applied. The specific shipborne terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0106] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which involves all or part of the processes in the above-mentioned embodiment method.

[0107] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A lightweight shipborne marine environment forecasting method, characterized in that: Applied to a shipborne terminal, the shipborne terminal is deployed on a public service or scientific research vessel, and the method comprises: Send data request instructions to the cloud server deployed on the shore; After the cloud server divides, compresses, and encapsulates the latest initial wind farm forecast data into data packets according to the data request instruction, receiving the initial wind farm forecast data packet sent by the cloud server; Decoding the initial wind farm forecast data packet, and transcoding the initial wind farm forecast data packet into a preset format to obtain transcoded initial wind farm forecast data; Inputting the initial wind field forecast data into a pre-trained wind field inference model, and outputting optimized target wind field forecast data; Using the empirical formula of wind-wave relationship, the corresponding initial wave field data is generated according to the optimized target wind field forecast data; Inputting the optimized target wind field forecast data and initial wave field data into a pre-trained wave field inference model, and outputting the target wave field forecast data; Based on the target wind field forecast data and the target wave field forecast data, and according to pre-set drawing procedures and parameters, draw and output wind and wave forecast products in various formats; The data request instruction includes regional information, element information, forecast duration information, and spatiotemporal resolution information; The wind field inference model is trained using a residual network model, and the wave field inference model is trained using a long-term and short-term convolutional network model; The training process of the wind field inference model includes: Collect historical wind field observation data and corresponding wind field forecast data for the world or sea area; Performing spatiotemporal fusion processing on the historical wind farm observation data and the corresponding wind farm forecast data to form a first real-time analysis set; Dividing the first actual situation analysis set into a first training set, a first validation set and a first test set according to a preset ratio; The residual network model is trained using the first training set to obtain a wind field inference model; during the training process, the model parameters of the residual network model are continuously adjusted through an optimization algorithm; Using the first validation set to evaluate the wind farm inference model, and optimizing the wind farm inference model according to the evaluation result; The training process of the wave field inference model includes: Collect historical wave field observation data and corresponding wave field forecast data for the world or sea area; Performing spatiotemporal fusion processing on the historical wave field observation data and the corresponding wave field forecast data to form a second real-time analysis set; Dividing the second actual situation analysis set into a second training set, a second verification set, and a second test set according to a preset ratio; The long- and short-term convolutional network model is trained using the second training set to obtain a wave field inference model; during the training process, the model parameters of the long- and short-term convolutional network model are continuously adjusted through an optimization algorithm; The wave field reasoning model is evaluated using the second validation set, and the wave field reasoning model is tuned according to the evaluation result.

2. The lightweight shipborne marine environment forecasting method according to claim 1 is characterized in that: The wind-wave relationship empirical formula is obtained by fitting wind field observation data and sea wave field observation data.

3. A lightweight shipborne marine environment forecasting system, characterized in that: It includes a forecast data cloud service system and a ship-borne intelligent forecast system. The forecast data cloud service system runs on a cloud server deployed on shore, and the ship-borne intelligent forecast system runs on a ship-borne terminal deployed on a public service or scientific research ship. The forecast data cloud service system includes a data acquisition and archiving module, a data request and response module, and a data segmentation and packaging module; the shipborne intelligent forecast system includes a data acquisition subsystem and a wind and wave intelligent forecast subsystem, the data acquisition subsystem includes a data customization request module and a data decoding and transcoding module, and the wind and wave intelligent forecast subsystem includes a wind field intelligent reasoning and calculation module, a sea wave initial field preprocessing module, a wave field intelligent reasoning and calculation module, and a forecast element drawing module; The data acquisition and archiving module is used to regularly monitor and scan the data servers of mainstream forecasting agencies at home and abroad, and obtain the latest initial wind field forecast data from the data servers of mainstream forecasting agencies at home and abroad and archive them; The data request response module is used to receive the data request instruction sent by the shipboard terminal; the data segmentation and packaging module is used to segment, compress and package the latest initial wind field forecast data into data packets according to the data request instruction, and send the initial wind field forecast data packets to the shipboard terminal; The data customization request module is used to send a data request instruction to the cloud server; the data decoding and transcoding module is used to receive the initial wind field forecast data packet sent by the cloud server, decode the initial wind field forecast data packet, and transcode the initial wind field forecast data packet into a preset format to obtain the transcoded initial wind field forecast data; the wind field intelligent reasoning and calculation module is used to input the initial wind field forecast data into a pre-trained wind field reasoning model, and output the optimized target wind field forecast data; the wave initial field preprocessing module is used to generate the corresponding initial wave field data according to the optimized target wind field forecast data by using the wind-wave relationship empirical formula; the wave field intelligent reasoning and calculation module is used to input the initial wave field data into a pre-trained wave field reasoning model, and output the target wave field forecast data; The forecast element drawing module is used to draw and output the optimized target wind field and target wave forecast data into wind and wave element forecast products in different formats; The data request instruction includes regional information, element information, forecast duration information, and spatiotemporal resolution information; The wind field inference model is trained using a residual network model, and the wave field inference model is trained using a long-term and short-term convolutional network model; The training process of the wind field inference model includes: Collect historical wind field observation data and corresponding wind field forecast data for the world or sea area; Performing spatiotemporal fusion processing on the historical wind farm observation data and the corresponding wind farm forecast data to form a first real-time analysis set; Dividing the first actual situation analysis set into a first training set, a first validation set and a first test set according to a preset ratio; The residual network model is trained using the first training set to obtain a wind field inference model; during the training process, the model parameters of the residual network model are continuously adjusted through an optimization algorithm; Using the first validation set to evaluate the wind farm inference model, and optimizing the wind farm inference model according to the evaluation result; The training process of the wave field inference model includes: Collect historical wave field observation data and corresponding wave field forecast data for the world or sea area; Performing spatiotemporal fusion processing on the historical wave field observation data and the corresponding wave field forecast data to form a second real-time analysis set; Dividing the second actual situation analysis set into a second training set, a second verification set, and a second test set according to a preset ratio; The long- and short-term convolutional network model is trained using the second training set to obtain a wave field inference model; during the training process, the model parameters of the long- and short-term convolutional network model are continuously adjusted through an optimization algorithm; The wave field reasoning model is evaluated using the second validation set, and the wave field reasoning model is tuned according to the evaluation result.

4. A shipboard terminal, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 2 are implemented.

Images