Computer program, sea condition predicting device, and sea condition prediction method

Abstract

[Problem] To provide a computer program, a sea condition predicting device, and a sea condition prediction method that implement prediction reflecting actual observation data while reducing a calculation load. [Solution] This computer program causes a computer to execute processing for: acquiring observation data at an observation position from a device that makes observations at the surface of the sea or underwater; acquiring predicted sea condition data obtained by predicting the sea conditions at the time of observation at a plurality of positions included in a sea area including the observation position; and outputting predicted sea condition data a predetermined period of time after the observation time point of the observation data corresponding to the observation data acquired from the device, by using a learning model that has been trained to output second predicted sea condition data at a second time point a predetermined period of time after a first time point, upon receiving input of the observation data acquired from the device at the first time point and first predicted sea condition data obtained by predicting the sea condition at the first time point.

Description

Computer program, sea condition prediction device, and sea condition prediction method

[0001] The present invention relates to a computer program for predicting sea conditions, a sea condition prediction device, and a sea condition prediction method.

[0002] Conventionally, a method for predicting sea conditions in a given sea area involves numerical forecasting, in which a physical simulation of an ocean model is performed reflecting actually measured sea state data, and sea state data is predicted a given time (Δt) later. Patent Document 1 discloses that calculation processing is performed using an on-board forecasting device to display high-resolution sea conditions. The numerical forecasting disclosed in Patent Document 1 divides the ocean into three-dimensional meshes, mathematically predicts sea state data at each mesh position (node), and converts the values ​​to match the actual measured values ​​at the observation point, thereby creating forecast data.

[0003] Statistical forecasting techniques are also used to predict ocean conditions. In statistical forecasting, an AI model is trained on analytical data obtained by analyzing ocean or submarine ocean conditions using past measured data, and the learning process is carried out so as to reproduce ocean condition data from ocean condition data from a predetermined time (Δt) before (Patent Document 2, etc.).

[0004] International Publication No. 2022 / 230333 Patent No. 7175415

[0005] Numerical forecasting requires large amounts of simulation data and takes time to calculate. Statistical forecasting using AI models makes it difficult to reflect observational information from the target ocean area. Furthermore, neither numerical forecasting nor statistical forecasting methods take into consideration accurately reflecting actual ocean state data observed at observation points that are different from the mesh location of the target area.

[0006] The present disclosure aims to provide a computer program, a sea condition prediction device, and a sea condition prediction method that realize predictions that reflect actual observation data while reducing the computational load.

[0007] A computer program according to one aspect of the present disclosure causes a computer to acquire observation data at an observation location from a device that observes the sea or underwater, acquire predicted sea condition data that predicts the sea conditions at the time of observation at multiple locations included in an ocean area that includes the observation location, and use a learning model that is trained to output second predicted sea condition data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea condition data that predicts the sea conditions at the first time point are input, and to execute a process of outputting predicted sea condition data for a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device.

[0008] In one aspect, a trained model is used that is trained to output predicted sea state data for a specified time period when observation data obtained from a device observing the sea or underwater and predicted sea state data are input. Statistical methods using models such as neural networks and multiple regression analysis have not taken into account the reflection of observation data. By using a model that applies observation data from the sea area including the observation location, the processing load is reduced compared to simulations, which require massive data and many computing resources, and sea state predictions that reflect observation data are possible.

[0009] In one aspect of the computer program of the present disclosure, the learning model is a graph neural network that inputs and outputs a graph described by edge data indicating the relationships between nodes and feature values ​​of each node, and each node of the graph input and output to the graph neural network corresponds to a position on or underwater, and multiple types of ocean condition data at the corresponding positions are associated with the nodes as feature values.

[0010] In one aspect, a graph neural network (GNN) is used as a learning model. CNNs (Convolutional Neural Networks) used for data learning are effective for image data composed of pixels arranged in a matrix. The observation positions of devices that observe the sea surface or underwater are arbitrary positions and do not match the grid points or mesh positions that are the target of the simulation. Therefore, since the observation positions deviate from regularity, adopting a GNN, which does not require regularity to be taken into account, rather than using a CNN, enables more accurate sea state predictions.

[0011] A computer program according to one aspect of the present disclosure causes the computer to input observation data at an observation position, sea condition data at multiple positions within the sea area, and edge data indicating the relationship between the observation position and nodes corresponding to the multiple positions into the learning model, and to execute a process of obtaining predicted sea condition data for the position on or underwater at sea after the specified time from the learning model.

[0012] In one aspect, edge data indicating the relationship between nodes indicates that the nodes are adjacent. By expressing the relationship between the grid points or mesh positions of the ocean condition prediction model and the observation position as edge data, predictions can be made using the GNN taking into account the relationship between positions.

[0013] A computer program of one aspect of the present disclosure causes the computer to perform a process of adding an observation node corresponding to the observation position of the observation data and observation data associated with the observation node to a graph corresponding to the first predicted sea condition data at the first time point, and adding edge data indicating the positional relationship between the observation node and the node of the first predicted sea condition data to the graph.

[0014] In one aspect, a graph is created by adding observation data from an arbitrary observation location to predicted sea state data that is predicted based on a regular grid or mesh, and the graph is then input into a learning model that is a GNN. This enables predictions using a learning model that reflects the observation data in the predicted sea state data.

[0015] In one aspect of the computer program of the present disclosure, the first predicted sea state data includes sea state data at a plurality of positions regularly arranged corresponding to a mesh that regularly divides the sea area, the observation position is located between the plurality of positions, and the edge data includes data on the distance between the positions.

[0016] In one aspect, by including data on the distance between the observation location and the grid points or mesh positions of the predicted sea condition model in the edge data input to the GNN, it becomes possible to realize predictions by the GNN that more accurately reflect the observation location.

[0017] In a computer program according to one aspect of the present disclosure, the type of data included in the observation data is different from the type of data included in the first predicted sea state data.

[0018] In one aspect, the data type of the observation data does not have to match the data type of the first predicted oceanographic state data. The observation data may only include water temperature and current speed, and the first predicted oceanographic state data may also include water temperature, current direction, and salinity. Even if different types of data than the first predicted oceanographic state data are used as observation data, the observation data can still be reflected in predictions based on the learning model.

[0019] In one aspect of the computer program of the present disclosure, the observation data, the first predicted sea condition data, and the second predicted sea condition data each include at least one of water temperature, current speed, current direction, salinity concentration, depth, and sea surface altitude.

[0020] In one aspect, the observation data, the first predicted oceanographic state data, and the second predicted oceanographic state data include at least one of water temperature, current speed, current direction, salinity, depth, and sea surface height. By using these data, it is possible to predict the water temperature distribution, current speed distribution, salinity distribution, depth distribution, or sea surface height distribution in the target sea area.

[0021] A computer program according to one aspect of the present disclosure causes the computer to acquire the observation position and the observation data from an observation device installed on a ship, a marine drone, an underwater drone, a buoy, a satellite, or a marine organism, and to execute a process of outputting the second predicted sea condition data for the sea area including the observation position using the acquired observation data and the learning model.

[0022] In one aspect, sea conditions can be predicted from observation data obtained by observation devices attached to movable objects such as ships, surface drones, underwater drones, buoys, satellites, or marine organisms.

[0023] In one aspect of the computer program of the present disclosure, the learning model is trained using training data generated by a simulator that calculates, by simulation, second predicted sea state data at a second time point a predetermined time after the first time point based on first predicted sea state data at a first time point and observation data obtained from the device at the first time point.

[0024] In one aspect, the learning model is trained using predicted sea state data that is sequentially output by an ocean model simulator and whose accuracy has been evaluated. The learning model is trained to reproduce the predictions of a proven simulator while reflecting the observed data.

[0025] A sea condition prediction device of one aspect of the present disclosure acquires observation data at an observation position from a device that observes the sea or underwater, acquires predicted sea condition data that predicts the sea conditions at the time of observation at multiple positions included in an ocean area that includes the observation position, and uses a learning model that is trained to output second predicted sea condition data at a second point in time that is a predetermined time after the first point in time when the observation data acquired from the device at a first point in time and first predicted sea condition data that predicts the sea conditions at the first point in time are input, and includes a processing unit that outputs predicted sea condition data a predetermined time after the observation point of the observation data that corresponds to the observation data acquired from the device.

[0026] A sea condition prediction method according to one aspect of the present disclosure involves a computer acquiring observation data at an observation location from a device that observes the sea or underwater, acquiring predicted sea condition data that predicts the sea conditions at the time of observation at multiple locations included in an ocean area that includes the observation location, and using a learning model that is trained to output second predicted sea condition data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea condition data that predicts the sea conditions at the first time point are input, and outputting predicted sea condition data for a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device.

[0027] 1 is a block diagram showing the configuration of a sea condition prediction device of the present disclosure; FIG. 2 is an explanatory diagram of predicted sea condition data; FIG. 3 is a schematic diagram of a graph of predicted sea condition data; FIG. 4 is a schematic diagram of a graph to which observation data has been added; FIG. 5 is a schematic diagram of a learning model; FIG. 6 is a flowchart showing an example of a prediction processing procedure of a first embodiment; FIG. 7 is a block diagram showing the configuration of a server device; FIG. 8 is a flowchart showing an example of a processing procedure by a sea condition prediction device and a server device in a second embodiment;

[0028] The present disclosure will be specifically described with reference to the drawings showing embodiments thereof.

[0029] [First embodiment] Fig. 1 is a block diagram showing the configuration of a sea condition prediction device 1 according to the present disclosure. The sea condition prediction device 1 is a device that predicts sea conditions in an area of ​​the sea through which a mobile body, such as a ship, moves, based on observation data obtained from a device 2 provided on the mobile body that moves on or underwater. The mobile body is primarily a ship that navigates based on predicted sea conditions. The mobile body may be a surface drone or an underwater drone. The mobile body may be a buoy floating on the sea, a satellite, or a marine organism.

[0030] The ocean condition prediction device 1 can connect to a device 2, such as a sensor, installed on a moving body and acquire observation data. The ocean condition prediction device 1 is a device that predicts ocean conditions after a predetermined time (Δt) using observation data obtained from the device 2. The device 2 is an instrument that measures temperature, current speed, current direction, salinity concentration, depth, and sea surface altitude. The device 2 includes a GPS receiver. The device 2 may be a camera that photographs the sea surface, a radar, sonar, a speedometer, a fish finder, etc., installed on the moving body. The device 2 may also include sensors that monitor the rudder angle of the ship's steering gear, the thrust of the thrust generating device, and the output from the power source.

[0031] The ocean condition prediction device 1 communicates via a network N including communication media such as satellite communication and a carrier network, and is capable of communicating with a server device or the like of a service that provides weather-related data.

[0032] The sea condition prediction device 1 is connected to a monitor device 3 or has a built-in monitor device 3, and can output current sea conditions and predicted sea conditions in color on a screen. The monitor device 3 uses a liquid crystal display, an organic EL (Electro Luminescence) display, or the like. The monitor device 3 may be equipped with a touch panel with a built-in display or physical buttons to accept switching of observation data and display range.

[0033] The sea condition prediction device 1 comprises a processing unit 10, a memory unit 11, an input / output unit 12, and a communication unit 13. The processing unit 10 includes one or more arithmetic processing devices such as a central processing unit (CPU), a micro-processing unit (MPU), a graphics processing unit (GPU), etc. The processing unit 10 also includes a temporary storage medium such as a static random access memory (SRAM) or a dynamic random access memory (DRAM). The processing unit 10 reads a computer program P1 stored in the memory unit 11 into the temporary storage medium and executes it, thereby causing a general-purpose computer to perform various processes described below.

[0034] The memory unit 11 is a relatively large-capacity non-volatile memory area such as a solid state drive (SSD) or a hard disk drive. The memory unit 11 stores a program (program product) required for the processing unit 10 to execute processing, and reference setting data. The program product includes a computer program P1, screen setting data to be displayed on the monitor device 3, and graph description data. The memory unit 11 stores a learning model M1. The memory unit 11 stores observation data obtained from the device 2.

[0035] One of the computer program (computer product) P1 and the learning model M1 stored in the memory unit 11 may be a computer program P9 or a learning model M9 stored in a computer-readable non-transitory storage medium 9 that is read by the processing unit 10 and stored in the memory unit 11. One of the computer program P1 and the learning model M1 may be downloaded by the processing unit 10 from a server device or another download server via the communication unit 13 and stored in the memory unit 11.

[0036] The input / output unit 12 is an interface with other devices. The input / output unit 12 includes, for example, ports compatible with serial communication standards such as RS485 and USB (Universal Serial Bus) and a connection terminal for a monitor device such as HDMI (High-Definition Multimedia Interface). The device 2 and the monitor device 3 are connected to the input / output unit 12. The processing unit 10 acquires data output from the device 2 via the input / output unit 12. The processing unit 10 can output screens showing text and graphs indicating numerical values ​​representing calculation results from the input / output unit 12 to the monitor device 3. The processing unit 10 can accept operation signals from an operation interface provided on the monitor device 3 or the sea condition prediction device 1 itself, and can accept instructions to start measurement or switch the output content to the monitor device 3. The processing unit 10 may output calculation results to an external device from the input / output unit 12.

[0037] The communication unit 13 is a communication module that realizes communication. The communication unit 13 is, for example, a network card. The communication unit 13 may be a communication module for satellite communication. The communication unit 13 may be a communication module that communicates with another server device via a land-based device using an AIS (Automatic Identification System) dedicated frequency. The processing unit 10 can obtain sea condition data for a specified date and time within a specified range from an external service that provides data obtained from satellites via the communication unit 13.

[0038] The sea condition prediction device 1 configured in this manner uses the learning model M1 stored in the memory unit 11 to predict and output future sea condition data based on the observation data obtained from the device 2 and the sea condition data predicted for the sea area around the observation position through analysis.

[0039] The processing unit 10 acquires observation data from the device 2. As described above, the observation data is at least one of temperature, current speed, current direction, salinity concentration, depth, and sea surface altitude. The processing unit 10 acquires the observation position from a GPS receiver that is one of the devices 2 or from a built-in GPS receiver.

[0040] The processing unit 10 can acquire predicted sea state data predicted for a predetermined range of sea area including the observation position of the observation data via the communication unit 13. For example, the processing unit 10 acquires the data from an ocean model server via the communication unit 13. The ocean model server is provided by an organization such as the Japan Meteorological Agency. The processing unit 10 may also acquire sea state data obtained from a meteorological satellite via the communication unit 13.

[0041] FIG. 2 is an explanatory diagram of predicted ocean state data. The predicted ocean state data is obtained by dividing an ocean model into meshes, and includes ocean state parameters such as temperature, current speed, current direction, and salinity concentration at corresponding points, such as the center of each mesh. The predicted ocean state data is generated by the ocean model server through geohydrodynamic simulations over multiple time periods, and is sequentially accumulated. The ocean model server sequentially simulates the ocean state parameters for each mesh at predetermined intervals and stores them in a database. For example, FIG. 2 shows that a specific sea area near Japan is divided into regular hexagonal meshes, and the ocean state parameters are predicted at a reference position in the center of each mesh.

[0042] The processing unit 10 acquires predicted sea state data such as that shown in Fig. 2 as data expressed in a graph described by node definitions, edge data showing the relationships between the nodes, and feature quantities of each node. Fig. 3 is a schematic diagram of a graph of predicted sea state data. The graph in Fig. 3 corresponds to the positional relationship of the predicted sea state data in Fig. 2. In Fig. 3, white circles indicate nodes, and directional arrows indicate the relationships between the nodes.

[0043] As shown in Figure 3, the nodes of the graph obtained from the ocean model server are regularly arranged corresponding to the central position of the mesh on the sea or underwater. Multiple types of ocean condition data at the corresponding position are associated with each node as features. The edge data is data described in a matrix that shows the relationship between nodes, as shown in the lower part of Figure 3. It is also possible to describe the distance between nodes using the values ​​of the matrix elements. The processing unit 10 represents the predicted ocean condition data for the sea area including the observation position of the observation data in a graph, and predicts the ocean condition data for a predetermined time period later using a graph neural network (GNN) using this graph.

[0044] The processing unit 10 adds observation nodes corresponding to the observation positions of the observation data and the observation data associated with the observation nodes to the predicted ocean state data shown in Figure 3 to create a graph to be input into the learning model M1. Figure 4 is a schematic diagram of the graph to which the observation data has been added. As shown in Figure 4, the observation positions of the observation data are between the meshes predicted by the ocean model server. Data indicating the relationship between the observation positions and the nodes corresponding to the multiple positions is added to the edge data. As shown in Figure 4, the nodes of the graph to which the observation data has been added are irregularly arranged. In order to handle a graph with nodes arranged irregularly in this way, a GNN is adopted as the learning model M1.

[0045] FIG. 5 is a schematic diagram of the learning model M1. As described above, the learning model M1 employs a GNN. The learning model M1 has an input layer M11, an output layer M12, and an intermediate layer M13. The input layer M11 accepts input of graph description data (node ​​definitions, edge data, and feature values ​​corresponding to the nodes). The output layer M12 outputs the graph description data. The intermediate layer M13 includes a convolutional layer that convolves the feature values ​​of surrounding nodes, a pooling layer that aggregates data, an activation function that extracts feature values ​​from input values, and the like.

[0046] The input layer M11 of the learning model M1 receives a graph created from the observation data shown in Figure 4 and the predicted sea state data obtained from the ocean model server. The output layer M12 of the learning model M1 outputs the predicted sea state data shown in Figure 3.

[0047] The intermediate layer M13 of the learning model M1 is trained using the actual results of predicted sea state data sequentially predicted with high accuracy by the ocean model server. The graph shown in Figure 4, generated based on the observation data obtained by the device 2 and the first predicted sea state data predicted by the ocean model server for the observation time (first time point) when the observation data was observed, is used as input data for the training data. The graph of the second predicted sea state data predicted by the ocean model server for a time point (second time point) a predetermined time after the observation time is used as output data for the training data. Here, the predicted sea state data for the first time point and the predicted sea state data for the second time point are both past actual data. The intermediate layer M13 progresses learning by updating parameters based on the difference between the graph of predicted sea state data output from the output layer M12 when input data of the training data is input to the input layer M11 and the output data in the training data corresponding to the input data, or the result of calculating the evaluation function. The learning process may be performed by a pre-modeling device or by the processing unit 10 before shipping the ocean condition prediction device 1.

[0048] 5 and a CNN that derives features by processing images included in the observation data obtained by the device 2. The learning model M1 may be trained to output image data showing the temperature distribution in the ocean area or image data showing the current velocity distribution in the ocean area.

[0049] The sea state prediction using the trained learning model M1 thus trained will be described with reference to a flowchart. Fig. 6 is a flowchart showing an example of a prediction processing procedure according to the first embodiment.

[0050] The processing unit 10 of the ocean condition prediction device 1 acquires observation data from the device 2 a predetermined time before the time to be predicted (step S101). The processing unit 10 determines the observation position and observation time by the device 2 (step S102). In step S102, the processing unit 10 may acquire the position data from a GPS receiver that is one of the devices 2, or may acquire the position data from another device that monitors the position of the device 2.

[0051] The processing unit 10 determines the sea area including the determined observation position (step S103), and acquires predicted sea condition data predicting the sea conditions at the time of observation in the determined sea area as first predicted sea condition data via the communication unit 13 (step S104).

[0052] The processing unit 10 creates a graph to be input into the learning model M1 based on the observation data acquired in step S101 and the first predicted sea state data acquired in step S104 (step S105).

[0053] The processing unit 10 inputs the graph created in step S105 into the trained learning model M1 (step S106). The processing unit 10 acquires second predicted sea state data for the time point of prediction described in the graph output from the learning model M1 (step S107).

[0054] The processing unit 10 generates a screen including an image showing the sea conditions in the sea area determined in step S103 from the acquired second predicted sea state data (step S108). The processing unit 10 outputs the generated screen to the monitor device 3 (step S109), and ends the processing.

[0055] As described above, by adopting the learning model M1 using a GNN, it is possible to reflect the values ​​of observation data in a model using a neural network that reproduces a reliable numerical forecast simulation. Numerical forecast simulations require a large amount of computational resources, but by using the learning model M1, it is possible to relatively conserve computational resources. As a result, by using a GNN instead of a CNN, which is a typical example of a learning model using a predictive neural network on ships that require compactness, such as small boats and drones, it is possible to reflect observation data from arbitrary positions that are not regularly located in the predictions.

[0056] [Second embodiment] In a second embodiment, the sea state prediction device 1 executes part of its calculations on a server device 4 installed on land. The server device 4 is installed on land and has more abundant calculation resources than the sea state prediction device 1 installed on a ship.

[0057] 7 is a block diagram showing the configuration of the server device 4. The server device 4 includes a processing unit 40, a storage unit 41, and a communication unit 42. The server device 4 may be configured as a single server computer, or may be configured to distribute processing among multiple server computers. The processing unit 40 includes one or more arithmetic processing units such as a CPU, an MPU, a GPU, etc. The processing unit 40 includes a temporary storage medium such as an SRAM or a DRAM.

[0058] The memory unit 41 is a relatively large-capacity non-volatile memory area such as an SSD or a hard disk. The memory unit 41 stores a computer program P4. The memory unit 41 stores a learning model M1. In the second embodiment, the learning model M1 does not need to be stored in the memory unit 11 of the sea condition prediction device 1. The processing unit 40 executes the processing procedure shown in FIG. 8 based on the computer program P4.

[0059] The communication unit 42 realizes communication with the sea condition prediction device 1 and the ocean model server via the network N. Specifically, the communication unit 42 is a network card. The communication unit 42 may be a wireless communication device that connects to a carrier network, or may be a wireless communication device for Wi-Fi. The processing unit 40 can send and receive data between the sea condition prediction device 1 and the ocean model server via the communication unit 42.

[0060] The server device 4 receives data from the sea condition prediction device 1 and executes part of the processing using the learning model M1 performed by the sea condition prediction device 1.

[0061] Figure 8 is a flowchart showing an example of a processing procedure by the sea condition prediction device 1 and the server device 4 in the second embodiment. Of the processing procedures of the sea condition prediction device 1 shown in Figure 8, steps that are common to the processing procedures shown in Figure 6 of the first embodiment are assigned the same step numbers and detailed explanations will be omitted.

[0062] The processing unit 10 of the ocean condition prediction device 1 acquires observation data from the device 2 a predetermined time before the prediction target time (S101) and determines the observation position and observation time by the device 2 (S102). The processing unit 10 transmits the observation position and observation data from the communication unit 13 to the server device 4 (step S123).

[0063] The processing unit 40 of the server device 4 receives the observation position and the observation data (step S401), and determines the sea area including the observation position (step S402). The processing unit 40 acquires predicted sea state data, which predicts the sea state at the time of observation in the determined sea area, from the ocean model server via the communication unit 42 as first predicted sea state data (step S403).

[0064] The processing unit 40 creates a graph to be input into the learning model M1 based on the observation data received in step S401 and the first predicted sea state data acquired in step S403 (step S404).

[0065] The processing unit 40 inputs the graph created in step S404 into the trained learning model M1 (step S405). The processing unit 40 acquires second predicted sea state data for the time point of prediction described in the graph output from the learning model M1 (step S406).

[0066] The processing unit 40 creates a screen including an image showing the sea conditions in the sea area determined in step S402 from the acquired second predicted sea condition data (step S407). The processing unit 40 transmits the created screen data to the sea condition prediction device 1 (step S408).

[0067] The processing unit 10 of the sea condition prediction device 1 receives screen data including an image showing sea conditions from the server device 4 (step S124), and outputs a screen based on the received data to the monitor device 3 (step S125), thereby ending the process.

[0068] The embodiments disclosed above are illustrative in all respects and are not limiting. The scope of the present invention is defined by the claims, and includes all modifications within the meaning and scope of the claims. Forms obtained by appropriately combining the technical means disclosed in each embodiment are also included in the technical scope of the present invention.

[0069] Furthermore, independent claims and dependent claims described in the claims can be combined with each other in any and all combinations, regardless of the reference format. Furthermore, although the claims use a format in which a claim references two or more other claims (multiple claim format), this is not limited to this format. Multiple claims that reference at least one other multiple claim (multi-multi claim format) may also be used.

[0070] The following additional notes are provided regarding the above-described embodiment.

[0071] (Supplementary Note 1) A computer program that causes a computer to execute the following processes: acquire observation data at an observation position from a device that observes the sea or underwater; acquire predicted sea state data that predicts sea conditions at the time of observation at multiple positions included in an ocean area that includes the observation position; use a learning model that is trained to output second predicted sea state data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea state data that predicts the sea conditions at the first time point are input; and output predicted sea state data a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device.

[0072] (Supplementary Note 2) The computer program according to Supplementary Note 1, wherein the learning model is a graph neural network that inputs and outputs a graph described by edge data indicating relationships between nodes and feature values ​​of each node, and each node of the graph input and output to the graph neural network corresponds to a position on or underwater, and multiple types of ocean condition data at the corresponding positions are associated with the nodes as feature values.

[0073] (Appendix 3) The computer program described in Appendix 2 causes the computer to execute a process of inputting observation data at an observation position, sea state data at multiple positions included in the sea area, and edge data indicating the relationship between the observation position and nodes corresponding to the multiple positions into the learning model, and obtaining predicted sea state data for the position on or underwater at the specified time after from the learning model.

[0074] (Appendix 4) A computer program as described in Appendix 2 or 3, which causes the computer to execute the following process: add an observation node corresponding to the observation position of the observation data and the observation data associated with the observation node to a graph corresponding to the first predicted sea state data at the first time point; and add edge data indicating the positional relationship between the observation node and the node of the first predicted sea state data to the graph.

[0075] (Appendix 5) The computer program described in Appendix 3 or Appendix 4, wherein the first predicted sea state data includes sea state data at a plurality of positions regularly arranged corresponding to meshes that regularly divide the sea area, the observation position is between the plurality of positions, and the edge data includes data on the distance between the positions.

[0076] (Supplementary Note 6) The computer program according to any one of Supplementary Notes 1 to 5, wherein the type of data included in the observation data is different from the type of data included in the first predicted sea state data.

[0077] (Appendix 7) The computer program according to any one of appendices 1 to 6, wherein the observation data, the first predicted sea state data, and the second predicted sea state data each include at least one of water temperature, current speed, current direction, salinity concentration, depth, and sea surface altitude.

[0078] (Appendix 8) The computer program described in any one of Appendices 1 to 7, which causes the computer to execute a process of acquiring the observation position and the observation data from an observation device attached to a ship, a marine drone, an underwater drone, a buoy, a satellite, or a marine organism, and outputting the second predicted sea condition data for the sea area including the observation position using the acquired observation data and the learning model.

[0079] (Supplementary Note 9) The computer program described in any one of Supplementary Notes 1 to 8, wherein the learning model is trained using training data generated by a simulator that calculates, by simulation, second predicted sea state data at a second time point that is a predetermined time after the first time point, based on first predicted sea state data at a first time point and observation data acquired from the device at the first time point.

[0080] (Supplementary Note 10) A sea condition prediction device comprising a processing unit that acquires observation data at an observation position from a device that observes the sea or underwater, acquires predicted sea condition data that predicts the sea conditions at the time of observation at a plurality of positions included in an ocean area that includes the observation position, uses a learning model that is trained to output second predicted sea condition data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea condition data that predicts the sea conditions at the first time point are input, and outputs predicted sea condition data for a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device.

[0081] (Appendix 11) A sea condition prediction method in which a computer acquires observation data at an observation position from a device that observes the sea or underwater, acquires predicted sea condition data that predicts the sea conditions at the time of observation at a plurality of positions included in an ocean area that includes the observation position, and uses a learning model that is trained to output second predicted sea condition data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea condition data that predicts the sea conditions at the first time point are input, and outputs predicted sea condition data for a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device. term

[0082] Not necessarily all objects or advantages may be achieved in accordance with any particular embodiment described herein. Thus, for example, one skilled in the art will appreciate that a particular embodiment may be configured to operate to achieve or optimize one or more advantages as taught herein without necessarily achieving other objects or advantages as taught or suggested herein.

[0083] All processes described herein may be embodied and fully automated by software code modules executed by a computing system including one or more computers or processors. The code modules may be stored on any type of non-transitory computer-readable medium or other computer storage device. Some or all of the methods may be embodied in dedicated computer hardware.

[0084] Many other variations beyond those described herein will be apparent from this disclosure. For example, depending on the embodiment, certain operations, events, or functions of any of the algorithms described herein may be performed in a different sequence, added, merged, or omitted entirely (e.g., not all described acts or events are necessary to execute an algorithm). Furthermore, in certain embodiments, operations or events may be performed in parallel rather than sequentially, e.g., via multithreading, interrupt processing, or multiple processors or processor cores, or on other parallel architectures. Furthermore, different tasks or processes may be performed by different machines and / or computing systems that may function together.

[0085] The various illustrative logical blocks and modules described in connection with the embodiments disclosed herein may be implemented or executed by a machine such as a processor. The processor may be a microprocessor, but alternatively, the processor may be a controller, microcontroller, or state machine, or a combination thereof. The processor may include electrical circuitry configured to process computer-executable instructions. In another embodiment, the processor includes an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable device that performs logical operations without processing computer-executable instructions. A processor may also be implemented as a combination of computing devices, such as a combination of a digital signal processor (DSP) and a microprocessor, multiple microprocessors, one or more microprocessors in combination with a DSP core, or any other such configuration. Although described herein primarily with reference to digital technology, a processor may also include primarily analog elements. For example, some or all of the signal processing algorithms described herein may be implemented by analog circuitry or mixed analog and digital circuitry. The computing environment can include any type of computer system, including, but not limited to, a microprocessor, mainframe computer, digital signal processor, portable computing device, device controller, or computer system based on a computational engine within an appliance.

[0086] Unless otherwise specified, conditional language such as "can," "could," "would," or "potential" is understood within the context in which it is generally used to convey that certain embodiments include certain features, elements, and / or steps, while other embodiments do not. Thus, such conditional language does not generally imply that features, elements, and / or steps are required in any manner in one or more embodiments, or that one or more embodiments necessarily include logic for determining whether those features, elements, and / or steps are included in or performed in any particular embodiment.

[0087] Disjunctive language such as "at least one of X, Y, Z," unless specifically stated otherwise, is understood in its general context to indicate that an item, term, etc. can be either X, Y, Z, or any combination thereof (e.g., X, Y, Z). Thus, such disjunctive language does not generally imply that a particular embodiment requires at least one of X, at least one of Y, or at least one of Z, respectively, to be present.

[0088] Any process descriptions, elements, or blocks in the flow diagrams described herein and / or illustrated in the accompanying drawings should be understood as potentially representing modules, segments, or portions of code, comprising one or more executable instructions for implementing a particular logical function or element in the process. Alternative embodiments are included within the scope of the embodiments described herein, in which elements or functions may be performed out of order, substantially simultaneously, or in reverse order from that shown or described, depending on the functionality involved, as will be understood by those skilled in the art.

[0089] Unless otherwise expressly stated, numeral terms such as "one" should generally be construed to include one or more described items. Thus, phrases such as "one device configured to" are intended to include one or more listed devices. Such one or more listed devices may also be collectively configured to perform the recited reference. For example, "a processor configured to perform the following A, B, and C" may include a first processor configured to perform A and a second processor configured to perform B and C. Additionally, even if a specific number of enumerations of the introduced embodiments are explicitly recited, those skilled in the art should construe such enumerations to typically mean at least the recited number (e.g., the mere enumeration of "two enumerations" without other modifiers typically means at least two enumerations, or two or more enumerations).

[0090] In general, it will be appreciated by those skilled in the art that the terms used herein generally intend "non-limiting" terms (e.g., the term "including" should be interpreted as "including but not limited to at least," the term "having" should be interpreted as "having at least," the term "including" should be interpreted as "including, but not limited to," etc.).

[0091] For purposes of description, the term "horizontal" as used herein is defined as a plane parallel to the plane or surface of the floor of the area in which the described system is used or the plane in which the described method is performed, regardless of its orientation. The term "floor" can be interchanged with the terms "ground" or "water surface." The term "vertical / plumb" refers to a direction perpendicular / vertical to a defined horizontal line. Terms such as "upper," "lower," "below," "top," "side," "higher," "lower," "above," "over," "below," etc. are defined relative to the horizontal plane.

[0092] As used herein, the terms "attach," "connect," "mate," and other related terms, unless otherwise noted, should be interpreted to include detachable, movable, fixed, adjustable, and / or removable connections or couplings. Connections / couplings include direct connections and / or connections with intermediate structures between the two components described.

[0093] Unless otherwise expressly stated, as used herein, numbers preceded by terms such as "approximately," "about," and "substantially" are inclusive of the recited number and also refer to an amount close to the recited amount that performs the desired function or achieves the desired result. For example, "approximately," "about," and "substantially" refer to values ​​less than 10% of the recited numerical value, unless otherwise expressly stated. As used herein, features of the disclosed embodiments preceded by terms such as "approximately," "about," and "substantially" refer to features that have some variability that also perform the desired function or achieve the desired result for that feature.

[0094] Many variations and modifications may be made to the above-described embodiments, and these elements should be understood to be among other acceptable examples. All such modifications and variations are intended to be included within the scope of this disclosure and are protected by the following claims.

[0095] 1 Ocean condition prediction device 10 Processing unit 11 Memory unit P1 Computer program M1 Learning model 2 Device

Claims

1. A computer program that causes a computer to execute the following process: acquire observation data at an observation location from a device that observes the sea or underwater; acquire predicted sea state data that predicts the sea state at the time of observation at multiple locations included in the sea area that includes the observation location; use a learning model that is trained to output second predicted sea state data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea state data that predicts the sea state at the first time point are input; and output predicted sea state data a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device.

2. The computer program of claim 1, wherein the learning model is a graph neural network that inputs and outputs a graph described by edge data indicating the relationships between nodes and feature values ​​of each node, and each node of the graph input and output to the graph neural network corresponds to a position on or underwater, and each node is associated with multiple types of ocean condition data at the corresponding position as feature values.

3. The computer program of claim 2, which causes the computer to execute a process of inputting observation data at an observation position, sea state data at multiple positions included in the sea area, and edge data indicating the relationship between the observation position and nodes corresponding to the multiple positions into the learning model, and obtaining predicted sea state data for the position on or underwater at the specified time after from the learning model.

4. A computer program as described in claim 2 or 3, which causes the computer to execute a process of adding an observation node corresponding to the observation position of the observation data and the observation data associated with the observation node to a graph corresponding to the first predicted sea state data at the first time point, and adding edge data indicating the positional relationship between the observation node and the node of the first predicted sea state data to the graph.

5. A computer program as described in claim 4, wherein the first predicted sea state data includes sea state data at a plurality of positions regularly arranged corresponding to meshes that regularly divide the sea area, the observation position is located between the plurality of positions, and the edge data includes data on the distance between the positions.

6. The computer program according to claim 1, wherein the type of data contained in the observation data is different from the type of data contained in the first predicted sea state data.

7. The computer program of claim 1, wherein the observation data, the first predicted sea state data, and the second predicted sea state data each include at least one of water temperature, current speed, current direction, salinity concentration, depth, and sea surface altitude.

8. The computer program according to claim 1, which causes the computer to execute a process of acquiring the observation position and the observation data from an observation device attached to a ship, a marine drone, an underwater drone, a buoy, a satellite, or a marine organism, and outputting the second predicted sea condition data for the sea area including the observation position using the acquired observation data and the learning model.

9. The computer program described in claim 1, wherein the learning model is trained using training data generated by a simulator that calculates, by simulation, second predicted sea state data at a second time point a predetermined time after the first time point based on first predicted sea state data at a first time point and observation data obtained from the device at the first time point.

10. A sea condition prediction device having a processing unit that acquires observation data at an observation position from a device that observes the sea or underwater, acquires predicted sea condition data that predicts the sea conditions at the time of observation at multiple positions included in the sea area including the observation position, and uses a learning model that is trained to output second predicted sea condition data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea condition data that predicts the sea conditions at the first time point are input, and outputs predicted sea condition data for a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device.

11. A sea condition prediction method in which a computer acquires observation data at an observation position from a device that observes the sea or underwater, acquires predicted sea condition data that predicts the sea conditions at the time of observation at multiple positions included in an ocean area that includes the observation position, and uses a learning model that is trained to output second predicted sea condition data at a second time point that is a predetermined time after the first time point when the observation data acquired from the device at a first time point and first predicted sea condition data that predicts the sea conditions at the first time point are input, and outputs predicted sea condition data a predetermined time after the observation time of the observation data that corresponds to the observation data acquired from the device.

Images