Evaluation device, evaluation method, and navigation control device

Abstract

An evaluation device according to the present technology comprises an evaluation calculation unit trained to obtain deposit evaluation data as output through machine learning that takes at least one of ocean topography data or particulate component measurement data for known seafloor deposits as input data for training. The evaluation device accepts at least one of ocean topography data or particulate component measurement data for an ocean area to be evaluated as input, and the evaluation calculation unit calculates deposit evaluation data for the ocean area to be evaluated.

Description

[Technical Field]

[0001] The present technology relates to the technical fields of an evaluation device and method for evaluating seabed mineral deposits, and a navigation control device that controls navigation for mineral deposit exploration based on the evaluation of seabed mineral deposits. [Background technology]

[0002] For example, as disclosed in Patent Document 1 below, various techniques have been proposed for exploring seabed mineral deposits. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2011-158343 Summary of the Invention [Problem to be solved by the invention]

[0004] In mineral deposit exploration, it is desirable to improve exploration efficiency, because low exploration efficiency leads to an increase in the number of navigation days and an increase in the cost required for exploration.

[0005] Furthermore, in conventional mineral deposit exploration, acoustic Doppler surveys (acoustic reconnaissance surveys) are often used to roughly estimate the location of mineral deposits. In these reconnaissance surveys, sound waves are emitted from an air gun, for example, from a research vessel, and sound waves are measured by analyzing the reflection from the seafloor. However, it has been pointed out that these waves may have a negative impact on the ecology of zooplankton, such as krill larvae and copepods, which serve as food for fish, posing a risk of environmental destruction.

[0006] This technology was developed in light of the above circumstances, and aims to reduce the costs associated with mineral deposit exploration while also reducing the risk of environmental destruction. [Means for solving the problem]

[0007] The evaluation device according to the present technology includes an evaluation calculation unit that is trained to obtain deposit evaluation data as output through machine learning using at least one of marine topography data and fine particle component measurement data for known seafloor deposits as learning input data, and the evaluation calculation unit calculates the deposit evaluation data for the sea area to be evaluated using at least one of the marine topography data and the fine particle component measurement data for the sea area to be evaluated as input. "Marine topography data" refers to ocean-related topography data, such as data showing the seafloor topography of a certain ocean area or data showing the coastline topography. "Fine particle component measurement data" refers to data obtained by measuring fine particle components in seawater, such as microbial data, which is measurement data on microorganisms such as plankton and bacteria in seawater; electrical conductivity, temperature, pH (hydrogen ion exponent), gas concentrations of specific gases such as methane, hydrogen, and helium; and metal concentrations of specific metals such as manganese and iron in seawater. "Mineral deposit evaluation data" refers to data indicating the evaluation of seafloor mineral deposits, such as mineral deposit existence probability, hydrothermal plume probability, which indicates the probability that a measurement point is within a hydrothermal plume, and estimated ore types, which indicate the types of ores that may be present in a mineral deposit. According to the above configuration, by using AI (artificial intelligence) as a calculation unit for evaluating seafloor mineral deposits, it is possible to improve the accuracy of the evaluation. Furthermore, by calculating mineral deposit evaluation data using the evaluation calculation unit described above, it is possible to realize a reconnaissance survey that has less risk of environmental destruction and is an alternative to acoustic reconnaissance surveys when evaluating mineral deposits in the sea area to be evaluated.

[0008] In the evaluation device according to the present technology described above, the evaluation calculation unit may be configured to have been trained to output a probability of existence of a mineral deposit as the mineral deposit evaluation data. This will improve the accuracy of the evaluation when assessing the probability of mineral deposits in the target sea area. Also, it will realize a method of evaluating the probability of mineral deposits, which is an alternative to acoustic surveys and poses less risk of environmental destruction. The probability of mineral deposit existence can also be used as an evaluation index for actions taken in mineral deposit exploration.

[0009] In the evaluation device according to the present technology described above, the evaluation calculation unit may be configured to be trained to output a hydrothermal plume probability as the deposit evaluation data by machine learning using at least microparticle component measurement data for known seafloor deposits as learning input data. Hydrothermal plume probability can be used as an indicator for the actions taken in mineral exploration.

[0010] In the evaluation device according to the present technology described above, the evaluation calculation unit may be configured to have learned to output a floating plume probability as the hot water plume probability. The floating plume probability is information indicating the probability that a measurement point is within a floating plume of a hydrothermal plume. This floating plume probability can be used as an evaluation index for actions taken in mineral deposit exploration.

[0011] In the evaluation device according to the present technology described above, the evaluation calculation unit may be configured to have learned to output the probability of a rising plume as the probability of a hot water plume. The emerging plume probability is information indicating the probability that a measurement point is within the emerging plume of a hydrothermal plume. This emerging plume probability can be used as an evaluation index for actions taken in mineral deposit exploration.

[0012] In the evaluation device according to the present technology described above, the evaluation calculation unit may be configured to be trained to obtain deposit evaluation data as output by machine learning using marine topography data and fine particle component measurement data for known seafloor deposits as learning input data. By learning based on multiple input elements related to seafloor mineral deposits, it is possible to improve the accuracy of mineral deposit evaluation data.

[0013] In the evaluation device according to the present technology described above, the evaluation calculation unit may be configured to be trained to output the hydrothermal plume probability by machine learning using microparticle component measurement data for known seafloor deposits and measurement location data indicating the measurement points as learning input data. Measurement location data can include, for example, data indicating latitude and longitude, and data indicating water depth. By using such measurement location data in combination with particulate component measurement data for learning, it is possible to improve the accuracy of hydrothermal plume probability.

[0014] The evaluation method according to the present technology is an evaluation method in which at least one of marine topography data and fine particle component measurement data for a target sea area is input to an evaluation calculation unit that has been trained to obtain deposit evaluation data as output through machine learning using at least one of marine topography data and fine particle component measurement data for known seafloor deposits as learning input data, and the mineral deposit evaluation data for the target sea area is calculated. Such an evaluation method also provides the same effects as the evaluation device according to the present technology described above.

[0015] In addition, the navigation control device according to the present technology includes an evaluation calculation unit that is trained to obtain mineral deposit evaluation data as output through machine learning using at least one of marine topography data and fine particle component measurement data for known seabed mineral deposits as learning input data, and a control information generation unit that generates navigation control information, which is control information related to navigation, based on the mineral deposit evaluation data for the sea area to be evaluated that is output by the evaluation calculation unit using at least one of the marine topography data and the fine particle component measurement data for the sea area to be evaluated as input. This means that when generating navigation control information for mineral deposit exploration based on mineral deposit evaluation data, using AI as the calculation unit for evaluation makes it possible to perform highly accurate evaluations of seabed mineral deposits. Furthermore, with the above configuration, it is possible to realize a reconnaissance survey that has less risk of environmental destruction and is an alternative to acoustic reconnaissance surveys when calculating mineral deposit evaluation data.

[0016] In the navigation control device related to the present technology described above, the evaluation calculation unit is trained to output a hydrothermal plume probability as the deposit evaluation data through machine learning using at least fine particle component measurement data for known seafloor deposits as learning input data, and the control information generation unit is configured to generate the navigation control information based on the hydrothermal plume probability output by the evaluation calculation unit using at least the fine particle component measurement data for the sea area to be evaluated as input. This makes it possible to generate navigation control information using the hydrothermal plume probability as an indicator for evaluating navigation behavior.

[0017] In the navigation control device related to the above-mentioned present technology, the evaluation calculation unit may include an evaluation calculation unit trained to output a floating plume probability and an evaluation calculation unit trained to output a floating plume probability, and the control information generation unit may be configured to generate the navigation control information to increase the floating plume probability when the floating plume probability is equal to or less than a predetermined value, and to generate the navigation control information to increase the floating plume probability when the floating plume probability exceeds the predetermined value and the floating plume probability is equal to or less than the predetermined value. This makes it possible to gradually guide the device subject to navigation control from outside the hydrothermal plume to inside the floating plume, and from inside the floating plume to inside the emerging plume.

[0018] In the above-described navigation control device according to the present technology, the control information generation unit may be configured to generate the navigation control information by reinforcement learning based on the ore deposit evaluation data. This makes it possible to generate navigation control information that maximizes the evaluation value (reward) for the action. [Brief explanation of the drawings]

[0019] [Figure 1] This is an explanatory diagram of seafloor deposits. [Figure 2]FIG. 1 is a diagram illustrating an example of the configuration of an evaluation system according to a first embodiment. [Figure 3] FIG. 2 is a diagram for explaining an example of a machine learning technique for generating an evaluation model in the first embodiment. [Figure 4] FIG. 10 is a schematic diagram of mineral deposit exploration in a second embodiment. [Figure 5] FIG. 10 is a diagram illustrating an example of the internal configuration of a navigation control device according to a second embodiment. [Figure 6] FIG. 1 is a diagram for explaining an example of a machine learning technique for generating an evaluation model for hydrothermal plume probability. [Figure 7] 10 is a flowchart showing a processing procedure for realizing navigation control according to a second embodiment. [Figure 8] 10 is a flowchart showing an example of a navigation control process according to a first modified example. [Figure 9] FIG. 10 is a diagram showing a configuration example of a mineral deposit exploration system as a second modified example. [Figure 10] FIG. 10 is a diagram showing another example of the configuration of a mineral deposit exploration system. DETAILED DESCRIPTION OF THE INVENTION

[0020] The embodiments will be described below in the following order. <1. First embodiment> [1-1. About seafloor deposits] [1-2. Evaluation system configuration] [1-3. First Example] [1-4. Second example] 2. Second Embodiment <3. Modifications> [3-1. First Modification] [3-2. Second Modification] [3-3. Other variations] <4. Summary of embodiments> <5. This technology>

[0021] <1. First embodiment> [1-1. About seafloor deposits] First, let us explain seafloor deposits with reference to Figure 1. Seafloor mineral deposits (also called seafloor hydrothermal deposits) are formed when seawater that has penetrated deep underground is heated by the heat of magma and other sources and erupts into the sea, causing the precipitation of minerals such as copper, lead, zinc, gold, silver, and other rare metals. Unlike the ocean surface, which relies on photosynthesis using solar energy, the areas near seafloor deposits are home to a unique ecosystem that uses hydrothermal vents as its energy source. The hot water that spews from hydrothermal vents contains high concentrations of sulfide, and bacteria called sulfur bacteria synthesize and excrete carbohydrates from these sulfides. Organisms that can survive without solar energy gather near the seafloor deposits by symbiotically living inside or on the surface of their bodies as primary producers, or by feeding on these sulfur bacteria, forming an ecosystem different from that found on land.

[0022] As shown in the figure, hydrothermal vents are formed at the top of chimneys that rise from the seafloor, and the hot water from the hydrothermal vents rises from the hydrothermal vents and spreads out like a mushroom cloud. The water mass affected by such hot water is called a hydrothermal plume. The seawater in hydrothermal plumes has chemosynthetic bacteria as its primary producers, which feed on plankton, which then feed on fish and other organisms in the food chain. Hydrothermal plumes have characteristics such as slightly higher water temperature, lower pH (hydrogen ion potential), and higher electrical conductivity compared to normal seawater at the same depth that is not affected by hydrothermal fluid. It has also been discovered that microbial cell density, along with various chemical components, increases in hydrothermal plumes, and it has become clear that microbial cell density can be up to several times higher than in the surrounding seawater. Therefore, by examining these characteristics, it is possible to estimate the presence or absence of hydrothermal plumes, and to conduct preliminary surveys (primary surveys that roughly explore the sea area) for seafloor mineral deposits.

[0023] The hot water that erupts from a hydrothermal vent into the seawater rises until it becomes equal in density to the surrounding seawater, and then spreads horizontally on deep currents. A hydrothermal plume that rises through the seawater is called a "buoyant plume," while a plume that spreads horizontally is called a "non-buoyant plume."

[0024] [1-2. Evaluation system configuration] FIG. 2 is a diagram showing an example of the configuration of an evaluation system as a first embodiment for evaluating seafloor mineral deposits. As shown in the figure, the evaluation system includes an evaluation device 1 and an operation / display terminal 2 . The evaluation device 1 includes a calculation unit 10 and a communication unit 15. The calculation unit 10 is configured with a microcomputer having, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory), and performs various calculations and controls to realize various operations of the evaluation device 1. The communication unit 15 performs data communication with an external device, particularly the operation and display terminal 2 in this example. The operation / display terminal 2 is a device that accepts operation inputs for various operational instructions to the evaluation device 1, transmits operation input information to the evaluation device 1, and displays information received from the evaluation device 1, and is composed of, for example, a PC (personal computer), a tablet terminal, a smartphone, etc.

[0025] The calculation unit 10 has the functions of a data acquisition unit 11 and an evaluation calculation unit 12 . The evaluation calculation unit 12 receives at least one of marine topography data and fine particle component measurement data for the sea area to be evaluated and calculates mineral deposit evaluation data for the sea area to be evaluated. The data acquisition unit 11 is a functional unit that acquires input data for the evaluation calculation unit 12.

[0026] Here, the marine topography data refers to topography data relating to the ocean, such as data showing the undersea topography of a certain sea area or data showing the coastline topography.

[0027] Furthermore, the particulate matter measurement data refers to data obtained by measuring particulate matter components in seawater. Examples of microparticle component measurement data include microbial data, which is measurement data on microorganisms such as plankton and bacteria in seawater, and water quality data, which is measurement data related to the water quality of seawater (excluding measurements related to microorganisms), such as electrical conductivity, temperature, and pH in seawater. Examples of microbial data include the number, type, and size of the measured microbial organisms. Here, the size may be multiple types of data such as the minimum size and the maximum size, or it may be the size in the thickness direction. Examples of water quality data include the electrical conductivity, temperature, and pH mentioned above, as well as gas concentrations of specific gases such as methane, hydrogen, and helium, and metal concentrations of specific metals such as manganese and iron. Here, the particulate component measurement data is associated with measurement location data, which is data indicating the measurement point. The measurement location data includes latitude and longitude data and water depth data.

[0028] Deposit evaluation data is data that indicates an evaluation of a seafloor deposit, and examples include deposit existence probability, which indicates the probability that a deposit exists, hydrothermal plume probability, which indicates the probability that the measurement point is within a hydrothermal plume, and estimated ore type, which indicates the type of ore that may be present in the deposit.

[0029] The evaluation calculation unit 12 has an evaluation model (trained evaluation model) that has been trained to obtain deposit evaluation data as output by machine learning using at least one of marine topography data and fine particle component measurement data for known seafloor deposits as learning input data, and calculates deposit evaluation data for the sea area to be evaluated (hereinafter referred to as the "evaluation target sea area") using at least one of marine topography data and fine particle component measurement data for the sea area to be evaluated. In this example, since measurement location data is associated with the fine particle component measurement data, when the fine particle component measurement data is used as input data, the measurement location data is also used as input data.

[0030] In the first embodiment, an example (first example) in which the evaluation calculation unit 12 calculates the deposit existence probability as the deposit evaluation data and an example (second example) in which the evaluation calculation unit 12 calculates the estimated ore type will be described.

[0031] [1-3. First Example] A first example will be described. Here, an example will be described in which the input data for evaluation is ocean topography data, fine particle component measurement data, and measurement location data associated with the fine particle component measurement data, and the evaluation calculation unit 12 calculates the deposit existence probability. An example of a machine learning technique for generating an evaluation model in the evaluation calculation unit 12 will be described with reference to FIG. First, for learning, a machine learning device 20 is used. In this example, the machine learning device 20 is a machine learning device that supports supervised learning.

[0032] The input data sets used for learning are a training data set for mineral deposit areas and a training data set for non-mineral deposit areas. Here, mineral deposit areas refer to areas where the existence of seafloor mineral deposits has been confirmed, and non-mineral deposit areas refer to areas where the existence of seafloor mineral deposits has been confirmed. As described above, in this example, oceanographic topography data, fine particle component measurement data, and measurement location data are input, so the training data set for ore deposit areas uses the oceanographic topography data, fine particle component measurement data, and measurement location data for ore deposit areas, and the training data set for non-ore deposit areas uses the oceanographic topography data, fine particle component measurement data, and measurement location data for non-ore deposit areas. In this case, data indicating the presence or absence of seafloor ore deposits is used as the training data as a correct label for the training data. That is, when training data about an ore-deposit sea area is input as training input data, data indicating the presence of seafloor ore deposits (i.e., probability of ore deposit existence = 100%) is provided as training data for the machine learning device 20, and when training data about an ore-free sea area is input as training input data, data indicating the absence of seafloor ore deposits (i.e., probability of ore deposit existence = 0%) is provided as training data for the machine learning device 20.

[0033] By using the learning data sets and teacher data for the above-mentioned mineral deposit and non-mineral deposit sea areas through machine learning, the machine learning device 20 can generate an evaluation model that takes marine topography data, fine particle component measurement data, and measurement location data as inputs and outputs the probability of mineral deposit existence.

[0034] The machine learning here can be realized using a known machine learning engine that handles supervised regression problems. As an example, it can be realized using a logistic regression learning engine, which is often used as a learning engine to predict the probability of a specific event occurring from multiple factors. It is known that the aforementioned hydrothermal plumes contain distinctive nutrients, which serve as food for the plankton, and minerals that affect microbial growth. Furthermore, the unique microbial species and densities within hydrothermal plumes indicate that they have a distinct ecological environment from other ocean regions. It is believed that these distinctive ecological environments within hydrothermal plumes also influence the environment outside the plume through food chains and ocean currents. By statistically analyzing these relationships using machine learning, we can estimate the probability of mineral deposits in the evaluation area without deep-sea diving, using only the particle component measurement data from the ocean surface and mid-depth layers (described below). This enables seafloor resource exploration that is safer, more efficient, and less costly in terms of environmental conservation than conventional acoustic surveys.

[0035] The evaluation model obtained by the machine learning device 20 through the above-described machine learning is used as the evaluation model for the evaluation calculation unit 12 shown in Fig. 2. In this case, the evaluation calculation unit 12 receives as input the marine topography data, the fine particle component measurement data, and the measurement location data for the sea area to be evaluated, and can calculate the probability of the existence of mineral deposits in the sea area to be evaluated.

[0036] [1-4. Second example] The second example is an example in which the evaluation calculation unit 12 calculates an estimated ore type using marine topography data, fine particle component measurement data, and measurement location data as input data. In the second example, the machine learning device 20 that supports supervised learning is also used in the machine learning. In the second example, only a training dataset for the ore deposit area is used as the input dataset for learning. Specifically, the training dataset includes oceanographic data, fine particle component measurement data, and measurement location data for the ore deposit area.

[0037] In the second example, information indicating the types of ores present in the ore deposit is used as the training data to be provided to the machine learning device 20. Specifically, when learning data about the sea area of ​​the ore deposit is input as learning input data, data indicating the types of ores that actually existed in the seabed ore deposit is provided as training data to the machine learning device 20.

[0038] By using the learning data set and teacher data for the ore deposit sea area as described above, machine learning device 20 can generate an evaluation model that takes marine topography data, fine particle component measurement data, and measurement location data as inputs and outputs an estimated ore type.

[0039] In the first and second examples, the input data for calculating the probability of existence of a mineral deposit and the estimated type of ore are oceanographic topography data, fine particle component measurement data, and measurement location data. However, the probability of existence of a mineral deposit and the estimated type of ore can also be calculated based on oceanographic topography data alone. In this case, the machine learning for generating an evaluation model uses only oceanographic topography data as the learning input data set in the learning method described above. While the fine particle component measurement data requires actual navigation to the designated sea area to conduct the measurements, known topographical data can be used for the oceanographical data. Therefore, if the evaluation is based solely on the oceanographical data as described above, it is possible to avoid conducting a reconnaissance survey in sea areas where the probability of the existence of mineral deposits is low, or to conduct a reconnaissance survey with a lower priority, which has the advantage of improving the efficiency of exploration.

[0040] The evaluation device 1 can also be configured to calculate the hydrothermal plume probability as the deposit evaluation data, but a method for generating an evaluation model for the hydrothermal plume probability will be described in the second embodiment below.

[0041] 2. Second Embodiment In the second embodiment, navigation control for mineral deposit exploration is performed based on mineral deposit evaluation data. As an example of navigation control, an example of navigation control of an AUV (Autonomous Underwater Vehicle: an untethered autonomous underwater robot) will be described below. For clarity, in this specification, "navigation" is used in a broad sense to include not only movement on water but also movement underwater.

[0042] FIG. 4 is a schematic diagram of mineral deposit exploration in the second embodiment. The mineral deposit exploration here is carried out by controlling navigation using the floating plume probability and the rising plume probability, which are hydrothermal plume probabilities, as control indicators to identify the location of the seafloor mineral deposit. Specifically, the mineral deposit exploration in this case is realized by step-by-step navigation control according to the values ​​of the floating plume probability and the rising plume probability, and the exploration is divided into the first exploration phase, the second exploration phase, and the third exploration phase as shown in the figure.

[0043] The first search phase is a phase for searching for floating plumes, and is a search phase in which navigation control is performed with the goal of making the floating plume probability exceed a predetermined value. The second search phase is a search phase in which, after the floating plume probability exceeds a predetermined value, navigation control is performed with the goal of making the floating plume probability exceed the predetermined value. The third search phase is a search phase in which navigation control is performed with the goal of recognizing a hydrothermal vent after the probability of the emerging plume exceeds a predetermined value.

[0044] In this example, navigation control for exploration is performed based on the results of behavioral learning performed by the AUV itself from its past actions and their results. Reinforcement learning is used for this behavioral learning. Below, we will explain the method for deriving optimal actions when Q-learning is used as reinforcement learning.

[0045] Q-learning is a machine learning method that learns through trial and error how to "maximize future rewards (points)." With the aim of using the Q-learning concept to control an AUV and guide it to a mine, we will use machine learning to automatically learn through trial and error so that the Q-value, which is an evaluation value for its actions, is maximized. The Q value is given by the state S at time t. t Action a t As a value function when t ,a t ) is defined as The following [Equation 1] is a mathematical model of Q-learning.

number

[0046] In [Equation 1], state S t represents the state information (probability of deposit existence) at time t, and action a t represents the AUV's control data (navigation control information) as an action at time t. AUV's control data change, i.e., action a t Due to the change, the status is S t+1 Changes to R t+1 represents the reward obtained by this change in state. Also, the second term on the right-hand side, maxQ(S t+1 ,a) represents the ideal value in the future. State S t+1 The Q value when the ideal future action a with the highest Q value is selected under the above condition is multiplied by the discount rate γ. Here, γ is a parameter in the range of 0<γ≦1, and is often set to 0.9 to 0.99. α is a learning coefficient, and is in the range of 0<α≦1, but a value of around 0.1 is usually used.

[0047] As shown above, the AUV's state S t In this study, we will change the AUV control data using machine learning equipment. t The function that gives the highest Q value when is taken as the value function (Q function), and the value function is repeatedly updated so that the Q value becomes higher.

[0048] FIG. 5 is a diagram showing an example of the internal configuration of an AUV 30 that performs automatic navigation using the behavior learning function described above. In the following description, parts that are the same as parts that have already been described will be given the same reference numerals and description thereof will be omitted.

[0049] As shown in the figure, the AVU 30 includes a seawater measurement unit 31, a sensor unit 32, a power unit 33, a power control unit 34, a power supply unit 35, a calculation unit 36, a memory 37, and a bus 38. The seawater measurement unit 31, the sensor unit 32, the power control unit 34, and the calculation unit 36 ​​are each connected to the bus 38 and are capable of communicating data with one another.

[0050] The seawater measurement unit 31 is configured to include various sensors, a microcomputer, and the like for measuring seawater. Specifically, the seawater measurement unit 31 is equipped with an imaging sensor (image capture element) as a sensor for measuring microorganisms. The seawater measurement unit 31 is configured to introduce seawater as a sample into a flow cell, and the imaging sensor can capture an image of the seawater introduced into the flow cell to obtain an image of the microorganisms. Based on this captured image, the seawater measurement unit 31 can obtain the above-mentioned microorganism data. Specifically, this data includes the number, type, size, and other information of the microorganisms. Here, the detection and type recognition of the microorganisms can be performed based on image analysis of the captured image. For example, this can be performed by image matching using a template image of the microorganisms. The seawater measurement unit 31 also includes various sensors for measuring the water quality data described above, such as electrical conductivity, temperature, pH, and the gas concentration of specific gases such as methane, hydrogen, and helium, as well as the metal concentration of specific metals such as manganese and iron. It should be noted that the seawater measurement unit 31 may be an in situ type measurement device that does not include a flow cell and directly images seawater.

[0051] The sensor unit 32 collectively refers to sensors for detecting the position of the AUV 30 and sensors for recognizing the external environment of the AUV 30. Here, regarding position information, it is conceivable to provide a GNSS (Global Navigation Satellite System) sensor that detects latitude and longitude information, or a sensor that detects water depth information. Furthermore, as a sensor for recognizing the external environment, for example, a camera (image sensor) or the like can be provided.

[0052] The power unit 33 comprehensively represents the parts that generate power for navigation, such as a motor that drives a propulsion screw and an actuator that adjusts the propulsion direction. The power control unit 34 controls the power unit 33 based on the navigation control information.

[0053] The power supply unit 35 is configured, for example, by a battery, and supplies power to each part of the AUV 30.

[0054] The calculation unit 36 ​​is configured with, for example, a microcomputer having a CPU, ROM, and RAM, and performs various calculations and controls to realize various operations of the AUV 30. The memory 37 is configured as a non-volatile storage device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive), and is used as a storage area for data used by the calculation unit 36 ​​in various processes.

[0055] Here, the calculation unit 36 ​​functions as the data acquisition unit 11 and evaluation calculation unit 12 described above. In this case, the evaluation calculation unit 12 uses one that has been trained to input marine topography data, fine particle component measurement data, and measurement location data and output a mineral deposit existence probability, similar to the one described as the first example of the first embodiment. As the marine topography data, for example, data stored in memory 37 is used. Specifically, in this case, the data acquisition unit 11 acquires marine topography data of the sea area corresponding to information on the position of the AUV 30 detected by the sensor unit 32 as input data for the evaluation calculation unit 12, and supplies this data to the evaluation calculation unit 12. In this case, the data acquisition unit 11 acquires the measurement data of the particulate matter component by the seawater measurement unit 31 and supplies the data to the evaluation calculation unit 12 .

[0056] The calculation unit 36 ​​also functions as a data acquisition unit 11A and an evaluation calculation unit 12A. The evaluation calculation unit 12A has an evaluation model trained to calculate the probability of a hot water plume. Specifically, the evaluation calculation unit 12A in this example is trained to calculate the above-mentioned floating plume probability and rising plume probability using the particulate component measurement data and measurement location data as input. Here, when calculating the floating plume probability and the rising plume probability individually, separate evaluation models for the floating plume probability and the rising plume probability would be provided, but in order to avoid complicating the illustration, these evaluation models are treated together as evaluation calculation unit 12A.

[0057] An example of a machine learning technique for generating an assessment model of hydrothermal plume probability will be described with reference to FIG. 6 . In this case, the evaluation model is generated using a machine learning device 20A. As with the machine learning device 20, the machine learning device 20A has a machine learning engine that handles well-known supervised regression problems, such as a logistic regression learning engine.

[0058] The input data sets used for learning are a learning data set for hydrothermal plumes and a learning data set for ocean areas other than those with hydrothermal plumes. Specifically, in generating an evaluation model for the probability of floating plumes, the learning data set for hydrothermal plumes uses measurement data of particulate components and measurement location data in ocean areas that are floating plumes, and the learning data set for ocean areas other than those with hydrothermal plumes uses measurement data of particulate components and measurement location data in ocean areas other than those with floating plumes and emerging plumes. In this case, data indicating whether or not it is a hydrothermal plume is used as training data. That is, in generating an evaluation model for the floating plume probability, when particulate component measurement data and measurement location data in an ocean area as a floating plume are used as training input data, data indicating that it is a floating plume (i.e., floating plume probability = 100%) is provided as training data for machine learning device 20A, and when particulate component measurement data and measurement location data in an ocean area other than a hydrothermal plume are used as training input data, data indicating that it is not a floating plume (floating plume probability = 0%) is provided. Furthermore, when generating an evaluation model for the probability of a rising plume, if the microparticle component measurement data and measurement location data in an ocean area representing a rising plume are used as learning input data, data indicating that it is a rising plume (i.e., the probability of a rising plume = 100%) is provided as training data for the machine learning device 20A, and if the microparticle component measurement data and measurement location data in an ocean area other than a hydrothermal plume are used as learning input data, data indicating that it is not a rising plume (the probability of a rising plume = 0%) is provided.

[0059] By performing machine learning on floating plumes and rising plumes using the machine learning device 20A with the above-described learning data sets and teacher data, respectively, it is possible to generate an evaluation model that uses marine topography data and measurement location data as input and obtains floating plume probability and rising plume probability as output.

[0060] 5, the data acquisition unit 11A acquires input data for the evaluation calculation unit 12A. Specifically, the data acquisition unit 11A in this example acquires measurement data by the seawater measurement unit 31 as particulate component measurement data and supplies it to the evaluation calculation unit 12, and acquires detection data by the sensor unit 32 as measurement location data and supplies it to the evaluation calculation unit 12.

[0061] The calculation unit 36 ​​also has a navigation control information generation unit 40, a value function calculation unit 41, and a reward calculation unit 42 as functional units for the reinforcement learning described above. The reward calculation unit 42 calculates the highest reward for the action (navigation control) that can be taken at that time, learned from the past under the current situation. The value function calculation unit 41 calculates the value function (state S t+1 The Q value when navigation control is performed as action a is calculated. The navigation control information generation unit 40 generates navigation control information that maximizes the value function (Q value) based on the value function calculated (updated) by the value function calculation unit 41. In this example, the navigation control information is, for example, information related to the control of the propulsion motor and the actuator for adjusting the propulsion direction. The navigation control information generation unit 40 outputs the generated navigation control information to the power control unit 34.

[0062] Here, in order to enable learning from past data, the calculation unit 36 ​​performs a process of recording the value function calculated by the value function calculation unit 41, the Q value obtained from the calculated value function, the state S (probability of mineral deposit existence), and the action a (navigation control information) in the memory 37.

[0063] FIG. 7 is a flowchart showing a processing procedure for realizing navigation control according to the second embodiment. The processing shown in FIG. 7 is executed by the calculation unit 36 ​​based on a program stored in a storage device such as an internal ROM or the memory 37.

[0064] First, in step S101, the calculation unit 36 ​​determines whether the search phase is the first search phase. If the exploration phase is the first exploration phase, the calculation unit 36 ​​proceeds to step S102 to start the first exploration phase, and then, in the following step S103, determines whether the floating plume probability at the measurement point exceeds a threshold. Specifically, in step S103, the calculation unit 36 ​​first causes the seawater measurement unit 31 to perform a measurement operation to obtain particulate component measurement data, and also causes the sensor unit 32 to detect latitude, longitude, and water depth to obtain measurement location data. Then, using the particulate component measurement data and measurement location data as input data, the calculation unit 36 ​​calculates the floating plume probability using an evaluation model for floating plume probability. It then determines whether the floating plume probability calculated in this manner exceeds a predetermined threshold.

[0065] In step S103, if it is determined that the floating plume probability at the measurement point does not exceed the threshold, the calculation unit 36 ​​proceeds to step S110 to reduce the reward (R), and then proceeds to step S111 to update the value function. That is, by performing the processing as the reward calculation unit 42 and value function calculation unit 41 described above, the value function (state S t+1 The Q value when navigation control is performed as action a is calculated. In calculating this value function, in this example, the probability of existence of a mineral deposit is used as state S. As can be understood from the above explanation, calculation unit 36 ​​calculates this probability of existence of a mineral deposit by processing as evaluation calculation unit 12, using as input data the marine topography data, the fine particle component measurement data, and the measurement location data acquired from memory 37, seawater measurement unit 31, and sensor unit 32, respectively, by processing as data acquisition unit 11.

[0066] In step S112 following step S111, calculation unit 36 ​​performs a process of recording the value function Q value, state S, and action a for each exploration phase in memory 37. Here, "for each exploration phase" refers to the first exploration phase, second exploration phase, and third exploration phase. The value function Q value refers to the Q value found from the updated value function.

[0067] In step S113 following step S112, the calculation unit 36 ​​generates, based on the updated value function, navigation control information that maximizes the value function, and outputs the navigation control information to the power control unit 34.

[0068] In step S114 following step S113, the calculation unit 36 ​​determines whether or not to end the search, that is, whether or not a predetermined condition set in advance as a search end condition is met. If the exploration end condition is not met and it is determined not to end the exploration, the calculation unit 36 ​​returns to step S101. As a result, during the first exploration phase (a situation in which the floating plume probability does not exceed the threshold), the processing of the above steps S110, S111, S112, and S113 is repeated unless it is determined in step S114 that the exploration should be ended, thereby realizing navigation control during the first exploration phase.

[0069] If it is determined in step S103 that the floating plume probability at the measurement point exceeds the threshold, the calculation unit 36 ​​proceeds to step S104 to increase the reward (R) and start the second exploration phase.

[0070] In response to starting the second exploration phase in step S104, the calculation unit 36 ​​proceeds to step S106 and determines whether the rising plume probability at the measurement point exceeds a threshold. That is, in step S106, the calculation unit 36 ​​first causes the seawater measurement unit 31 to perform a measurement operation to acquire particulate component measurement data, and also causes the sensor unit 32 to detect latitude, longitude, and water depth to acquire measurement location data. Then, using the particulate component measurement data and measurement location data as input data, the calculation unit 36 ​​calculates the rising plume probability using an evaluation model for the rising plume probability, and determines whether the calculated rising plume probability exceeds a predetermined threshold.

[0071] If it is determined in step S106 that the probability of the emerging plume at the measurement point does not exceed the threshold, the calculation unit 36 ​​proceeds to step S110 to reduce the reward. Note that the processing flow from step S110 onwards has been explained above, so a duplicate explanation will be avoided.

[0072] If it is determined in step S101 that the search phase is not the first search phase, the calculation unit 36 ​​proceeds to step S105 to determine whether the search phase is the third search phase. If it is determined that the search phase is not the third search phase (i.e., the search phase is the second search phase), the calculation unit 36 ​​proceeds to step S106. As a result, during the second exploration phase (a situation in which the probability of the emerging plume does not exceed the threshold), the processing of the above-mentioned steps S110, S111, S112, and S113 is repeated unless it is determined in step S114 that the exploration has ended, and navigation control during the second exploration phase is realized.

[0073] If it is determined in step S106 that the probability of the emerging plume at the measurement point exceeds the threshold, the calculation unit 36 ​​proceeds to step S107 to increase the reward and start the third exploration phase. Then, in step S108 following step S107, the calculation unit 36 ​​determines whether or not a hydrothermal vent has been detected. This determination can be made by image analysis of the image captured by the camera in the sensor unit 32. Note that the detection target in step S108 is not limited to a hydrothermal vent, and the detection target may be, for example, a part of a mineral deposit, such as a part of a chimney.

[0074] If it is determined in step S108 that no hydrothermal vents have been detected, the calculation unit 36 ​​proceeds to step S110. Here, if it is determined in the previous step S105 that the search phase is the third exploration phase, the calculation unit 36 ​​proceeds to the process in step S108. As a result, during the third exploration phase (a situation in which no hydrothermal vents have been detected), the processes of steps S110, S111, S112, and S113 are repeated unless it is determined in step S114 that the exploration has ended, thereby realizing navigation control during the third exploration phase.

[0075] If it is determined in step S108 that a hydrothermal vent has been detected, the calculation unit 36 ​​increases the reward in step S109, and then proceeds to step S111.

[0076] Furthermore, in response to determining in step S114 that the search is to be ended, the calculation unit 36 ​​ends the series of processes shown in FIG.

[0077] <3. Modifications> [3-1. First Modification] Here, the embodiment is not limited to the specific example described above, and various modified configurations can be adopted. For example, in the second embodiment, an example was described in which the floating plume probability and the rising plume probability are used as indicators for switching between exploration phases to perform navigation control divided into multiple exploration phases, but it is not necessary to explicitly divide the exploration phases.

[0078] FIG. 8 is a flowchart showing an example of a navigation control process that corresponds to a case where the exploration phase is not explicitly separated. In the example of Fig. 7, only the probability of existence of an ore deposit is used as the state S in the value function, but in Fig. 8, the probability of existence of an ore deposit and the probability of a hydrothermal plume are used as the state S. In this case, the value of state S is a value that reflects both the probability of existence of an ore deposit and the probability of a hydrothermal plume, for example, by setting it to the average value of the probability of existence of an ore deposit and the probability of a hydrothermal plume.

[0079] In this case, the calculation unit 36 ​​performs the data acquisition process of step S201, which is to acquire input data for the evaluation calculation units 12 and 12A for calculating the ore deposit existence probability and the hydrothermal plume probability, respectively. Specifically, in the case where the ore deposit existence probability is calculated using ocean topography data, fine particle component measurement data, and measurement location data as inputs, and the hydrothermal plume probability is calculated using fine particle component measurement data and measurement location data as inputs, as in this example, the ocean topography data, fine particle component measurement data, and measurement location data are acquired. Then, in step S202 following step S201, the calculation unit 36 ​​calculates the probability of existence of an ore deposit and the probability of a hydrothermal plume based on the acquired data using an evaluation model serving as the evaluation calculation units 12 and 12A.

[0080] The processing of steps S203 to S205 following step S202 is processing related to reinforcement learning by Q-learning. Specifically, in step S203, the calculation unit 36 ​​calculates the highest reward for the action (navigation control) that can be taken at that time, learned from the past under the current situation, and then in the following step S204, it calculates the value function when the navigation control setting that was the basis for the reward calculation is made (updating the value function). That is, in the state S t+1 The Q value is calculated when navigation control is performed as action a. Then, in step S205 following step S204, calculation unit 36 ​​performs processing to record the value of value function Q, state S, and action a in memory 37.

[0081] In step S206 following step S205, the calculation unit 36 ​​generates navigation control information that maximizes the value function based on the updated value function, and outputs the information to the power control unit 34.

[0082] Furthermore, in step S207 following step S206, the calculation unit 36 ​​determines whether or not to end the search (determines whether or not the search end condition is met, as in the previous step S114), and if it determines not to end the search, the process returns to step S201. On the other hand, if it is determined in step S207 that the search is to be ended, the calculation unit 36 ​​ends the series of processes shown in FIG.

[0083] [3-2. Second Modification] The second modification relates to a configuration for mineral deposit exploration. FIG. 9 is a diagram showing an example of the configuration of a mineral deposit exploration system as a second modified example. As shown in the figure, the mineral deposit exploration system as the second modified example includes at least an AUV 30A, a navigation control device 50, and an offshore repeater 60. This mineral deposit exploration system can be described as one in which the functions of the calculation unit 36 ​​possessed by the AUV 30 shown in Figure 5 have been transferred to the external navigation control device 50.

[0084] AUV30A differs from AUV30 in that it does not include the calculation unit 36 ​​and is provided with an acoustic communication unit 39 to enable communication underwater.

[0085] The navigation control device 50 includes a calculation unit 36, a memory 37, and a communication unit 51. The navigation control device 50 may be installed, for example, on a ship sailing on the ocean or on land. The offshore repeater 60 is placed on the ocean and includes an acoustic communication unit 61 and a communication unit 62. The communication unit 62 performs data communication with the communication unit 51 in the navigation control device 50 via a network NT, which is a communication network such as the Internet, a LAN (Local Area Network), or a satellite communication network.

[0086] With this configuration, the calculation unit 36 ​​in the navigation control device 50 can acquire data such as particulate component measurement data from the AUV 30A, which is required to generate navigation control information using the reinforcement learning described above, and can output the generated navigation control information to the AUV 30A to control the navigation of the AUV 30A.

[0087] In the navigation control system shown in Figure 9, the navigation control information generation unit 40 can also be provided on the AUV 30B side as shown in Figure 10. The AUV 30B differs from the AUV 30A in that the navigation control information generation unit 40 is connected to a bus 38. In Figure 10, the calculation unit 36 ​​from which the navigation control information generation unit 40 is omitted is referred to as calculation unit 36B, and the navigation control device 50 in which the calculation unit 36B is provided instead of the calculation unit 36 ​​is referred to as the navigation control device 50B.

[0088] [3-3. Other variations] The above provides an example of generating an evaluation model for hydrothermal plume probability through machine learning using only particulate component measurement data and measurement location data, but the evaluation model for hydrothermal plume probability can also be generated through machine learning using particulate component measurement data, measurement location data, and ocean topography data.

[0089] In addition, although AUVs have been cited above as an example of navigation devices used in mineral deposit exploration, the navigation devices can take other forms, such as ROVs (Remotely Operated Vehicles), manned submersibles, underwater drones, surface drones, underwater gliders, towed navigation devices, and research vessels.

[0090] Furthermore, depending on the type of seafloor deposit, there may be cases where data on the habitat of microorganisms that are presumed to be related to mineral deposits or data on characteristic seawater components are measured, even though they do not have water masses like hydrothermal plumes. As a result of learning based on data measured in such sea areas, it is expected that the probability of a hydrothermal plume being present may be low, but only the probability of a mineral deposit being present may be high. In such a case, in both the first and second embodiments, there are cases where only the probability of existence of a mineral deposit is used as reference data for exploration or reference data for navigation control.

[0091] <4. Summary of embodiments> As described above, the evaluation device of the embodiment (same 1, AUV 30, navigation control device 50, 50B) is equipped with an evaluation calculation unit (same 12, 12A) that has been trained to obtain deposit evaluation data as output through machine learning using at least one of marine topography data and fine particle component measurement data for known seabed deposits as learning input data, and the evaluation calculation unit calculates deposit evaluation data for the sea area to be evaluated using at least one of marine topography data and fine particle component measurement data for the sea area to be evaluated as input. According to the above configuration, by using AI (artificial intelligence) as a calculation unit for evaluating seafloor mineral deposits, it is possible to improve the accuracy of the evaluation. Therefore, the efficiency of mineral deposit exploration can be improved, and costs associated with mineral deposit exploration can be reduced. Furthermore, by calculating mineral deposit evaluation data using the evaluation calculation unit described above, it is possible to realize a reconnaissance survey that has less risk of environmental destruction and is an alternative to acoustic reconnaissance surveys when evaluating mineral deposits in the sea area to be evaluated. Therefore, the risk of environmental destruction can be reduced. It is understood that in the future, it will become an international standard to conduct environmental impact assessments of target seabed resource development areas before and after the development in order to balance seabed resource development with environmental conservation. One approach under consideration is to measure the habitat conditions of microorganisms such as plankton as one indicator of environmental assessment and compare them before and after resource development. In this technology, when using microparticle component measurement data as input data for calculating mineral deposit assessment data, measuring microorganisms not only has the advantage of being a resource exploration method with low risk of environmental destruction, but also has the significant advantage of being efficient and cost-effective, since it can also be used for the aforementioned environmental assessment measurements conducted during resource development.

[0092] In the evaluation device according to the embodiment, the evaluation calculation unit (12) is trained to output the probability of occurrence of a mineral deposit as the mineral deposit evaluation data. This will improve the accuracy of the evaluation when assessing the probability of mineral deposits in the target sea area. Also, it will realize a method of evaluating the probability of mineral deposits, which is an alternative to acoustic surveys and poses less risk of environmental destruction. Therefore, it is possible to improve the efficiency of mineral deposit exploration while reducing the risk of environmental destruction. The probability of mineral deposit existence can also be used as an evaluation index for actions taken in mineral deposit exploration. Therefore, the output of the evaluation calculation unit can be used as an action evaluation index in mineral deposit exploration, which can contribute to improving the efficiency of exploration.

[0093] Furthermore, in the evaluation device as an embodiment, the evaluation calculation unit (12A) is trained to output hydrothermal plume probability as deposit evaluation data through machine learning using at least microparticle component measurement data for known seafloor deposits as learning input data. Hydrothermal plume probability can be used as an indicator for the actions taken in mineral exploration. Therefore, the output of the evaluation calculation unit can be used as an action evaluation index in mineral deposit exploration, which can contribute to improving the efficiency of exploration.

[0094] Furthermore, in the evaluation device according to the embodiment, the evaluation calculation unit is trained to output the floating plume probability as the hot water plume probability. The floating plume probability can be used as an indicator for the actions taken in mineral exploration. Therefore, the output of the evaluation calculation unit can be used as a behavior evaluation index, which can contribute to improving the efficiency of exploration.

[0095] In the evaluation device according to the embodiment, the evaluation calculation unit is trained to output the probability of a rising plume as the probability of a hot water plume. The probability of a rising plume can be used as an indicator for the actions taken in mineral exploration. Therefore, the output of the evaluation calculation unit can be used as a behavior evaluation index, which can contribute to improving the efficiency of exploration.

[0096] Furthermore, in the evaluation device as an embodiment, the evaluation calculation unit is trained to obtain deposit evaluation data as output by machine learning using marine topography data and fine particle component measurement data for known seafloor deposits as learning input data. By learning based on multiple input elements related to seafloor mineral deposits, it is possible to improve the accuracy of mineral deposit evaluation data. Therefore, the accuracy of evaluation of seabed mineral deposits can be improved.

[0097] Furthermore, in the evaluation device as an embodiment, the evaluation calculation unit is trained to output a hydrothermal plume probability by machine learning using fine particle component measurement data for known seafloor deposits and measurement location data indicating the measurement points as learning input data. By learning using not only particulate component measurement data but also measurement location data, it is possible to improve the accuracy of hydrothermal plume probability. This will improve the accuracy of evaluations of seafloor mineral deposits.

[0098] In addition, the evaluation method as an embodiment is an evaluation method in which at least one of marine topography data and fine particle component measurement data for the sea area to be evaluated is input to an evaluation calculation unit that has been trained to obtain deposit evaluation data as output through machine learning using at least one of marine topography data and fine particle component measurement data for known seabed deposits as learning input data, and deposit evaluation data for the sea area to be evaluated is calculated. This evaluation method can also provide the same functions and effects as the evaluation device of the above embodiment.

[0099] An embodiment of the navigation control device (AUV30, navigation control device 50) comprises an evaluation calculation unit (12, 12A) that is trained to obtain mineral deposit evaluation data as output through machine learning using at least one of marine topography data and fine particle component measurement data for known seabed mineral deposits as learning input data, and a control information generation unit (navigation control information generation unit 40) that generates navigation control information, which is control information related to navigation, based on the mineral deposit evaluation data for the sea area to be evaluated that is output by the evaluation calculation unit using at least one of marine topography data and fine particle component measurement data for the sea area to be evaluated as input. This means that when generating navigation control information for mineral deposit exploration based on mineral deposit evaluation data, using AI as the calculation unit for evaluation makes it possible to perform highly accurate evaluations of seabed mineral deposits. Therefore, the accuracy of navigation control can be improved, which can improve the efficiency of mineral deposit exploration and reduce the costs associated with mineral deposit exploration. Furthermore, with the above configuration, it is possible to realize a reconnaissance survey that has less risk of environmental destruction and is an alternative to acoustic reconnaissance surveys when calculating mineral deposit evaluation data. Therefore, the risk of environmental destruction can be reduced.

[0100] In addition, in the navigation control device as an embodiment, the evaluation calculation unit (12A) is trained to output hydrothermal plume probability as deposit evaluation data through machine learning using at least fine particle component measurement data for known seafloor deposits as learning input data, and the control information generation unit generates navigation control information based on the hydrothermal plume probability output by the evaluation calculation unit using at least fine particle component measurement data for the sea area to be evaluated as input. This makes it possible to generate navigation control information using the hydrothermal plume probability as an indicator for evaluating navigation behavior. Therefore, the efficiency of mineral deposit exploration can be improved.

[0101] Furthermore, in the navigation control device as an embodiment, the evaluation calculation unit includes an evaluation calculation unit trained to output the floating plume probability and an evaluation calculation unit trained to output the floating plume probability, and the control information generation unit generates navigation control information to increase the floating plume probability when the floating plume probability is below a predetermined value, and generates navigation control information to increase the floating plume probability when the floating plume probability exceeds the predetermined value and the floating plume probability is below the predetermined value. This makes it possible to gradually guide the device subject to navigation control from outside the hydrothermal plume to inside the floating plume, and from inside the floating plume to inside the emerging plume. Therefore, the efficiency of mineral deposit exploration can be improved.

[0102] Furthermore, in the navigation control device according to the embodiment, the control information generating unit generates the navigation control information by reinforcement learning based on the deposit evaluation data. This makes it possible to generate navigation control information that maximizes the evaluation value (reward) for the action. Therefore, the efficiency of mineral deposit exploration can be improved.

[0103] The effects described in this specification are merely examples and are not limiting, and other effects may also be present.

[0104] <5. This technology> The present technology can also be configured as follows. (1) an evaluation calculation unit that is trained to obtain deposit evaluation data as an output by machine learning using at least one of oceanographic topography data and microparticle component measurement data for known seafloor deposits as learning input data; The evaluation calculation unit calculates the mineral deposit evaluation data for the evaluation target sea area by inputting at least one of the marine topography data and the fine particle component measurement data for the evaluation target sea area. Evaluation equipment. (2) The evaluation calculation unit is trained to output a probability of existence of a mineral deposit as the mineral deposit evaluation data. The evaluation device according to (1) above. (3) The evaluation calculation unit The system is trained to output a hydrothermal plume probability as deposit evaluation data by machine learning using at least particulate component measurement data for known seafloor deposits as learning input data. The evaluation device according to (1) or (2). (4) The evaluation calculation unit is trained to output a floating plume probability as the hot water plume probability. The evaluation device according to (3) above. (5) The evaluation calculation unit is trained to output the probability of a rising plume as the probability of a hot water plume. The evaluation device according to (3) above. (6) The evaluation calculation unit is trained to obtain deposit evaluation data as an output by machine learning using oceanographic topography data and fine particle component measurement data for known seafloor deposits as learning input data. The evaluation device according to any one of (1) to (5) above. (7) The evaluation calculation unit The system is trained to output the hydrothermal plume probability by machine learning using the particulate component measurement data for known seafloor deposits and the measurement location data indicating the measurement points as learning input data. The evaluation device according to any one of (3) to (6) above. (8) An evaluation calculation unit is trained to obtain deposit evaluation data as an output by machine learning using at least one of oceanographic topography data and fine particle component measurement data for known seabed deposits as learning input data, and calculates the deposit evaluation data for the evaluation target sea area by inputting at least one of the oceanographic topography data and the fine particle component measurement data for the evaluation target sea area. Evaluation method. (9) an evaluation calculation unit that is trained to obtain deposit evaluation data as an output by machine learning using at least one of marine topography data and microparticle component measurement data for known seafloor deposits as learning input data; and a control information generating unit that generates navigation control information, which is control information related to navigation, based on the mineral deposit evaluation data for the sea area to be evaluated that is output by the evaluation calculation unit using at least one of the marine topography data and the microparticle component measurement data for the sea area to be evaluated as input. Navigation control device. (10) The evaluation calculation unit learning to output a hydrothermal plume probability as the deposit evaluation data by machine learning using at least particulate component measurement data for known seafloor deposits as learning input data; The control information generation unit The navigation control information is generated based on the hydrothermal plume probability output by the evaluation calculation unit using at least the particulate component measurement data for the evaluation target sea area as an input. The navigation control device according to (9) above. (11) The evaluation calculation unit includes an evaluation calculation unit trained to output a floating plume probability and an evaluation calculation unit trained to output a rising plume probability, The control information generation unit If the floating plume probability is equal to or less than a predetermined value, the navigation control information is generated to increase the floating plume probability; When the floating plume probability exceeds the predetermined value and the rising plume probability is equal to or less than the predetermined value, the navigation control information is generated to increase the rising plume probability. The navigation control device according to (10) above. (12) The control information generation unit The navigation control information is generated by reinforcement learning based on the mineral deposit evaluation data. A navigation control device according to any one of (9) to (11) above. [Explanation of symbols]

[0105] 1 Evaluation device 2 Operation and display terminal 10 Arithmetic section 11,11A Data acquisition section 12,12A Evaluation calculation section 15 Communications Department 20,20A Machine Learning Device 30, 30A, 30B AUV 31 Seawater measurement section 32 Sensor section 33 Power section 34 Power control section 35 Power supply section 36,36B calculation section 37 Memory 38 Bus 39,61 Acoustic Communication Department 40 Navigation control information generation unit 41 Value function calculation unit 42 Remuneration Calculation Department 50,50B Navigation control device 51,62 Communications Department 60 Offshore repeater NT Network

Claims

1. an evaluation calculation unit that has been trained to obtain ore deposit evaluation data as an output based on feature amounts of the oceanographic topography data or feature amounts of the fine particle component measurement data and feature amounts of the measurement location data, by machine learning using marine topography data for known seafloor deposits as learning input data, or by machine learning using fine particle component measurement data and measurement location data for the fine particle component measurement data for known seafloor deposits as learning input data, The evaluation calculation unit calculates the mineral deposit evaluation data for the evaluation target sea area using the marine topography data for the evaluation target sea area or the fine particle component measurement data and the measurement location data for the evaluation target sea area as input. Evaluation equipment.

2. The evaluation calculation unit is trained to output a probability of existence of a mineral deposit as the mineral deposit evaluation data. The evaluation device according to claim 1 .

3. The evaluation calculation unit The system is trained to output a hydrothermal plume probability as the deposit evaluation data by machine learning using the particulate component measurement data and the measurement location data for known seafloor deposits as learning input data. The evaluation device according to claim 1 .

4. The evaluation calculation unit is trained to output a floating plume probability as the hot water plume probability. The evaluation device according to claim 3 .

5. The evaluation calculation unit is trained to output the probability of a rising plume as the probability of a hot water plume. The evaluation device according to claim 3 .

6. The evaluation calculation unit is trained to obtain deposit evaluation data as an output by machine learning using oceanographic topography data, fine particle component measurement data, and measurement location data of the fine particle component measurement data for known seafloor deposits as learning input data. The evaluation device according to claim 1 .

7. An evaluation calculation unit is trained to obtain deposit evaluation data as an output based on the feature amounts of the marine topography data or the feature amounts of the fine particle component measurement data and the feature amounts of the measurement location data by machine learning using marine topography data for known seabed deposits as learning input data, or by machine learning using fine particle component measurement data and measurement location data for the fine particle component measurement data for known seabed deposits as learning input data, and the marine topography data for the sea area to be evaluated, or the fine particle component measurement data and the measurement location data for the sea area to be evaluated, is inputted to calculate the deposit evaluation data for the sea area to be evaluated. Evaluation method.

8. an evaluation calculation unit that is trained to obtain deposit evaluation data as an output based on the feature amounts of the oceanographic topography data or the feature amounts of the fine particle component measurement data and the feature amounts of the measurement location data, by machine learning using marine topography data for known seafloor deposits as learning input data, or by machine learning using fine particle component measurement data and measurement location data for the fine particle component measurement data for known seafloor deposits as learning input data; a control information generation unit that generates navigation control information, which is control information related to navigation, based on the marine topography data for the sea area to be evaluated, or the mineral deposit evaluation data for the sea area to be evaluated that is output by the evaluation calculation unit using the marine topography data for the sea area to be evaluated or the fine particle component measurement data and the measurement location data for the sea area to be evaluated as input. Navigation control device.

9. The evaluation calculation unit learning to output a hydrothermal plume probability as the deposit evaluation data by machine learning using the particulate component measurement data and the measurement location data for known seafloor deposits as learning input data; The control information generation unit The navigation control information is generated based on the hydrothermal plume probability output by the evaluation calculation unit using the particulate component measurement data and the measurement location data for the sea area to be evaluated as input. The navigation control device according to claim 8.

10. The evaluation calculation unit includes an evaluation calculation unit trained to output a floating plume probability and an evaluation calculation unit trained to output a rising plume probability, The control information generation unit If the floating plume probability is equal to or less than a predetermined value, the navigation control information is generated to increase the floating plume probability; When the floating plume probability exceeds the predetermined value and the rising plume probability is equal to or less than the predetermined value, the navigation control information is generated to increase the rising plume probability. The navigation control device according to claim 9.

11. The control information generation unit The navigation control information is generated by reinforcement learning based on the mineral deposit evaluation data. The navigation control device according to claim 8.

Images