Intelligent underwater vehicle hot charging method

Abstract

The application belongs to the technical field of underwater equipment and artificial intelligence, and discloses a kind of intelligent underwater submersible heat charging method, to solve the problems of low charging efficiency, underwater navigation difficulty and insufficient safety guarantee of existing underwater submersible. The method is long-distance positioning to underwater heat source and underwater volcano through GPS signal when sailing on water surface, mid-distance positioning is carried out through inertial navigation, sonar signal and charging capacity of thermal power generation plate during diving process, precise short-distance positioning is carried out through ultra-short baseline positioning system, inertial navigation system, doppler sonar speed measurement system and charging capacity of thermal power generation plate at the position of underwater heat source and underwater volcano during slow sailing charging process, the optimal path of obstacle avoidance energy consumption is planned based on cost function, and the thermal power generation plate converts the thermal energy of underwater heat source into electric energy to charge the battery. The application realizes accurate perception of underwater thermal environment, significantly improves underwater navigation efficiency and charging safety, and prolongs the endurance time of submersible.

Description

Technical Field

[0001] This invention relates to the field of underwater equipment combined with artificial intelligence technology, specifically a thermal charging method for intelligent underwater vehicles. Background Technology

[0002] Against the backdrop of rapid development in marine development and exploration technologies, intelligent underwater vehicles (UUVs) play an irreplaceable role in deep-sea scientific research, seabed resource exploration, marine environmental monitoring, and underwater engineering inspection. These devices need to complete long-duration, large-scale operations in complex underwater thermal environments without human intervention; their endurance and charging efficiency directly determine the continuity and effectiveness of their missions. The energy supply of intelligent UUVs relies on the electrical energy stored in their batteries, and the charging efficiency and stability of these batteries are crucial for ensuring continuous operation. CN120739641A discloses a wave energy-based charging system for unmanned underwater vehicles; CN120414933A discloses an autonomous wireless charging location matching method, charging module, and system for underwater robots; and CN120364096A discloses a biomimetic pufferfish robotic fish with intelligent automatic charging and an intelligent charging method.

[0003] Underwater hydrothermal vents, underwater volcanoes, and underwater heat flows contain virtually limitless energy resources, offering new avenues for thermal charging of underwater vehicles. However, the unique characteristics of the underwater thermal environment pose multiple challenges to the charging process: thermal power generation is dynamically affected by factors such as the underwater thermal environment, volcanoes, and water depth, leading to significant fluctuations in the energy conversion efficiency of thermal power generation panels; complex underwater terrain and variable currents make it difficult for underwater vehicles to accurately locate and reach suitable charging areas; simultaneously, the timing of charging must be coordinated with mission progress and energy consumption levels, otherwise, mission interruptions due to untimely charging or increased operating costs due to frequent charging may occur. Therefore, how to achieve efficient, stable, and intelligent charging in the dynamically changing underwater thermal environment has become a core issue that urgently needs to be addressed in the development of intelligent underwater vehicle technology.

[0004] The current charging process for intelligent underwater vehicles faces several bottlenecks: First, locating heat sources underwater is difficult due to poor visibility and complex terrain; underwater charging efficiency is low, and environmental perception and feature extraction are inaccurate. Data collection using only a single sensor is insufficient to comprehensively capture key factors such as marine life, obstacles, and water flow speed, and the lack of in-depth data mining results in inadequate characterization of the underwater thermal environment. Second, underwater heat sources are often located near underwater volcanoes, posing a certain threat to the safety of underwater vehicles; and the scientific basis for predicting charging time and location is insufficient, failing to establish a quantitative model by combining historical charging data with real-time environmental parameters. The evaluation model cannot accurately select the optimal charging time and space combination, which is prone to "charging too early" or "insufficient power". Third, underwater path planning and navigation are more difficult, and there is a lack of reference points. The stability of path planning and charging docking is poor. The ability to adapt to ocean currents and obstacles during navigation is weak. The low alignment accuracy between the deployment of the thermal power generation plate and the charging interface may lead to charging interruption. Fourth, the charging process lacks dynamic monitoring and intelligent adjustment. It is impossible to optimize the charging strategy in real time according to the battery status and the efficiency of the thermal power generation plate. There are safety hazards such as overcharging and overheating, and the ability to cope with sudden environmental changes is insufficient. Summary of the Invention

[0005] The purpose of this invention is to provide a thermal charging method for intelligent underwater vehicles to solve the problems mentioned in the background art.

[0006] A method for hot-charging intelligent underwater vehicles includes the following steps:

[0007] Step 1: When the underwater vehicle is navigating on the surface, it uses GPS signals to locate the underwater heat source and the underwater volcano over long distances. On the surface, it uses the shortest path algorithm to navigate to the surface above the underwater heat source and the underwater volcano. The intelligent chip uses CNN and GRURNN models to extract 256-dimensional underwater thermal environment feature vectors. Step 2: During the descent, the underwater vehicle uses inertial navigation, sonar signals, and the charging capability of the thermal power plant to locate the underwater heat source and underwater volcano at a medium distance. It then uses a shortest path algorithm to navigate from the surface to the location of the underwater heat source and underwater volcano. The intelligent chip, based on GRU, constructs an attention mechanism model to learn from inertial navigation, sonar signals, and the charging capability of the thermal power plant to assist in correcting the shortest path algorithm during the descent. Combined with a matrix factorization collaborative filtering algorithm, a time-location scoring matrix is ​​generated to predict the optimal charging time window and three-dimensional geographic coordinates. Step 3: The underwater vehicle enters slow-speed charging mode. It uses an ultra-short baseline positioning system, inertial navigation system, Doppler sonar velocity measurement system, and the charging capability of the thermal power generation plate to accurately locate the underwater heat source and underwater volcano. Based on a cost function, it plans its route, avoiding the underwater volcano and heading to the target charging location. The thermal power generation plate converts the thermal energy of the underwater heat source into electrical energy to charge the battery. The intelligent chip monitors the environment and charging parameters in real time, and has fault warning and emergency evacuation functions.

[0008] Step 1 above includes the following steps: Step 1-1: When the underwater vehicle is navigating on the water surface, it collects the longitude and latitude coordinates of its current position through the GPS positioning module and obtains the depth of the underwater vehicle's current position through the depth table. Together, they form the three-dimensional coordinates of the underwater vehicle, namely longitude, latitude, and depth. Steps 1-2: While navigating on the surface, the underwater vehicle searches for maps of underwater thermal environments and initially locates the GPS positions of suitable underwater thermal environments for charging. It performs long-range positioning of underwater heat sources and underwater volcanoes. The intelligent chip uses the shortest path algorithm to travel from the current position in Step 1-1 to the location above the water surface of suitable underwater thermal environments for charging, i.e., above the water surface of underwater heat sources and underwater volcanoes. During the journey, it collects environmental data through cameras and sonar, avoiding marine life and obstacles. Steps 1-3: When the underwater vehicle navigates to the surface above the underwater heat source and underwater volcano, it activates the camera and sonar to collect environmental data, the depth gauge to collect depth data of the current location, and the thermal power generation plate to collect underwater thermal environment data, generating a 3D image dataset of the GPS position from the underwater vehicle to the suitable underwater thermal environment area for charging. Before diving, the underwater vehicle confirms the accuracy of the GPS signal of the current location and the GPS position of the suitable underwater thermal environment area for charging in the underwater thermal environment area map, and checks the accuracy of the position of the 3D image dataset and the GPS signal. If the accuracy meets the requirements, it proceeds to step 1-4 to begin preparing for the diving process in step 2. If the accuracy does not meet the requirements, it continues to navigate on the surface and repeats steps 1-1 and 1-2. Steps 1-4: Before diving, the underwater vehicle runs an artificial intelligence algorithm to extract underwater thermal environment features at its current location; preferably, the intelligent chip runs a deep learning algorithm to extract underwater thermal environment features, including underwater temperature, depth, marine life, and obstacles; the chip uses CNN and GRURNN models to process the underwater thermal environment data and generate a 256-dimensional feature vector. The intelligent chip first adjusts the 3D image dataset to a size of 224×224 and normalizes it to [0,1]. It then filters underwater temperature, depth, marine life, and obstacle features by using the maximum information coefficient (MIC=0.8). The chip extracts 128-dimensional image features through ResNet-18 and inputs 10-step time series data into GRU to generate 64-dimensional time series features. Finally, it combines these features with an attention mechanism to weightedly fuse them into a 256-dimensional comprehensive feature vector.

[0009] The aforementioned intelligent chip verifies the effectiveness of features using mean square error and evaluates generalization ability using cross-validation, as shown in the following formulas: ; in, This represents the actual value of the underwater thermal environment at time step T+1. For the true value, For predicted values, T For the sample size, The parameters representing all three steps can be learned through backpropagation using the Adam optimizer;

[0010] in Indicates the first A realistic underwater thermal environment Indicates the first The predicted underwater thermal environment, where T is the prediction period.

[0011] Step 2 above includes the following steps: Step 2-1: The underwater vehicle uses artificial intelligence algorithms to learn habitual navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the charging capacity of the thermal power plate. It performs mid-range positioning of underwater heat sources and underwater volcanoes, runs the shortest path algorithm to find the shortest path from the surface to the GPS position of the underwater thermal environment area suitable for charging in Step 1-2, and calculates the underwater thermal environment attention at the current position. The intelligent chip is based on a deep learning framework and has built an attention mechanism model for underwater thermal environments. By mining the temporal features of inertial navigation and sonar signals and weighting key information, it can achieve efficient representation and navigation of complex underwater environmental features without GPS signals. This process adopts a neural network architecture based on gated recurrent units, and gradually extracts the core elements in the environmental data through multi-level calculations using inertial navigation and sonar signals. Step 2-2: The underwater vehicle calculates the charging capacity of the thermal power generation plate in the current underwater thermal environment to help correct the underwater thermal environment attention of the inertial navigation and sonar signals and the current position; preferably, when the inertial navigation and sonar signals deviate, the closer the underwater vehicle is to the underwater heat source and the underwater volcano, the stronger the charging capacity of the thermal power generation plate, and vice versa, so as to calculate the deviation of the underwater vehicle from the GPS position of the underwater thermal environment area suitable for charging in Step 1-2; Power is calculated based on the intensity of thermal power generation and the efficiency of the thermal power generation panel, using the following formula:

[0012] Where I represents thermal power generation intensity, and A represents area. To improve efficiency, charging capacity data is generated. This calculation process takes into account the effects of underwater thermal power generation intensity decay and the cleanliness of the thermal power panel surface. Accurate charging capacity data is generated through real-time data calibration, providing a quantitative basis for energy supply side for charging decisions.

[0013] Step 2-3: Based on the timing characteristics of the inertial navigation and sonar signals from Step 2-1, and the charging capacity of the thermal power panels in the current underwater thermal environment (Step 2-2), the underwater submersible uses an artificial intelligence algorithm to correct the shortest path during the descent process in Step 2-1. It uses the shortest path to the area with stronger charging capacity near the thermal power panels, calculates a scoring matrix for underwater thermal environment charging time and charging location, updates the 3D image dataset of the GPS position from the underwater submersible to the suitable underwater thermal environment area for charging, and provides the shortest path for the descent process within the 3D image dataset. The matrix factorization collaborative filtering algorithm described above constructs an initial time-location score matrix, fills missing values ​​with the average location score, calculates Pearson similarity to generate a similarity matrix, obtains a k-dimensional latent factor matrix through SVD decomposition, calculates the predicted score through inner product, and generates a comprehensive score by combining POI weighted correction. Steps 2-4: The underwater vehicle adjusts its direction and speed according to the shortest path of the descent process as corrected in Step 2-3, and heads to the GPS location of the underwater thermal environment area suitable for charging in Step 1-2. It continuously calculates the scoring matrix of the optimal charging time and charging location in the underwater thermal environment and updates the optimal decision.

[0014] The above POI weighted adjustment includes: via the formula:

[0015] Calculate the initial score for each POI, where b is the index of the surrounding POI category, a is an integer between 0 and 18, x is the number of POIs of the corresponding category, and W is the weighting coefficient for that category of POIs; then use the formula:

[0016] Make corrections. Average POI score for charging time and charging location in all underwater thermal environments.

[0017] After the above SVD decomposition, the time latent factor matrix is ​​calculated using the following formula:

[0018] U is the time latent factor matrix of the underwater thermal environment, with each row representing a time slice. k Feature vectors in the latent space. S It is m A diagonal matrix of n, representing R Singular values ​​of decomposition The location latent factor matrix is ​​calculated using the following formula:

[0019] Where V is the latent factor matrix of the underwater thermal environment location, and each row corresponds to the latent space feature vector of a candidate location. S It is m A diagonal matrix of n, representing R Singular values ​​of the decomposition.

[0020] The aforementioned intelligent chip introduces Dropout regularization and L2 regularization during model training to prevent overfitting and improve the model's adaptability to the dynamic underwater thermal environment.

[0021] Step 3 above includes the following steps: Step 3-1: The underwater vehicle reaches the GPS location of the suitable underwater thermal environment area for charging as described in Step 1-2, maintaining a slow speed for charging so that it can quickly avoid danger without restarting. The smart chip reads the optimal underwater thermal environment charging time and charging location from sub-step 2-4. The smart chip converts the finally determined optimal underwater thermal environment charging time and charging location pair into a standard navigation format, generates a visual navigation path preview through a greedy algorithm, and initiates a time synchronization mechanism to ensure that it proceeds to charging as planned.

[0022] Step 3-2: The intelligent chip performs precise short-range positioning using the angle of arrival and time difference of the beacon in the ultra-short baseline positioning system, the acceleration and angular velocity of the inertial navigation system, the Doppler sonar velocity measurement system that emits sound waves to the seabed and measures the echo Doppler frequency shift, and the charging capacity of the thermal power generation panel. Based on the current location, the underwater heat source and underwater volcano changes, surrounding marine life, obstacles, and remaining power, it plans a collision avoidance driving path to avoid underwater volcanic areas where the charging capacity of the thermal power generation panel is too strong. The chip uses artificial intelligence algorithms to generate a path by combining power and distance, stores the path point set, and provides slow-speed charging collision avoidance driving path prompts in the 3D image dataset. The cost function is as follows:

[0023] middle, To determine the actual navigation distance from the starting point to the current point, avoiding surrounding marine life and obstacles. This represents the estimated distance from the current node to the target. This is the weighting factor for electricity consumption. The percentage of remaining battery power when reaching node n; Step 3-3: The underwater vehicle travels along the planned route to the optimal charging location for charging. The thermal power generation panel converts the thermal energy of the underwater heat source into electrical energy to charge the battery. The environment and charging parameters are monitored in real time. If the underwater vehicle issues a malfunction warning or gets too close to the volcano, it will evacuate in an emergency.

[0024] In step 2 above, the intelligent chip optimizes the scoring matrix parameters using the Adam optimizer. The update gate weight formula for GRU is:

[0025] in, To update the door, It is the sigmoid activation function. This is the weight matrix, applied to the current input. , The mapping between input features and hidden states is learned through training; This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t 1's hidden state The formula for resetting the door weight is:

[0026] in, To reset the door, It is the sigmoid activation function. This is the weight matrix, applied to the current input. , The mapping between input features and hidden states is learned through training. This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t The hidden state of 1.

[0027] When the optimal charging time and location are output in step 2 above, the top 5 suboptimal solutions are backed up simultaneously. The optimality of the decision is verified by precision, recall, and MAPE. The MAPE calculation formula is:

[0028] in Indicates the first A realistic underwater thermal environment Indicates the first The underwater thermal environment for each charging prediction is given, where T is the total number of charging prediction cycles.

[0029] Compared with the prior art, the beneficial effects of the present invention are: 1. By employing different positioning methods in three stages—surface navigation, diving process, and slow-speed charging—underwater heat sources are accurately identified, resulting in high underwater charging efficiency. Thermal power generation is utilized, with thermal power panels directly converting heat energy into electrical energy, minimizing conversion losses. This invention enhances environmental perception capabilities through multi-sensor fusion and deep learning algorithms. It utilizes GPS, cameras, and depth gauges to collect location, images, and underwater thermal environment data, which are then processed by CNN and GRURNN models to generate 256-dimensional feature vectors, accurately extracting key features such as thermal power generation and volcanoes. 2. Precise short-range positioning is achieved through an ultra-short baseline positioning system, an inertial navigation system, and a Doppler sonar velocity measurement system, accurately avoiding underwater volcanoes and making underwater charging safer; an intelligent charging decision model is constructed, which combines an attention mechanism to calculate environmental attention weights, generates a time-location score matrix through a matrix factorization collaborative filtering algorithm, and integrates the charging capacity of the thermal power generation panel with POI weighted data to accurately predict the optimal charging time and location; 3. Intelligent underwater navigation with high positioning accuracy; the intelligent chip is based on GRU to build an attention mechanism model to learn habitual navigation and sonar signals, optimize path planning and charging docking process, design navigation path based on cost function, and combine PID control and ocean current compensation technology to ensure navigation accuracy. The robotic arm automatically adjusts the attitude of the thermal power generation plate to achieve efficient docking and charging. 4. Optimize underwater path planning to improve operational efficiency; introduce full-process monitoring and safety mechanisms. The intelligent chip uses CNN and GRURNN models to extract 256-dimensional underwater thermal environment feature vectors, monitors battery voltage, current, temperature and environmental parameters in real time, and realizes overcharge protection, fault warning and dynamic strategy adjustment through the intelligent chip, which significantly improves charging efficiency, stability and safety, and ensures that the underwater vehicle can perform underwater missions for a long time with high precision. Attached Figure Description

[0030] Figure 1 This is a schematic diagram illustrating the operation of the underwater vehicle thermal charging method provided in an embodiment of the present invention.

[0031] Figure 2 The system flowchart of the underwater vehicle thermal charging method provided in the embodiment of the present invention is shown. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figure 1 and Figure 2 This invention provides a method for hot-charging an intelligent underwater vehicle, comprising an underwater vehicle, an intelligent chip, a camera, a sonar, a GPS positioning module, a depth gauge, a thermal power generation plate, and a battery. The underwater vehicle serves as the main body, providing a stable installation platform and electrical connection for the other modules. The camera and sonar are installed inside the underwater vehicle, as is the intelligent chip, which is connected to each module to achieve intelligent control of the underwater vehicle. It receives ground control commands, adjusts its navigation attitude based on GPS positioning module location information, monitors battery power status, controls the charging process, and has fault diagnosis and early warning functions, improving the safety and reliability of the underwater vehicle. The camera acquires underwater image information in real time. The sonar acquires underwater acoustic information in real time. The GPS positioning module provides accurate longitude and latitude coordinates for the underwater vehicle's positioning, transmits the location information to the intelligent chip, and assists in planning the navigation path and determining the relative position to charging facilities during charging. A depth gauge calculates the depth of the underwater vehicle by detecting underwater pressure and converting it into a corresponding depth value. This value, along with the longitude and latitude coordinates from the GPS positioning module, forms the underwater vehicle's three-dimensional coordinates. A thermoelectric generator plate, installed on the outer surface of the underwater vehicle, converts thermal energy into electrical energy to charge the battery. It features intelligent charging management, automatically adjusting the charging current and voltage based on the battery's charge level. The battery stores electrical energy to power the underwater vehicle's navigation.

[0034] Implementation steps: Step 1: The underwater vehicle uses the GPS positioning module to initially locate itself and head to a suitable underwater thermal environment area for charging. It then collects and learns underwater thermal environment data through multiple sensors. Step 2: The underwater vehicle further searches for the optimal thermal charging location in the underwater thermal environment area suitable for charging. Through multi-sensor acquisition and learning, it performs multi-dimensional data modeling and dynamic analysis of the underwater thermal environment to accurately predict the optimal charging time window and three-dimensional geographic coordinates. Step 3: The underwater vehicle proceeds to the charging location according to the optimal charging time window and thermal charging location. The thermal power generation plate converts the thermal energy of the underwater heat source into electrical energy to charge the battery. Step 1 involves the following specific steps: Step 1-1: When the underwater vehicle is navigating on the water surface, it collects the longitude and latitude coordinates of its current position through the GPS positioning module and obtains the depth of the underwater vehicle's current position through the depth table. Together, they form the three-dimensional coordinates of the underwater vehicle, namely longitude, latitude, and depth. Steps 1-2: While navigating on the surface, the underwater vehicle searches for maps of underwater thermal environments and initially locates the GPS positions of suitable underwater thermal environments for charging. It performs long-range positioning of underwater heat sources and underwater volcanoes. The intelligent chip uses the shortest path algorithm to travel from the current position in Step 1-1 to the location above the water surface of suitable underwater thermal environments for charging, i.e., above the water surface of underwater heat sources and underwater volcanoes. During the journey, it collects environmental data through cameras and sonar, avoiding marine life and obstacles. Steps 1-3: When the underwater vehicle navigates to the surface above the underwater heat source and underwater volcano, it activates the camera and sonar to collect environmental data, the depth gauge to collect depth data of the current location, and the thermal power generation plate to collect underwater thermal environment data, generating a 3D image dataset of the GPS position from the underwater vehicle to the suitable underwater thermal environment area for charging. Before diving, the underwater vehicle confirms the accuracy of the GPS signal of the current location and the GPS position of the suitable underwater thermal environment area for charging in the underwater thermal environment area map, and checks the accuracy of the position of the 3D image dataset and the GPS signal. If the accuracy meets the requirements, it proceeds to step 1-4 to begin preparing for the diving process in step 2. If the accuracy does not meet the requirements, it continues to navigate on the surface and repeats steps 1-1 and 1-2. Steps 1-4: Before diving, the underwater vehicle runs an artificial intelligence algorithm to extract underwater thermal environment features at its current location; preferably, the intelligent chip runs a deep learning algorithm to extract underwater thermal environment features, including underwater temperature, depth, marine life, and obstacles; the chip uses CNN and GRURNN models to process the underwater thermal environment data and generate a 256-dimensional feature vector.

[0035] Sub-step 1-4-1: The smart chip preprocesses the three-dimensional image dataset from step 1-3, including underwater temperature, depth, marine life and obstacles, adjusts the size to 224×224, and normalizes the pixels to [0,1] using the following formula to generate a normalized image of the underwater thermal environment.

[0036] For sequence probability, is the pixel value, T is the time series length, and sequence decomposition is used for normalization.

[0037] Sub-step 1-4-2: The smart chip uses the maximum information coefficient to filter relevant features of the underwater thermal environment data; the chip calculates the MIC, such as underwater temperature, depth, marine life and obstacles, MIC=0.8, and generates the filtered dataset; Sub-steps 1-4-3: The intelligent chip uses CNN to extract features from the 3D image dataset in step 1-3, runs ResNet-18 (18 layers), generates a 128-dimensional feature map, and generates underwater thermal environment image features using the following formula;

[0038] In hidden state, Here is a parameter, and T is the length of the time series. Output feature probabilities; Sub-step 1-4-4: The smart chip uses GRURNN to process the underwater thermal environment data sequence. The chip inputs 10 steps of underwater thermal environment data and calculates 64-dimensional time series features using the following formula.

[0039] in, For the activation of the GRU at time t, To update the door, It is the previous activation and candidate activation Linear interpolation between; Sub-steps 1-4-5: The smart chip calculates the GRU update gate weights using the following formulas to optimize the time-series modeling of underwater thermal environment data.

[0040] in, To update the door, It is the sigmoid activation function. This is the weight matrix, applied to the current input. , The mapping between input features and hidden states is learned through training; This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t The hidden state of 1.

[0041] In sub-steps 1-4-6, the smart chip calculates GRU candidate activations using the following formula to enhance the representation of underwater thermal environment features.

[0042]

[0043] in, tanh is the reset gate, and tanh is the activation function. This represents element-wise multiplication. When... When the value is close to 0, the reset gate effectively makes the cell behave as if it is reading the first symbol of the input sequence, allowing it to forget the previously computed state; Sub-steps 1-4-7: The intelligent chip calculates the GRU reset gate weight using the following formula, optimizes the underwater thermal environment memory mechanism, and resets the gate. The calculation process is similar to that of the update gate:

[0044] in, To reset the door, It is the sigmoid activation function. This is the weight matrix, applied to the current input. , The mapping between input features and hidden states is learned through training. This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t The hidden state of 1.

[0045] Sub-steps 1-4-8: The intelligent chip uses an attention mechanism to weight key features and generates underwater thermal environment attention weights using the following formula to highlight key features.

[0046] in yes Input, yes The output of .

[0047] Sub-steps 1-4-9: The intelligent chip integrates underwater thermal environment images and temporal features, merges features through a fully connected layer, generates a 256-dimensional vector, and generates comprehensive underwater thermal environment features. Sub-steps 1-4-10: The smart chip evaluates the quality of underwater thermal environment characteristics through mean square error (MSE) and verifies the effectiveness of the characteristics;

[0048] in, This represents the actual value of the underwater thermal environment at time step T+1. For the true value, For predicted values, T The number of samples; The parameters representing all three steps can be learned through backpropagation using the Adam optimizer; Sub-step 1-4-11: The smart chip evaluates its generalization ability in underwater thermal environments using the following cross-validation formula;

[0049] in Indicates the first A realistic underwater thermal environment Indicates the first The predicted underwater thermal environment, where T is the prediction period.

[0050] In sub-steps 1-4-12, the smart chip outputs the final feature vector (256-dimensional) of the underwater thermal environment, including underwater temperature, depth, marine life, and obstacles, for subsequent prediction.

[0051] Step 2 involves the following specific steps: Step 2-1: The underwater vehicle uses artificial intelligence algorithms to learn habitual navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the charging capacity of the thermal power plate. It performs mid-range positioning of underwater heat sources and underwater volcanoes, runs the shortest path algorithm to find the shortest path from the surface to the GPS position of the underwater thermal environment area suitable for charging in Step 1-2, and calculates the underwater thermal environment attention at the current position. The intelligent chip is based on a deep learning framework and has built an attention mechanism model for underwater thermal environments. By mining the temporal features of inertial navigation and sonar signals and weighting key information, it can achieve efficient representation and navigation of complex underwater environmental features without GPS signals. This process adopts a neural network architecture based on gated recurrent units (GRUs) and inertial navigation and sonar signals, and gradually extracts the core elements in the environmental data through multi-level calculations. Sub-step 2-1-1: The underwater vehicle acquires inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the charging capacity of the thermal power generation plate. The intelligent chip first collects environmental data in real time from the inertial navigation and sonar sensor array and the thermal power generation plate, including the underwater vehicle's acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, underwater temperature, depth, longitude, latitude, marine life, obstacles, salinity, water flow velocity, historical charging records, and multi-dimensional parameters of the thermal power generation plate's charging status. After standardization, these parameters are integrated into a 256-dimensional high-dimensional feature vector, which serves as the initial input to the model. This feature vector covers key environmental indicators affecting charging efficiency, providing a data foundation for subsequent time-series analysis.

[0052] Sub-step 2-1-2: Configure the smart chip with 128 GRU units, initialize GRURNN, and prepare for underwater thermal environment time series modeling.

[0053] in, In hidden state, It is a nonlinear function responsible for fusing and transforming the input characteristics of inertial navigation acceleration and angular velocity, the angle of arrival and time difference of sonar signals to underwater vehicle beacons, and the charging capacity of thermal power panels with historical states. This initialization process lays the computational foundation for the dynamic modeling of subsequent time series data.

[0054] Sub-step 2-1-3: The smart chip calculates the GRU update gate using the following formula to control the update of underwater thermal environment information;

[0055] in, To update the door, It is the sigmoid activation function. This is the weight matrix, applied to the current input. , By training habitual navigation acceleration and angular velocity, the angle of arrival and time difference of sonar signals to underwater vehicle beacons, and the mapping between input characteristics of the charging capability of thermal power panels and the hidden state. This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t The hidden state of 1. The core function of the update gate is to determine how much information should be retained from the historical state of inertial navigation acceleration and angular velocity, sonar signals to the angle of arrival and time difference of the underwater vehicle beacon, and the charging capacity of the thermal power generation panel, as well as how much new input feature should be included, so as to achieve dynamic control of the information flow.

[0056] Sub-step 2-1-4: The smart chip calculates the GRU reset gate using the following formula to control the retention of underwater thermal environment history.

[0057] in, The reset gate weights are responsible for controlling the impact of the historical hidden state on the current inertial navigation acceleration and angular velocity, the angle of arrival and time difference of the sonar signal to the underwater vehicle beacon, and the charging capacity of the thermal power generation plate. When the reset gate value is close to 0, the model will forget some historical information of the underwater thermal environment, thus focusing more on the current input features and improving its adaptability to dynamic environments.

[0058] Sub-step 2-1-5: The smart chip calculates candidate activations using the following formula to generate a new state of the underwater thermal environment; ; in, The parameters include inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the candidate states of the thermal power generation panel's charging capability. Indicates a set of reset doors. Indicates multiplication between elements. This is the weight matrix, applied to the current input. , The mapping between input features of the underwater thermal environment and the hidden state is learned through training. This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t The hidden state of 1. This candidate state integrates the current inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, the charging capability input features of the thermal power panel, and the historical state filtered by the reset gate. After activation by the tanh function, a new state representation is generated, providing alternative information for the hidden state update.

[0059] Sub-step 2-1-6: The smart chip updates the hidden state using the following formula to complete the time-series modeling of the underwater thermal environment; ; Among them, the update door Determines the degree of activation for each cell update. When As the value approaches 1, the model relies more on the newly generated inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the candidate states of the thermal power generation panel's charging capability; when As the velocity approaches zero, a significant amount of historical information is preserved, including inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the charging capacity of the thermal power generation panels. This mechanism enables the model to adaptively handle changes in underwater thermal environment characteristics at different time scales.

[0060] Sub-step 2-1-7: The intelligent chip calculates the attention weight using the following formula to highlight the key features of the underwater thermal environment;

[0061] in yes Input, yes The output of this process, through weighted aggregation of inertial navigation acceleration and angular velocity, the angle of arrival and time difference of sonar signals to underwater vehicle beacons, and the time series characteristics of the charging capacity of the thermal power generation plate, enables the model to automatically identify underwater thermal environment indicators that have a significant impact on charging decisions.

[0062] In sub-step 2-1-8, the smart chip evaluates the underwater thermal environment weight quality through MSE and verifies the effectiveness of the weight using the following formula.

[0063]

[0064] in, This represents the inertial navigation acceleration and angular velocity at time step T+1, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the actual value of the charging capacity of the thermal power generation panel. For the true value, For predicted values, T This represents the number of samples. The parameters for all three steps can be learned through backpropagation using the Adam optimizer.

[0065] Sub-step 2-1-9, to enhance the model's generalization ability, introduces a Dropout regularization mechanism during training, randomly discarding neurons in the neural network with a probability of 0.3 to prevent overfitting. This operation forces the model to learn more robust feature representations, avoiding overfitting to training data on inertial navigation acceleration and angular velocity, the angle of arrival and time difference between sonar signals and underwater vehicle beacons, and the charging capability of the thermal power plant.

[0066] In sub-step 2-1-10, the intelligent chip outputs a 256-dimensional attention weight vector for subsequent prediction. Each element in this vector corresponds to the inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the importance score of the initial input features related to the charging capability of the thermal power plant. Features with higher weight values ​​in the underwater thermal environment have a greater impact on the charging decision, providing a crucial feature importance distribution for subsequent predictions of charging time and location.

[0067] Step 2-2: The underwater vehicle calculates the charging capacity of the thermal power generation plate in the current underwater thermal environment to help correct the underwater thermal environment attention of the inertial navigation and sonar signals and the current position; preferably, when the inertial navigation and sonar signals deviate, the closer the underwater vehicle is to the underwater heat source and the underwater volcano, the stronger the charging capacity of the thermal power generation plate, and vice versa, so as to calculate the deviation of the underwater vehicle from the GPS position of the underwater thermal environment area suitable for charging in Step 1-2; Step 2-3: Based on the timing characteristics of the inertial navigation and sonar signals in Step 2-1 and the charging capacity of the thermal power plant in the current underwater thermal environment in Step 2-2, the underwater vehicle runs an artificial intelligence algorithm to correct the shortest path of the descent process in Step 2-1. It uses the shortest path to approach the area with stronger charging capacity of the thermal power plant, calculates the scoring matrix of underwater thermal environment charging time and charging location, updates the three-dimensional image dataset of GPS position from the underwater vehicle to the underwater thermal environment area suitable for charging, and prompts the shortest path of the descent process in the three-dimensional image dataset. Sub-step 2-3-1: The intelligent chip uses the matrix factorization collaborative filtering algorithm to calculate the scoring matrix of charging time and location based on the inertial navigation acceleration and angular velocity in step 2-1, the angle of arrival and time difference of the sonar signal to the underwater vehicle beacon, the charging capacity characteristics of the thermal power generation plate, and the charging capacity of the thermal power generation plate under the current underwater thermal environment in step 2-2, and generates the scoring matrix. The intelligent chip constructs a time-location scoring matrix, providing the basis for scoring and generating an initial matrix. It extracts charging records from a historical database of inertial navigation acceleration and angular velocity, the angle of arrival and time difference between sonar signals and underwater vehicle beacons, and the charging capacity of thermal power panels. Using time slices (e.g., one-hour units) and geographic locations (e.g., latitude, longitude, and depth grids) as V-indexes, it constructs the initial scoring matrix. R Matrix elements This represents the historical charging efficiency score at time t and position p (e.g., a maximum score of 5 points), forming a basic representation of the spatiotemporal charging pattern.

[0068] In sub-step 2-3-2, the smart chip fills the missing values ​​with the average score of the underwater thermal environment location to complete the matrix. This filling strategy is based on the assumption that "the charging suitability of the same location has a certain stability at different times", making the matrix structure more complete.

[0069] Sub-step 2-3-3: The intelligent chip calculates the Pearson similarity (time-position similarity), generates the inertial navigation acceleration and angular velocity, the angle of arrival and time difference of the sonar signal to the underwater vehicle beacon, and the charging capability similarity matrix of the thermal power generation panel;

[0070] in, The similarity matrix is ​​generated for the inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the charging capacity of the thermal power panel, where and Historical users underwater thermal environment The charging rating, and and User underwater thermal environment The average charging rating, This indicates that the user The underwater thermal environment was jointly rated. These are common scoring items.

[0071] Sub-steps 2-3-4 involve the intelligent chip setting hidden factor dimensions based on inertial navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, the historical data scale of the thermal power generation panel's charging capacity, and computational resource limitations. k (like k =32), this dimension determines the level of abstraction of spatiotemporal features by the model. The parameter matrix required for Singular Value Decomposition (SVD) is then initialized, including the user latent factor matrix. U Singular value matrix S and item latent factor matrix V This prepares for the dimensionality reduction decomposition of high-dimensional scoring matrices.

[0072] In sub-steps 2-3-5, the smart chip performs SVD decomposition on R using the following formula and generates an underwater thermal environment decomposition matrix.

[0073]

[0074] Where U is An orthogonal matrix, where m is the number of users; yes An orthogonal matrix, where n is the number of underwater thermal environment regions and the thermal charging location; S It is m A diagonal matrix of n, representing R Singular values ​​of the decomposition.

[0075] In sub-steps 2-3-6, the smart chip retains the first k singular values ​​of S, simplifies the matrix to k dimensions, and thus generates a low-dimensional matrix of the underwater thermal environment.

[0076] In sub-steps 2-3-7, the intelligent chip calculates the time latent factor of the underwater thermal environment using the following formula and generates the time latent factor.

[0077]

[0078] Where U is the underwater thermal environment time latent factor matrix, and each row represents a time slice in... k Feature vectors in the latent space. S It is m A diagonal matrix of n, representing R The singular values ​​of the decomposition. This operation, through singular value scaling, ensures that the length of the latent factor vector is consistent with the energy distribution of the original scoring matrix, enhancing the physical meaning of the underwater thermal environment characteristics representation.

[0079] In sub-steps 2-3-8, the intelligent chip calculates the underwater thermal environment location latent factor using the following formula and generates the location latent factor.

[0080]

[0081] Where V is the latent factor matrix of the underwater thermal environment location, and each row corresponds to the latent space feature vector of a candidate location. S It is m A diagonal matrix of n, representing R The singular values ​​of the decomposition. The time and location latent factor matrices together constitute a low-dimensional feature representation of the original scoring matrix, providing a computational basis for subsequent score prediction.

[0082] Sub-steps 2-3-9: The intelligent chip calculates the inner product and generates... The underwater thermal environment charging time and charging location prediction score are calculated using the following formula, and an underwater thermal environment charging time and charging location score matrix is ​​generated.

[0083]

[0084] in, , This involves generating latent factor vectors for the charging time and charging location of the underwater thermal environment. This operation maps the underwater thermal environment features from the low-dimensional latent space back to the original scoring space, generating a complete prediction scoring matrix for the underwater thermal environment's charging time and charging location. Each element This represents the predicted charging suitability score at time i and location j.

[0085] Sub-step 2-3-10: The intelligent chip applies POI weighted adjustment scoring, through calculation... This is used to generate underwater thermal environment charging time and charging location POI scores.

[0086]

[0087] in, It refers to the charging time in the underwater thermal environment and the number of POI types in the surrounding area. is the maximum interval of b. Since there are 18 types of POIs, a is an integer between 0 and 18. x is the number of a certain type of POI around the underwater thermal environment j. W is the weighting coefficient of the corresponding type of POI. This score reflects the impact of the surrounding environment on the convenience and safety of charging operations, and provides supplementary information on the spatial dimension of the scoring matrix.

[0088] Sub-step 2-3-11: The intelligent chip corrects the charging time and charging location POI score in the underwater thermal environment; the chip calculates... This is used to correct the underwater thermal environment charging time and charging location POI score.

[0089]

[0090] in, This represents the average POI score for all underwater thermal environment charging times and charging locations. This correction operation converts the POI score into a deviation value relative to the global average level, making the POI impact at different locations comparable and facilitating fusion with the original score matrix.

[0091] In sub-step 2-3-12, the smart chip calculates and evaluates the underwater thermal environment charging time and charging location score quality using the MSE formula, and verifies the validity of the score.

[0092]

[0093] in, Represents the i-th real underwater thermal environment. Indicates the first The underwater thermal environment is predicted for each charging cycle, where T is the total number of charging prediction cycles. The accuracy of the underwater thermal environment charging time and charging location scoring matrix is ​​verified by comparing the MSE value with a preset threshold. After successful verification, the shortest path for the descent process in step 2-1 is corrected. The scoring matrix for underwater thermal environment charging time and charging location is calculated, and the 3D image dataset of the GPS location from the underwater vehicle to a suitable underwater thermal environment area for charging is updated. The shortest path for the descent process is then indicated in the 3D image dataset.

[0094] Steps 2-4: The underwater vehicle adjusts its direction and speed according to the shortest path of the descent process as corrected in Step 2-3, and heads to the GPS location of the underwater thermal environment area suitable for charging in Step 1-2. It continuously calculates the scoring matrix of the optimal charging time and charging location in the underwater thermal environment and updates the optimal decision.

[0095] Sub-step 2-4-1: The intelligent chip adjusts its direction and speed according to the shortest path of the descent process corrected in step 2-3, and descends along this path, referring to the generated underwater thermal environment charging time and charging position prediction scoring matrix. This matrix contains underwater thermal environment charging time and charging location suitability scores based on historical data. It also loads basic location-related information, such as longitude, latitude, and depth, providing a data interface for subsequent multi-factor fusion.

[0096] In sub-step 2-4-2, during the descent, the intelligent chip continuously calculates the scoring matrix of the optimal charging time and charging location in the underwater thermal environment. It then obtains the charging capacity of the thermal power generation panel under the current underwater thermal environment (as in step 2-2) and maps the power to the score (e.g., 88W → 4.5), facilitating a comprehensive score of the underwater thermal environment's charging time and charging location. This mapping converts the energy supply capacity into a numerical scale compatible with spatiotemporal scoring, facilitating weighted fusion.

[0097] Sub-steps 2-4-3 involve the smart chip applying the corrected underwater thermal environment charging time and charging location POI score, and generating a weighted score using the following formula.

[0098]

[0099] in, A score for predicting charging time and charging location in underwater thermal environments with POI weighting. This is the original score. To correct the POI score, this weighted formula uses multi-factor fusion to ensure that the final score simultaneously reflects historical patterns, environmental characteristics, and real-time energy supply.

[0100] In sub-step 2-4-4, the smart chip optimizes the underwater thermal environment charging time and charging location scoring matrix through the Adam optimizer.

[0101]

[0102] in, This represents the actual value of the underwater thermal environment at time step T+1. For the true value, For predicted values, T This represents the number of samples. The parameters representing all three steps can be learned via backpropagation using the Adam optimizer to determine the underwater thermal environment charging time and location.

[0103] In sub-steps 2-4-5, the smart chip calculates the accuracy using the following formula and verifies the accuracy of charging time and charging location in the underwater thermal environment.

[0104]

[0105] Where P represents the underwater thermal environment charging time and charging location accuracy, T represents the number of data points with accurate predictions in the extracted samples, and n represents the total number of samples extracted from the overall dataset. This metric reflects the model's ability to accurately identify the optimal solution among candidate solutions; higher accuracy indicates stronger decision reliability.

[0106] In sub-steps 2-4-6, the smart chip calculates the recall rate using the following formula and verifies the charging time and charging location coverage in the underwater thermal environment.

[0107]

[0108] Here, R represents the underwater thermal environment charging time and charging location coverage, T represents the number of accurately predicted data extracted from the overall sample, and N represents the number of accurately predicted data in the entire dataset.

[0109] In sub-steps 2-4-7, the intelligent chip uses L2 regularization to prevent overfitting, thereby stabilizing the scoring of charging time and charging location in underwater thermal environments. This operation, by penalizing excessively large parameter values, forces the model to learn more generalized feature representations, improving the stability of charging time and charging location decisions in unknown underwater thermal environments.

[0110] In sub-steps 2-4-8, the intelligent chip globally sorts the fused underwater thermal environment charging time and charging location score matrix, arranging all time-location pairs in descending order of score to generate a candidate decision list. This list contains a ranking of charging suitability for different spatiotemporal combinations, providing a candidate set for selecting the final optimal solution, while retaining suboptimal solutions as backup options.

[0111] In sub-steps 2-4-9, the smart chip verifies the optimal score for charging time and location in the underwater thermal environment using MAPE, calculates the percentage error using the following formula, and thus determines the optimality, ensuring that the dive does not deviate from the course and accurately reaches the GPS location of the suitable underwater thermal environment area for charging in step 1-2.

[0112]

[0113] in Indicates the first A realistic underwater thermal environment Indicates the first The underwater thermal environment for each charging prediction is given, where T is the total number of charging prediction cycles.

[0114] In sub-steps 2-4-10, the smart chip outputs the highest-scoring underwater thermal environment charging time and charging location pair and stores the results, while backing up the top 5 suboptimal solutions so that the strategy can be quickly switched when encountering obstacles or changes in yaw.

[0115] Step 3, proceed with the following specific steps: Step 3-1: The underwater vehicle reaches the GPS location of the suitable underwater thermal environment area for charging as described in Step 1-2, maintaining a slow speed for charging so that it can quickly avoid danger without restarting. The smart chip reads the optimal underwater thermal environment charging time and charging location from sub-step 2-4. The smart chip converts the finally determined optimal underwater thermal environment charging time and charging location pair into a standard navigation format, generates a visual navigation path preview through a greedy algorithm, and initiates a time synchronization mechanism to ensure that it proceeds to charging as planned.

[0116] Step 3-2: The intelligent chip performs precise short-range positioning using the angle of arrival and time difference of the beacon in the ultra-short baseline positioning system, the acceleration and angular velocity of the inertial navigation system, the Doppler sonar velocity measurement system that emits sound waves to the seabed and measures the echo Doppler frequency shift, and the charging capacity of the thermal power generation panel. Based on the current location, the underwater heat source and underwater volcano changes, surrounding marine life, obstacles, and remaining power, it plans a collision avoidance driving path to avoid underwater volcanic areas where the charging capacity of the thermal power generation panel is too strong. The chip uses artificial intelligence algorithms to generate a path by combining the power (e.g., 50%) and distance, and stores the path point set. It also provides prompts for a slow-speed charging collision avoidance driving path in the 3D image dataset.

[0117] Sub-step 3-2-1: The intelligent chip acquires the angle of arrival and time difference of the beacon of the ultra-short baseline positioning system, the acceleration and angular velocity of the inertial navigation system, the Doppler sonar velocity measurement system emits sound waves to the bottom of the water and measures the echo Doppler frequency shift, and the charging capacity of the thermal power generation plate to perform accurate short-range positioning. The intelligent chip accurately measures the speed and displacement of the underwater vehicle relative to the bottom of the water, and the speed and displacement relative to the GPS position of the underwater thermal environment area suitable for charging in step 1-2. The cost function is designed based on the current position, surrounding marine life, obstacles and remaining power.

[0118] in, To determine the actual navigation distance from the starting point to the current point, avoiding surrounding marine life and obstacles. This is the estimated distance (Euclidean distance) from the current node to the target. The weighting factor for electricity consumption is 0.5. This represents the percentage of remaining battery power when reaching node n. This function prioritizes the path with lower energy consumption, based on the shortest path algorithm.

[0119] In sub-step 3-2-2, the smart chip reads the pre-stored 3D image dataset (10-meter resolution) and combines it with real-time sonar scanning data (500-meter detection range) to construct a dynamic environment model that includes obstacles and ocean current velocity fields, as well as a collision avoidance driving path to the GPS location of the underwater thermal environment area suitable for charging in step 1-2, ensuring that the collision avoidance driving path avoids dangerous areas, surrounding marine life, and obstacles.

[0120] In sub-step 3-2-3, the intelligent chip generates a path point set at 100-meter intervals based on the planning results. Each point contains information such as longitude, latitude coordinates, depth, and estimated arrival time, which is stored in the navigation cache for real-time control.

[0121] Step 3-3: The underwater vehicle travels along the planned route to the optimal charging location for charging. The thermal power generation panel converts the thermal energy of the underwater heat source into electrical energy to charge the battery. The environment and charging parameters are monitored in real time. If the underwater vehicle issues a malfunction warning or gets too close to the volcano, it will evacuate in an emergency. In sub-step 3-3-1, the underwater vehicle follows the collision avoidance path described in sub-step 3-2 to reach the GPS location of the suitable underwater thermal environment area for charging, as indicated in step 1-2. The intelligent chip collects the power generation voltage and current of the thermal power generation panel in real time, and uses a PID controller to generate the power generation voltage and current of the thermal power generation panel, ensuring voltage deviation... Meters, current deviation .

[0122] Sub-step 3-3-2: The intelligent chip detects and prevents the effects of overvoltage and overcurrent to ensure the safety of battery charging.

[0123] Sub-step 3-3-3: The underwater vehicle maintains a slow charging speed, preferably below 10 meters per minute; the smart chip monitors underwater temperature, depth, longitude, latitude, marine life, obstacles, salinity, and water flow speed in real time. When danger occurs, such as a volcanic eruption, attack by large marine life or a school of fish, the charging circuit is automatically cut off and an alarm is triggered, and the vehicle quickly leaves the water area without restarting.

[0124] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0125] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, equipment, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0126] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or equipment, and may be electrical, mechanical, or other forms.

[0127] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0128] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0129] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program instructions, such as USB flash drives, portable hard drives, read-only storage servers, random access storage servers, magnetic disks, or optical disks.

[0130] Furthermore, it should be noted that the combination of the various technical features in this case is not limited to the combination methods described in the claims of this case or the combination methods described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way, unless they contradict each other.

[0131] It should be noted that the above examples are merely specific embodiments of the present invention, and the present invention is obviously not limited to the above embodiments, with many similar variations. All modifications that can be directly derived or conceived by those skilled in the art from the content disclosed in this invention should fall within the protection scope of this invention.

[0132] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A thermal charging method for intelligent underwater vehicles, characterized in that, Includes the following steps: Step 1: When the underwater vehicle is navigating on the surface, it uses GPS signals to locate the underwater heat source and the underwater volcano over long distances. On the surface, it uses the shortest path algorithm to navigate to the surface above the underwater heat source and the underwater volcano. The intelligent chip uses CNN and GRURNN models to extract 256-dimensional underwater thermal environment feature vectors. Step 2: During the descent, the underwater vehicle uses inertial navigation, sonar signals, and the charging capability of the thermal power plant to locate the underwater heat source and underwater volcano at a medium distance. It then uses a shortest path algorithm to navigate from the surface to the location of the underwater heat source and underwater volcano. The intelligent chip, based on GRU, constructs an attention mechanism model to learn from inertial navigation, sonar signals, and the charging capability of the thermal power plant to assist in correcting the shortest path algorithm during the descent. Combined with a matrix factorization collaborative filtering algorithm, a time-location scoring matrix is ​​generated to predict the optimal charging time window and three-dimensional geographic coordinates. Step 3: The underwater vehicle enters slow-speed charging mode. It uses an ultra-short baseline positioning system, inertial navigation system, Doppler sonar velocity measurement system, and the charging capability of the thermal power generation plate to accurately locate the underwater heat source and underwater volcano. Based on a cost function, it plans its route, avoiding the underwater volcano and heading to the target charging location. The thermal power generation plate converts the thermal energy of the underwater heat source into electrical energy to charge the battery. The intelligent chip monitors the environment and charging parameters in real time, and has fault warning and emergency evacuation functions.

2. The method for thermal charging of an intelligent underwater vehicle according to claim 1, characterized in that, In step 1, Includes the following steps, Step 1-1: When the underwater vehicle is navigating on the water surface, it collects the longitude and latitude coordinates of its current position through the GPS positioning module and obtains the depth of the underwater vehicle's current position through the depth table. Together, they form the three-dimensional coordinates of the underwater vehicle, namely longitude, latitude, and depth. Steps 1-2: While navigating on the surface, the underwater vehicle searches for maps of underwater thermal environments and initially locates the GPS positions of suitable underwater thermal environments for charging. It performs long-range positioning of underwater heat sources and underwater volcanoes. The intelligent chip uses the shortest path algorithm to travel from the current position in Step 1-1 to the location above the water surface of suitable underwater thermal environments for charging, i.e., above the water surface of underwater heat sources and underwater volcanoes. During the journey, it collects environmental data through cameras and sonar, avoiding marine life and obstacles. Steps 1-3: When the underwater vehicle navigates to the surface above the underwater heat source and underwater volcano, it activates the camera and sonar to collect environmental data, the depth gauge to collect depth data of the current location, and the thermal power generation plate to collect underwater thermal environment data, generating a 3D image dataset of the GPS position from the underwater vehicle to the suitable underwater thermal environment area for charging. Before diving, the underwater vehicle confirms the accuracy of the GPS signal of the current location and the GPS position of the suitable underwater thermal environment area for charging in the underwater thermal environment area map, and checks the accuracy of the position of the 3D image dataset and the GPS signal. If the accuracy meets the requirements, it proceeds to step 1-4 to begin preparing for the diving process in step 2. If the accuracy does not meet the requirements, it continues to navigate on the surface and repeats steps 1-1 and 1-2. Steps 1-4: Before diving, the underwater vehicle runs an artificial intelligence algorithm to extract underwater thermal environment features at its current location; preferably, the intelligent chip runs a deep learning algorithm to extract underwater thermal environment features, including underwater temperature, depth, marine life, and obstacles; the chip uses CNN and GRURNN models to process the underwater thermal environment data and generate a 256-dimensional feature vector. The intelligent chip first adjusts the 3D image dataset to a size of 224×224 and normalizes it to [0,1]. It then filters underwater temperature, depth, marine life, and obstacle features by using the maximum information coefficient (MIC=0.8). The 128-dimensional image features are extracted by ResNet-18, and the 10-step time series data is input into GRU to generate 64-dimensional time series features. These features are then weighted and fused into a 256-dimensional comprehensive feature vector by combining the attention mechanism.

3. The method for thermal charging of an intelligent underwater vehicle according to claim 2, characterized in that, The intelligent chip verifies the effectiveness of features using mean squared error and evaluates generalization ability using cross-validation, with the following formulas: ； in, This represents the actual value of the underwater thermal environment at time step T+1. For the true value, For predicted values, T For the sample size, The parameters representing all three steps can be learned through backpropagation using the Adam optimizer; in Indicates the first A realistic underwater thermal environment Indicates the first The predicted underwater thermal environment, where T is the prediction period.

4. The method for thermal charging of an intelligent underwater vehicle according to claim 1, characterized in that, Step 2 includes the following steps: Step 2-1: The underwater vehicle uses artificial intelligence algorithms to learn habitual navigation acceleration and angular velocity, the angle of arrival and time difference between the sonar signal and the underwater vehicle beacon, and the charging capacity of the thermal power plate. It performs mid-range positioning of underwater heat sources and underwater volcanoes, runs the shortest path algorithm to find the shortest path from the surface to the GPS position of the underwater thermal environment area suitable for charging in Step 1-2, and calculates the underwater thermal environment attention at the current position. The intelligent chip is based on a deep learning framework and has built an attention mechanism model for underwater thermal environments. By mining the temporal features of inertial navigation and sonar signals and weighting key information, it can achieve efficient representation and navigation of complex underwater environmental features without GPS signals. This process adopts a neural network architecture based on gated recurrent units, and gradually extracts the core elements in the environmental data through multi-level calculations using inertial navigation and sonar signals. Step 2-2: The underwater vehicle calculates the charging capacity of the thermal power generation plate in the current underwater thermal environment to help correct the underwater thermal environment attention of the inertial navigation and sonar signals and the current position; preferably, when the inertial navigation and sonar signals deviate, the closer the underwater vehicle is to the underwater heat source and the underwater volcano, the stronger the charging capacity of the thermal power generation plate, and vice versa, so as to calculate the deviation of the underwater vehicle from the GPS position of the underwater thermal environment area suitable for charging in Step 1-2; Power is calculated based on the intensity of thermal power generation and the efficiency of the thermal power generation panel, using the following formula: Where I represents thermal power generation intensity, and A represents area. To improve efficiency, charging capacity data is generated. This calculation process takes into account the effects of underwater thermal power generation intensity decay and the cleanliness of the thermal power panel surface. Accurate charging capacity data is generated through real-time data calibration, providing a quantitative basis for energy supply side for charging decisions. Step 2-3: Based on the timing characteristics of the inertial navigation and sonar signals from Step 2-1, and the charging capacity of the thermal power panels in the current underwater thermal environment (Step 2-2), the underwater submersible uses an artificial intelligence algorithm to correct the shortest path during the descent process in Step 2-1. It uses the shortest path to the area with stronger charging capacity near the thermal power panels, calculates a scoring matrix for underwater thermal environment charging time and charging location, updates the 3D image dataset of the GPS position from the underwater submersible to the suitable underwater thermal environment area for charging, and provides the shortest path for the descent process within the 3D image dataset. The matrix factorization collaborative filtering algorithm constructs an initial time-location score matrix, fills missing values ​​with the average location score, calculates Pearson similarity to generate a similarity matrix, obtains a k-dimensional latent factor matrix through SVD decomposition, calculates the predicted score through inner product, and generates a comprehensive score by combining POI weighted correction. Steps 2-4: The underwater vehicle adjusts its direction and speed according to the shortest path of the descent process as corrected in Step 2-3, and heads to the GPS location of the underwater thermal environment area suitable for charging in Step 1-2. It continuously calculates the scoring matrix of the optimal charging time and charging location in the underwater thermal environment and updates the optimal decision.

5. The intelligent underwater vehicle thermal charging method according to claim 4, characterized in that, The POI weighted correction includes: via the formula: Calculate the initial score for each POI, where b is the index of the surrounding POI category, a is an integer between 0 and 18, x is the number of POIs of the corresponding category, and W is the weighting coefficient for that category of POIs; then use the formula: Make corrections. Average POI score for charging time and charging location in all underwater thermal environments.

6. The method for thermal charging of an intelligent underwater vehicle according to claim 4, characterized in that, After the SVD decomposition, the time latent factor matrix is ​​calculated using the following formula: U is the time latent factor matrix of the underwater thermal environment, with each row representing a time slice. k Feature vectors in the latent space. S It is m A diagonal matrix of n, representing R Singular values ​​of decomposition The location latent factor matrix is ​​calculated using the following formula: Where V is the latent factor matrix of the underwater thermal environment location, and each row corresponds to the latent space feature vector of a candidate location. S It is m A diagonal matrix of n, representing R Singular values ​​of the decomposition.

7. The method for thermal charging of an intelligent underwater vehicle according to claim 1, characterized in that, The intelligent chip incorporates Dropout regularization and L2 regularization during model training to prevent overfitting and improve the model's adaptability to underwater dynamic thermal environments.

8. The method for thermal charging of an intelligent underwater vehicle according to claim 1, characterized in that, In step 3 Includes the following steps, Step 3-1: The underwater vehicle reaches the GPS location of the suitable underwater thermal environment area for charging as described in Step 1-2, maintaining a slow speed for charging so that it can quickly avoid danger without restarting. The smart chip reads the optimal underwater thermal environment charging time and charging location from sub-step 2-4. The smart chip converts the finally determined optimal underwater thermal environment charging time and charging location pair into a standard navigation format, generates a visual navigation path preview through a greedy algorithm, and initiates a time synchronization mechanism to ensure that it proceeds to charging as planned. Step 3-2: The intelligent chip performs precise short-range positioning using the angle of arrival and time difference of the beacon in the ultra-short baseline positioning system, the acceleration and angular velocity of the inertial navigation system, the Doppler sonar velocity measurement system that emits sound waves to the seabed and measures the echo Doppler frequency shift, and the charging capacity of the thermal power generation panel. Based on the current location, the underwater heat source and underwater volcano changes, surrounding marine life, obstacles, and remaining power, it plans a collision avoidance driving path to avoid underwater volcanic areas where the charging capacity of the thermal power generation panel is too strong. The chip uses artificial intelligence algorithms to generate a path by combining power and distance, stores the path point set, and provides slow-speed charging collision avoidance driving path prompts in the 3D image dataset. The cost function is: middle, To determine the actual navigation distance from the starting point to the current point, avoiding surrounding marine life and obstacles. This represents the estimated distance from the current node to the target. This is the weighting factor for electricity consumption. The percentage of remaining battery power when reaching node n; Step 3-3: The underwater vehicle travels along the planned route to the optimal charging location for charging. The thermal power generation panel converts the thermal energy of the underwater heat source into electrical energy to charge the battery. The environment and charging parameters are monitored in real time. If the underwater vehicle issues a malfunction warning or gets too close to the volcano, it will evacuate in an emergency.

9. The method for thermal charging of an intelligent underwater vehicle according to claim 1, characterized in that, In step 2, the intelligent chip optimizes the scoring matrix parameters using the Adam optimizer. The update gate weight formula for GRU is: in, To update the door, It is the sigmoid activation function. This is the weight matrix, applied to the current input. , The mapping between input features and hidden states is learned through training; This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t 1's hidden state The formula for resetting the door weight is: in, To reset the door, It is the sigmoid activation function. This is the weight matrix, applied to the current input. , The mapping between input features and hidden states is learned through training. This is the weight matrix, applied to the hidden state at the previous time step. It is used to learn the dependencies between historical hidden states and the current state. For the previous time step t The hidden state of 1.

10. The method for thermal charging of an intelligent underwater vehicle according to claim 1, characterized in that, When outputting the optimal charging time and location in step 2, the top 5 suboptimal solutions are simultaneously backed up. The optimality of the decision is verified by precision, recall, and MAPE. The MAPE calculation formula is: in Indicates the first A realistic underwater thermal environment Indicates the first The underwater thermal environment for each charging prediction is given, where T is the total number of charging prediction cycles.

Images