A method and system for autonomous path planning of ROVs in deep-sea cold seep environments

Abstract

This application relates to an autonomous path planning method and system for ROVs in deep-sea cold seep environments, belonging to the field of deep-sea robot autonomous navigation and environmental detection technology. The method includes: acquiring multi-source data of a deep-sea cold seep area and preprocessing it to obtain preprocessed multi-source data; constructing a three-dimensional environmental data map of the deep-sea cold seep area based on the preprocessed multi-source data; obtaining multiple initial candidate paths through a candidate path generation algorithm based on the three-dimensional environmental data map and the real-time status parameters of the ROV; selecting the optimal path through a DQN algorithm based on the multiple initial candidate paths and generating an optimal path command; and transmitting the optimal path command to the ROV, enabling the ROV to navigate according to the optimal path. Compared with existing technologies, this application can accurately and autonomously plan paths to efficiently and accurately identify target sampling areas and improve ROV energy utilization. Furthermore, the method exhibits strong environmental dynamic adaptability, high detection efficiency, and high sampling accuracy.

Description

Technical Field

[0001] This application relates to the field of autonomous navigation and environmental detection technology for deep-sea robots, and in particular to an autonomous path planning method and system for ROVs in deep-sea cold seep environments. Background Technology

[0002] Deep-sea cold seeps, as unique ecosystems rich in chemicals such as methane and hydrogen sulfide, possess irreplaceable value for global climate research and marine resource exploration due to their distinctive biological communities and geological features. ROVs (Remotely Operated Vehicles), as the core equipment for deep-sea cold seep detection, directly determine detection efficiency, sampling quality, and energy utilization through their autonomous path planning capabilities.

[0003] Currently, existing path planning systems suffer from three major flaws. First, they lack the ability to integrate intelligent data specifically tailored to cold seep ecosystems. Existing solutions, such as two-layer path planning based on the ISSA algorithm and real-time planning using width reinforcement learning, focus solely on obstacle avoidance and path minimization, failing to incorporate deep learning algorithms for intelligent feature extraction from multi-dimensional environmental data such as biological distribution and chemical concentration. This makes it impossible to accurately identify high-value sampling areas. Second, their decision-making logic is simplistic and lacks physical constraints. Multi-sensor fusion navigation systems and dung beetle-based swarm planning prioritize obstacle avoidance or energy reduction, neglecting the value of sampling areas in deep-sea cold seeps and the energy constraints of ROVs, leading to energy waste and insufficient sampling targeting. Third, their dynamic adaptability is weak. Existing environmental models are mostly static or weakly dynamic. For example, deep learning-based obstacle avoidance methods do not incorporate physical processes such as biological migration and chemical diffusion, making it difficult for planned paths to adapt to the complex dynamic changes in cold seep environments. Summary of the Invention

[0004] Based on this, it is necessary to address the aforementioned technical problems of failing to accurately identify high-value target sampling areas by combining multi-source ecological environment data of deep-sea cold seeps, having a single path planning decision lacking physical constraints and weak environmental dynamic adaptability. A method and system for autonomous path planning in deep-sea cold seep environments should be provided that can accurately and autonomously plan paths to efficiently and accurately identify target sampling areas, improve ROV energy utilization, enhance environmental dynamic adaptability, and achieve high detection efficiency and sampling accuracy.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: In a first aspect, the present invention provides an autonomous path planning method for ROVs in deep-sea cold seep environments, comprising: Multi-source data from the deep-sea cold seep region were acquired and preprocessed to obtain preprocessed multi-source data. Based on the preprocessed multi-source data, a three-dimensional environmental data map of the deep-sea cold seep area is constructed. Based on the aforementioned 3D environmental data map and the real-time status parameters of the ROV, multiple initial candidate paths are obtained through a candidate path generation algorithm. Based on multiple initial candidate paths, the optimal path is selected using the DQN algorithm, and the optimal path instruction is generated. The optimal path command is transmitted to the ROV, causing the ROV to navigate along the optimal path.

[0006] Secondly, the present invention also provides an autonomous path planning system for ROVs in deep-sea cold seep environments, utilizing the autonomous path planning method for ROVs in deep-sea cold seep environments as described above, including: The deep-sea cold seep data acquisition module is used to acquire multi-source data from the deep-sea cold seep region and perform preprocessing to obtain preprocessed multi-source data. An environmental data map construction module is used to construct a three-dimensional environmental data map of the deep-sea cold seep area based on the preprocessed multi-source data. The initial candidate path generation module is used to obtain multiple initial candidate paths based on the three-dimensional environment data map and the real-time status parameters of the ROV through a candidate path generation algorithm. The optimal path instruction generation module is used to select the optimal path based on multiple initial candidate paths using the DQN algorithm and generate the optimal path instruction. The optimal route autonomous planning module is used to transmit the optimal path command to the ROV, so that the ROV can navigate according to the optimal path.

[0007] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention accurately locates high-value areas using a 3D environmental data map model, and path planning directly guides to the target sampling area, significantly improving sampling effectiveness and efficiency. Combined with the real-time status parameters of the ROV, the optimal path is selected through path planning, improving the ROV's energy utilization and ensuring accurate arrival at the target sampling area. The 3D environmental data map model integrates multi-source data on deep-sea cold seep environments from historical data, enabling real-time response to dynamic changes in the cold seep environment. Autonomous path planning eliminates the need for manual intervention in target sampling area identification and path adjustment, significantly improving operational efficiency and path execution accuracy. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating an autonomous path planning method for ROVs in a deep-sea cold seep environment, as described in some embodiments of this application. Figure 2 This is a schematic diagram of the framework of an ROV autonomous path planning system in a deep-sea cold seep environment, as described in some embodiments of this application. Detailed Implementation

[0009] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the description of embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of units is not necessarily limited to those units, but may include other units not explicitly listed or inherent to those processes, methods, products, or apparatuses. The term "determine" broadly covers a wide variety of actions, including acquiring, calculating, processing, deriving, investigating, searching (e.g., searching in a table, database, or other data structure), probing, and similar actions; it may also include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), and similar actions; it may also include generating, creating, establishing, and similar actions; and parsing, selecting, choosing, and similar actions, etc. Definitions of other terms will be given in the following description.

[0010] It should be noted that when one element is considered to be "connected" to another element, it can be directly connected to the other element or connected to the other element through an intermediary element. Furthermore, in the following embodiments, "connection" should be understood as "electrical connection," "communication connection," etc., if there is transmission of electrical signals or data between the connected objects.

[0011] It should be emphasized that the acquisition, transmission, storage, use, and processing of data in the technical solutions of this application all comply with the relevant provisions of national laws and regulations.

[0012] In the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0013] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0014] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Example 1 This embodiment provides an autonomous path planning method for ROVs in deep-sea cold seep environments. (See reference...) Figure 1 ,include: Multi-source data from the deep-sea cold seep region were acquired and preprocessed to obtain preprocessed multi-source data. Based on the preprocessed multi-source data, a three-dimensional environmental data map of the deep-sea cold seep area is constructed. Based on the aforementioned 3D environmental data map and the real-time status parameters of the ROV, multiple initial candidate paths are obtained through a candidate path generation algorithm. Based on multiple initial candidate paths, the optimal path is selected using the DQN algorithm, and the optimal path instruction is generated. The optimal path command is transmitted to the ROV, causing the ROV to navigate along the optimal path.

[0016] Compared to existing technologies, this invention accurately locates high-value areas through a 3D environmental data map model, and the path planning directly guides to the target sampling area, significantly improving sampling effectiveness and efficiency. Combined with the real-time status parameters of the ROV, the optimal path is selected through path planning, improving the ROV's energy utilization and ensuring accurate arrival at the target sampling area. The 3D environmental data map model integrates multi-source data on deep-sea cold seep environments from historical data, enabling real-time response to dynamic changes in the cold seep environment. Autonomous path planning eliminates the need for manual intervention in target sampling area identification and path adjustment, significantly improving operational efficiency and path execution accuracy.

[0017] In some preferred embodiments, multi-source data of the deep-sea cold seep environment are acquired and preprocessed to obtain preprocessed multi-source data, including: The multi-source data includes biological data, chemical data, and environmental data; Outliers in the multi-source data are removed by using the isolated forest deep learning algorithm, and then normalized to obtain normalized multi-source data. A multilayer perceptron is used to assign corresponding fusion weights to the normalized multi-source data. The sensor error coefficients corresponding to the normalized multi-source data are calculated using the physical sensor error model. Based on the normalized multi-source data and its corresponding fusion weights and sensor error coefficients, preprocessed multi-source data is obtained.

[0018] In this embodiment, the multi-source data of the deep-sea cold seep environment includes biological data, such as the distribution, density, and activity range of unique cold seep biological communities collected by underwater high-definition cameras and biosensors; chemical data, such as the concentration data of methane, hydrogen sulfide, and dissolved oxygen collected by chemical sensors; environmental data, such as the topography, water depth, ocean current speed and direction collected by sonar and inertial navigation systems; and historical data, such as historical biochemical distribution heat maps and coordinates of high-value sampling points.

[0019] Furthermore, the multi-source data preprocessing workflow adopts a three-level architecture of deep learning anomaly removal, normalization, and physical constraint weighted fusion to provide high-quality data support for subsequent modeling. The specific steps are as follows: To address the issue of sensor noise interference, an Isolation Forest deep learning algorithm is introduced to construct an outlier identification model. The degree of data anomaly is quantified using a formula, expressed as:

[0020] in, This is represented as the data sample to be tested; The score represents the outlier score; T represents the number of decision trees. Represented as the sample in the th The path length of each decision tree; Let be the average path length of the t-th decision tree; when Data exceeding θ (where θ is the outlier threshold) is considered noise and removed.

[0021] An adaptive calculation method is used to obtain the anomaly threshold θ, which is expressed as: θ = 0.5 + 0.3 × (1 - mean data confidence level). The data confidence level is calculated by the sensor error coefficient δ (e.g., confidence level is 0.9 when δ = 0.1), ensuring that the threshold is dynamically adjusted according to the sensor accuracy. This formula achieves accurate noise filtering by learning the distribution characteristics of normal data.

[0022] To eliminate the dimensional differences between data of different dimensions, the min-max normalization formula is used to map all data to the interval [0,1], which is expressed as:

[0023] in, Represented as normalized data, Represented as raw data, and These represent the extreme values ​​of the data in that dimension, laying the foundation for multi-source data fusion.

[0024] Finally, considering the differences in reliability and physical error characteristics of data from different sensors, a multilayer perceptron (MLP) deep learning algorithm is introduced to dynamically allocate fusion weights, and this is combined with a physical sensor error model for constraint, achieving accurate fusion of multi-source data, as expressed below:

[0025] in, The merged data set; These are the dynamic weights for biological, chemical, and environmental data (obtained by the MLP model through adaptive learning based on data distribution characteristics). The distribution is represented by the preprocessed data in each dimension; These are respectively the sensor error coefficients calculated by the physical sensor error model, through... The impact of correction errors on data credibility is assessed to ensure the reliability of the fusion results. In some preferred embodiments, a three-dimensional environmental data map of the deep-sea cold seep area is constructed based on the preprocessed multi-source data, including: Establish grid standards for deep-sea cold seep areas; Based on the grid standard and the preprocessed multi-source data, multi-source grid data is obtained; The multi-source grid data is input into the GNN to learn the spatial topology between grid cells in the multi-source grid data, generating a node topology feature matrix. The node feature update formula is expressed as:

[0026] in, Represented as the first Layer nodes eigenvectors; Represented as nodes The set of neighboring nodes; and These are represented as network weights and biases, respectively. The activation function is used; this formula provides intelligent feature support for 3D mapping by learning the spatial correlation between biologically dense areas and topography, and between chemically anomalous areas and ocean currents. Based on preprocessed multi-source biological data, biological community distribution changes were simulated using a biological migration and diffusion sub-model, yielding simulation results, which are expressed as follows:

[0027] in, Represented as biological density; Expressed as the diffusion coefficient; This is expressed as migration speed; Represented as a source term for reproduction or a source term for extinction; Based on the preprocessed multi-source chemical data, chemical diffusion was simulated using a chemical concentration diffusion sub-model, yielding the simulation results, which are expressed as follows:

[0028] in, Expressed as chemical concentration; Expressed as the diffusion coefficient; Expressed as ocean current velocity; This can be represented as either a source item being generated or a source item being consumed. Based on the preprocessed multi-source environmental data, ocean current motion is simulated using a hydrodynamic ocean current sub-model to obtain the dynamic simulation results of the ocean currents, which are expressed as follows:

[0029] in, Expressed as seawater density; Expressed as ocean current velocity; Expressed as pressure; Expressed as dynamic viscosity; Expressed as gravitational acceleration; The results of the biological migration and diffusion simulation, the results of the chemical concentration diffusion simulation, and the results of the ocean current dynamic parameters simulation are integrated into the node topology feature matrix to construct a three-dimensional environmental data map of the deep-sea cold seep region.

[0030] In this embodiment, based on preprocessed multi-source data, a three-dimensional environmental data map of the deep-sea cold seep area is constructed using graph neural network (GNN) deep learning algorithm and physical model, realizing ecological, chemical, and environmental mapping.

[0031] Based on detection accuracy and sensor resolution, a grid standard was set for the deep-sea cold seep area. The planar grid is 1m×1m (matching sonar sampling resolution), and the depth grid is 0.5m (matching water depth measurement accuracy). Core nodes (centers of high-value areas) are further densified to 0.5m×0.5m×0.2m to ensure accurate features in key areas. The 3D environmental data map is dynamically updated according to a preset time.

[0032] In some preferred embodiments, based on preset annotation conditions, high-value areas, non-value areas, obstacle areas, and ocean current interference areas in the three-dimensional environmental data map are annotated. Among them, the coordinates of high-value areas, the distribution of obstacles, and environmental dynamic parameters are converted into instructions and directly transmitted to the ROV, providing visualization and quantitative support for path decision-making.

[0033] In some preferred embodiments, based on the 3D environmental data map and the real-time status parameters of the ROV, multiple initial candidate paths are obtained through a candidate path generation algorithm, including: Set each grid node in the valueless area of ​​the three-dimensional environmental data map as a disabled path node; Set each grid node in the high-value area of ​​the three-dimensional environmental data map as a mandatory path node; Set all grid nodes in the 3D environment data map, except for disabled path nodes and mandatory path nodes, as optional path nodes; Based on the 3D coordinates of the mandatory path nodes, the 3D coordinates of the optional path nodes, and the real-time status parameters of the ROV, multiple initial candidate paths are obtained through the improved A* algorithm, represented as follows:

[0034] in, Represented as the distance from the starting point to the node Energy consumption; Represented as nodes Distance to the nearest high-value area; The number of initial candidate paths is determined based on the number of high-value regions.

[0035] In this embodiment, N+2 initial candidate paths are generated based on the number of high-value areas N to ensure path diversity.

[0036] In some preferred embodiments, based on multiple initial candidate paths, the optimal path is selected using the DQN algorithm, and an optimal path instruction is generated, including: Based on the biological dynamics, chemical dynamics and ocean current dynamics of each high-value area, the degree to which each initial candidate path covers the high-value area is evaluated by the DQN algorithm, and the result is used as the value score of the initial candidate path. Based on the real-time status parameters of ROV, the ROV energy consumption score of each initial candidate path is evaluated and selected through a physical energy consumption model. A dynamic adjustment formula for value weight and energy consumption weight is constructed to achieve a dynamic balance between value and energy consumption, which is expressed as:

[0037] in, and These are represented as value weight and energy consumption weight, respectively. Represented as initial value weights; Represented as the AI ​​adjustment coefficient; Represented as surplus energy; This represents the maximum range; the formula adaptively adjusts the weight ratio based on the remaining energy state to match the detection demand with the energy state. The DQN algorithm is used to iteratively optimize the path through a preset number of rounds (10-15 rounds), and the initial candidate path with the highest comprehensive score is selected as the optimal path, and the optimal path instruction is generated. The comprehensive scoring formula is expressed as follows:

[0038]

[0039] in, This is expressed as a value score (the extent to which the assessment path covers high-value areas); This is represented as an energy consumption score; This is expressed as ocean current resistance; Expressed as thruster energy consumption; Represented as path length; The optimal path instruction includes the three-dimensional coordinates and order of each path node in the optimal path, as well as the maximum speed of the ROV.

[0040] In some preferred embodiments, transmitting the optimal path command to the ROV, causing the ROV to navigate according to the optimal path, includes: The optimal path command is transmitted to the ROV. By parsing the optimal path command, the three-dimensional coordinates and order of each path node of the optimal path, as well as the maximum sailing speed, are obtained. Set the three-dimensional coordinates of the necessary path nodes in the path nodes of the optimal path as the three-dimensional coordinates of the target sampling area; The ROV travels towards the target sampling area in sequence according to the path nodes of the optimal path, and determines in real time whether the ROV has reached the target sampling area based on the three-dimensional coordinates of the necessary path nodes of the optimal path and according to the preset arrival judgment criteria. During its journey towards the target sampling area, the ROV collects and fuses real-time location information from its multiple positioning devices to obtain the ROV's real-time location, denoted as [missing information]. ; Based on the maximum speed, the future target time is predicted using a trajectory prediction model. The position is represented as:

[0041] =round(Vmax×5) in, Represented as Real-time location of ROV ; Represented as the length of the historical data window; Vmax represents the maximum sailing speed; determined based on ROV sailing speed. = round(Vmax×5) (when Vmax=1.5m / s) =7), which means covering the historical trajectory data of the past 7 seconds, balancing prediction accuracy and computing power. This formula learns the temporal characteristics of the trajectory, corrects deviations in advance, and ensures accurate navigation according to the instructions of the path optimization module.

[0042] Based on the aforementioned future target time The position is corrected in advance to ensure that the ROV travels to the target sampling area in the order of each path node of the optimal path; The control torque is calculated based on the thruster dynamics formula to ensure that the ROV resists ocean current interference during its journey towards the target sampling area. This torque is expressed as:

[0043]

[0044] in, Represented as control torque; Expressed as moment of inertia; Expressed as thruster angular velocity; Expressed as the damping coefficient; For ocean current interference torque; It is expressed as a torque coefficient (determined by the shape of the ROV, with a value of 0.05-0.1). Real-time ocean current speed; This is expressed as the ROV length, ensuring that torque calculations can be derived from known parameters, thus enabling the ROV to withstand ocean current interference.

[0045] In this embodiment, the ROV is equipped with multimodal sensors, an inertial navigation system, a thruster, and a sampling and execution structure. Its core function is to receive optimal path instructions and accurately execute navigation operations.

[0046] First, the ROV receives the optimal path instruction, parses the instruction to determine the core navigation parameters, and extracts the 3D coordinates, universal UTM projection coordinates, and depth coordinates of the high-value selection area from the instruction, denoted as... The boundary range of the target sampling area, generally a circular or polygonal boundary parameter.

[0047] Further, the path sequence is parsed, and the order of path nodes (e.g., node 1 → node 2 → … → target sampling area) and the set of coordinates of each path node are extracted, denoted as . .

[0048] Furthermore, general navigation constraints are determined, including maximum navigation speed Vmax, minimum safe distance, etc.

[0049] Furthermore, the system's collaborative positioning is initiated, including a collaborative positioning system of IMU (Inertial Measurement Unit), DVL (Doppler Velocimeter), and USBL (Ultra-Short Baseline Positioning System). This system ensures positioning accuracy through multi-source data fusion, supporting subsequent navigation decisions.

[0050] In some preferred embodiments, the method further includes the ROV collecting and fusing corresponding location information in real time through its multiple positioning devices to obtain the real-time location of the ROV, including: The ROV collects a preset number of attitude and 3D acceleration data per second using its IMU. Based on this data, the instantaneous position is calculated using inertial navigation principles and denoted as . ; The ROV measures the velocity vector of the seawater relative to the ROV using DVL (Direct Velocity Flow Measurement), corrects for the accumulated drift error of the IMU (Integrated Mutual Measure), and obtains the corrected position, denoted as . ; The ROV receives acoustic positioning signals from the surface mother ship every 2 seconds via USBL to provide absolute position calibration, obtaining the calibrated position, which is denoted as . ; The real-time location of the ROV was obtained by fusing three sets of location data using a weighted average method. The fusion formula is expressed as:

[0051] in, , and Let these be IMU weights, DVL weights, and USBL weights, respectively, satisfying... + + =1; The weights of each location data point in the fusion formula are dynamically adjusted based on the real-time errors of each positioning device (e.g., during USBL calibration). Increased to 0.5 to ensure absolute position accuracy, with a final positioning error ≤0.3m.

[0052] In this embodiment, the IMU acquires 100 sets of ROV attitude data (pitch angle α, roll angle β, yaw angle γ) and three-dimensional acceleration data per second. Calculate instantaneous position based on inertial navigation principle .

[0053] DVL corrects the accumulated drift error of the IMU by measuring the velocity vector of seawater relative to the ROV, and outputs the corrected position. .

[0054] USBL receives acoustic positioning signals from the surface mother ship every 2 seconds, provides absolute position calibration, and outputs the calibrated position. .

[0055] In some preferred embodiments, based on the three-dimensional coordinates of the necessary path nodes of the optimal path, the ROV is determined in real time according to a preset arrival determination criterion to determine whether it has reached the target sampling area, including: During navigation towards the target sampling area, the ROV's real-time position is compared in real time with the three-dimensional coordinates of the necessary path nodes of the optimal path: If the deviation between the 3D coordinates of the ROV's real-time location and the 3D coordinates of the necessary path nodes of the optimal path is not greater than a preset threshold, it is determined that the target sampling area has been successfully reached.

[0056] In this embodiment, to confirm arrival at the target sampling area, the real-time precise position of the ROV is continuously compared. Center coordinates of the target sampling area A successful arrival is determined when the following preset arrival criteria are met: The three-dimensional coordinate deviation of the set position is ≤0.5m, that is: and and

[0057] In the specific implementation, the detection scenario was a deep-sea cold seep area containing two densely populated areas of tubular worms (biologically high-value areas M and N) and two hydrogen sulfide-methane complex seepage points (chemically high-value areas P and Q). Surrounding these areas were one reef obstacle area and one area with moderate ocean current interference. The detection target was the comprehensive high-value complex area, balancing sampling effectiveness with energy control. The ROV model was XYZ-ROV700, equipped with an IMU+DVL+USBL collaborative positioning system, a biosensor (XYZ-B200), a chemical sensor (XYZ-C300), a 10kW thruster, and a hydraulic sampling clamp. It had a maximum speed of 1.5 m / s, a range of 100 kWh, and a positioning error of ≤0.3 m.

[0058] First, input the density data of tubular worms collected by the biosensor, the hydrogen sulfide / methane concentration data collected by the chemical sensor, the topographic / water depth data collected by the sonar (XYZ-S400), and the coordinates of 15 historical composite high-value sampling points.

[0059] Furthermore, outlier removal is performed: Using the Isolation Forest algorithm:

[0060] The number of decision trees is set to T=120. The data reliability is calculated by the sensor error coefficient (biosensor δ_b=0.06→reliability 0.94, chemical sensor δ_c=0.07→reliability 0.93, environmental sensor δ_e=0.09→reliability 0.91). The anomaly threshold θ=0.5+0.3×(1-(0.94+0.93+0.91) / 3)=0.53 is adaptively calculated, and 3.2% of the noisy data is removed.

[0061] Further, normalization is performed: Through min-max normalization:

[0062] Data such as biological density, chemical concentration, and terrain height are uniformly mapped to the [0,1] interval to eliminate dimensional differences.

[0063] Furthermore, a weighted fusion based on physical constraints is performed: Dynamic weight allocation based on MLP deep learning algorithm:

[0064] Calculate the fusion weights by incorporating sensor error coefficient correction: With w_b=0.38, w_c=0.42 and w_e=0.2 (chemical data have slightly higher weights to suit composite region detection), a standardized dataset is generated.

[0065] Based on the above standardized multi-source data, according to GNN topology learning and grid division, the grid standard is set according to "detection accuracy + sensor resolution", with a planar grid of 1m×1m and a depth grid of 0.5m. The high-value area center (M, N, P, Q) is densified to 0.5m×0.5m×0.2m.

[0066] Furthermore, node updates are performed using the GNN node feature update formula, which is expressed as:

[0067] The network is set to have 3 layers and the activation function is σ=ReLU. The spatial correlation between biologically dense areas and chemical seepage points, as well as the interaction between ocean currents and topography, are studied to generate a node topology feature matrix.

[0068] Further, physical model simulations are performed: Incorporating biological migration and diffusion formulas ( =0.09m² / s =0.03m / s), chemical concentration diffusion formula ( =0.085m² / s (calculated from fluid mechanics formulas), and fluid mechanics formulas ( =1025kg / m³ =0.001Pa (s), simulating the migration of tubular worms, the diffusion of hydrogen sulfide-methane, and the movement of ocean currents.

[0069] Further, 3D map generation and output are performed: Construct a 1:500 scale 3D environmental data map, label the composite high-value areas M, N, P, Q, reef obstacle areas, and ocean current interference areas (current velocity 0.3-0.5 m / s), output the 3D coordinates and dynamic parameters (ocean current velocity / direction, biological density change rate) of each area, and transmit them to the ROV in real time.

[0070] First, input the area coordinates and dynamic parameters output from the environmental data map, and the real-time status of the ROV (initial remaining energy 100kWh, maximum speed 1.5m / s).

[0071] Furthermore, an improved A* algorithm is used to generate candidate paths, with composite high-value areas as mandatory nodes and obstacle / valueless areas as prohibited nodes, based on a heuristic function. This generates N+2=6 initial candidate paths (N=4 high-value areas) to ensure path diversity.

[0072] Furthermore, dynamic weight adjustments are performed: Based on DQN deep reinforcement learning, initial value weights are set. =0.62, AI adjustment coefficient =0.55, according to the formula Adaptive weight allocation. Initial energy remaining is 100% (>60%). =0.65、 =0.35. Automatically adjust to 0.35 when remaining energy drops to 55%. =0.5、 =0.5.

[0073] Furthermore, comprehensive scoring and iterative optimization are conducted: The comprehensive scoring is performed using a formula, which is expressed as follows:

[0074] in, The completeness of the path coverage of high-value areas is assessed by DQN (full coverage earns 1 point). The formula is calculated using the physical energy consumption model. , For ocean current resistance, For propulsion energy consumption.

[0075] After 13 rounds of iterative optimization, the optimal path with the highest comprehensive score was selected: starting point → M → P → N → Q → starting point. This path avoids ocean current interference areas, with a total energy consumption of 38 kWh, and covers all complex high-value areas. The path coordinates, node sequence, and navigation speed commands are transmitted to the ROV.

[0076] The ROV receives the optimal path command and parses it to extract the three-dimensional coordinates (UTM projection coordinates + depth) of the high-value area: M(324512.3, 3567891.2, 1245.6), P(324531.8, 3567902.5, 1243.8), N(324542.1, 3567915.7, 1246.3), Q(324528.4, 3567923.9, 1244.1), thus determining the order of path nodes and the maximum navigation speed of 1.2 m / s (reducing speed in areas affected by ocean currents).

[0077] Furthermore, collaborative positioning and trajectory prediction are performed: an IMU+DVL+USBL positioning system is activated. The IMU collects 100 sets of attitude and acceleration data per second, DVL corrects drift, and USBL calibrates the absolute position every 2 seconds, dynamically allocating weights according to a weighted fusion formula (w3=0.5 during USBL calibration, and w1=0.3, w2=0.4, w3=0.3 under normal conditions). Based on the LSTM trajectory prediction formula, the historical data window length is set to k=round(1.5×5)=7, predicting the position in the future τ=2s, and correcting navigation deviations in advance.

[0078] Furthermore, propeller control is performed: according to the propeller dynamics formula. Calculate the control torque, where , =0.07 (ROV shape adaptation). Real-time ocean current velocity (from environmental data map). =3.2m (ROV length) to resist ocean current interference; after arriving at each high-value area, sampling is started when the three-dimensional coordinate deviation is ≤0.5m, and real-time location, remaining energy and biochemical data are transmitted back every 10 seconds to support the dynamic updating of environmental data maps.

[0079] By identifying high-value areas, the route planning directly guides the sampling area, significantly improving the biological / chemical specificity of the sampled specimens. Compared to existing technologies, the sample effectiveness rate has increased from 30% to over 75%.

[0080] By avoiding detours through low-value areas and traversing high-energy-consuming areas, and combining dynamic weight optimization and physical energy consumption modeling, ROV energy consumption is reduced by more than 40%, and the time for a single detection operation is extended by 50%, allowing coverage of more high-value areas.

[0081] Based on a mechanism that dynamically updates 3D environmental data maps every 30 seconds, and integrates physical simulations such as biological migration and chemical diffusion with AI predictions, it can respond to dynamic changes in the cold seep environment in real time. Compared with fixed path solutions, the environmental adaptation accuracy is improved by 60%.

[0082] ROV autonomously completes the entire process of "data integration-modeling-planning-sampling" without the need for manual intervention in high-value area identification and path adjustment, improving operational efficiency by more than 70%. The combination of IMU+DVL+USBL collaborative positioning and LSTM trajectory prediction results in a positioning error of ≤0.3m, and the path execution accuracy is significantly better than existing technologies.

[0083] Example 2 This embodiment further provides an autonomous path planning system for ROVs in deep-sea cold seep environments, based on Embodiment 1. It utilizes the autonomous path planning method for ROVs in deep-sea cold seep environments described above. (See attached document for more details.) Figure 2 ,include: The deep-sea cold seep data acquisition module is used to acquire multi-source data from the deep-sea cold seep region and perform preprocessing to obtain preprocessed multi-source data. An environmental data map construction module is used to construct a three-dimensional environmental data map of the deep-sea cold seep area based on the preprocessed multi-source data. The initial candidate path generation module is used to obtain multiple initial candidate paths based on the three-dimensional environment data map and the real-time status parameters of the ROV through a candidate path generation algorithm. The optimal path instruction generation module is used to select the optimal path based on multiple initial candidate paths using the DQN algorithm and generate the optimal path instruction. The optimal route autonomous planning module is used to transmit the optimal path command to the ROV, so that the ROV can navigate according to the optimal path.

[0084] The deep-sea cold seep data acquisition module of this system uses the isolated forest algorithm to remove sensor noise, combines it with MLP deep learning to dynamically allocate fusion weights, incorporates physical sensor error coefficients to correct data reliability, and achieves accurate integration of biological, chemical, environmental and historical multidimensional heterogeneous data through standardized formulas. This breaks through the limitations of traditional data fusion, which is fragmented and has fixed weights, and provides high-quality data support for environmental modeling.

[0085] The environmental data map construction module of this system integrates the GNN graph neural network algorithm to learn the spatial topological relationships of environmental elements. It achieves adaptive grid densification (core area 0.5m×0.5m×0.2m) according to "detection accuracy + sensor resolution". Combined with physical models such as biological migration and diffusion, chemical concentration diffusion, and fluid dynamics, it constructs a 1:500 scale three-dimensional model to realize the dynamic mapping of ecology, chemistry, and environment, quantitatively define high / no-value areas, and provide accurate spatial support for path planning.

[0086] The initial candidate path generation module of this system generates diverse candidate paths based on the improved A* algorithm. The optimal path instruction generation module constructs a dynamic weight adjustment formula through DQN deep reinforcement learning, adaptively allocates value and energy consumption weights according to the remaining energy of the ROV, and calculates a comprehensive score by combining the physical energy consumption model (ocean current resistance + thruster energy consumption). After 10-15 rounds of iteration, the optimal path is selected, abandoning the traditional single-objective-oriented logic and achieving a dynamic balance between sampling value and energy consumption.

[0087] The optimal route autonomous planning module of this system constructs a seamless link of "data integration - model building - path optimization - autonomous execution", and the modules interact in real time through data flow; the ROV is equipped with an IMU+DVL+USBL collaborative positioning system, which combines LSTM trajectory prediction and thruster dynamics formula to resist ocean current interference and achieve positioning accuracy of 0.3m. At the same time, it supports dynamic model updates through real-time data feedback to ensure accurate adaptation of path planning and execution.

[0088] It is understood that the system in this embodiment corresponds to the method in Embodiment 1 above, and the options in Embodiment 1 above are also applicable to this embodiment, so they will not be described again here.

[0089] Example 3 This embodiment provides a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set. The at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor, causing the processor to perform some or all of the steps of the method provided in Embodiment 1 of this application.

[0090] It is understood that the storage medium can be transient or non-transient. Exemplarily, the storage medium includes, but is not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] By way of example, the processor may be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0092] By way of example, the read-only memory includes, but is not limited to, MASK ROM, PROM, EPROM, EEPROM, Flash, etc.

[0093] By way of example, the random access memory includes, but is not limited to, DRAM, SRAM, SDRAM, DDR SDRAM, etc.

[0094] In some examples, a computer program product is provided, which can be implemented by hardware, software, or a combination thereof. As a non-limiting example, the computer program product can be embodied in the storage medium, or it can be embodied in a software product, such as an SDK (Software Development Kit).

[0095] As a non-limiting example, a computer program product is provided, comprising a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer program or computer-executable instructions from the computer-readable storage medium, and executes the computer-executable instructions, causing the electronic device to perform some or all of the steps of the method described in the embodiments of this application.

[0096] In some examples, a computer program is provided, including computer-readable code, wherein, when the computer-readable code is run in a computer device, a processor in the computer device performs some or all of the steps for implementing the method.

[0097] This embodiment also proposes an electronic device, including a memory and a processor. The memory stores at least one instruction, at least one program, code set, or instruction set. When the processor executes the at least one instruction, at least one program, code set, or instruction set, it implements some or all of the steps of the method described in Embodiment 1.

[0098] In some examples, a hardware entity of the electronic device is provided, including: a processor, a memory, and a communication interface; wherein the processor typically controls the overall operation of the electronic device; the communication interface is used to enable the electronic device to communicate with other terminals or servers via a network; the memory is configured to store instructions and applications executable by the processor, and may also cache data to be processed or already processed (including but not limited to image data, audio data, voice communication data, and video communication data) to be processed by the processor and various modules in the electronic device, and may be implemented using flash memory (FLASH), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or random access memory (RAM).

[0099] A processor may include one or more processing elements. Therefore, a processor may include one or more integrated circuits (ICs) configured to perform the functions of the processor. Furthermore, each integrated circuit may include circuitry (e.g., a first circuit, a second circuit, and other circuitry, etc.) configured to perform the functions of the processor.

[0100] Furthermore, data can be transferred between the processor, communication interface, and memory via a bus, which can include any number of interconnected buses and bridges, connecting various circuits of one or more processors and memories together.

[0101] It is understood that the options in Embodiment 1 above also apply to this embodiment, so they will not be described again here.

[0102] The same or similar labels correspond to the same or similar parts; The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this application. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0103] In different specific implementations, the methods or systems described in this application can be implemented in software, hardware, or a combination thereof. Furthermore, the order of the method steps can be changed, and various elements can be added, reordered, combined, omitted, or modified.

[0104] Obviously, the above embodiments of this application are merely examples for clearly illustrating this application, and are not intended to limit the implementation of this application, nor are they intended to limit this application. For those skilled in the art, other variations or modifications can be made based on the above description. The separate structural / functional modules or units can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. The structure and function of the separate components can be implemented as a combined structure or component. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of the claims of this application.

Claims

1. A method for ROV autonomous path planning in a deep-sea cold-spring environment, characterized in that, include: Multi-source data from the deep-sea cold seep region were acquired and preprocessed to obtain preprocessed multi-source data. Based on the preprocessed multi-source data, a three-dimensional environmental data map of the deep-sea cold seep area is constructed. Based on the aforementioned 3D environmental data map and the real-time status parameters of the ROV, multiple initial candidate paths are obtained through a candidate path generation algorithm. Based on multiple initial candidate paths, the optimal path is selected using the DQN algorithm, and the optimal path instruction is generated. The optimal path command is transmitted to the ROV, causing the ROV to navigate along the optimal path.

2. The ROV autonomous path planning method for deep-sea cold spring environment according to claim 1, characterized in that, Multi-source data of the deep-sea cold seep environment were acquired and preprocessed to obtain preprocessed multi-source data, including: The multi-source data includes biological data, chemical data, and environmental data; Outliers in the multi-source data are removed by using the isolated forest deep learning algorithm, and then normalized to obtain normalized multi-source data. A multilayer perceptron is used to assign corresponding fusion weights to the normalized multi-source data. The sensor error coefficients corresponding to the normalized multi-source data are calculated using the physical sensor error model. Based on the normalized multi-source data and its corresponding fusion weights and sensor error coefficients, preprocessed multi-source data is obtained.

3. The ROV autonomous path planning method for deep-sea cold seep environments according to claim 2, characterized in that, Based on the preprocessed multi-source data, a three-dimensional environmental data map of the deep-sea cold seep region is constructed, including: Establish grid standards for deep-sea cold seep areas; Based on the grid standard and the preprocessed multi-source data, multi-source grid data is obtained; The multi-source grid data is input into the GNN to learn the spatial topology between grid cells in the multi-source grid data, generating a node topology feature matrix. The node feature update formula is expressed as: in, Represented as the first Layer nodes eigenvectors; Represented as nodes The set of neighboring nodes; and These are represented as network weights and biases, respectively. For activation functions; Based on preprocessed multi-source biological data, the biological migration and diffusion simulation results are obtained through a biological migration and diffusion sub-model, which are expressed as follows: in, Represented as biological density; Expressed as the diffusion coefficient; This is expressed as migration speed; Represented as a source term for reproduction or a source term for extinction; Based on the preprocessed multi-source chemical data, the chemical concentration diffusion simulation results are obtained through a chemical concentration diffusion sub-model, which are expressed as follows: in, Expressed as chemical concentration; Expressed as the diffusion coefficient; Expressed as ocean current velocity; This can be represented as either a source item being generated or a source item being consumed. Based on the preprocessed multi-source environmental data, the dynamic simulation results of ocean currents are obtained through a hydrodynamic ocean current sub-model, which are expressed as follows: in, Expressed as seawater density; Expressed as ocean current velocity; Expressed as pressure; Expressed as dynamic viscosity; Expressed as gravitational acceleration; The results of the biological migration and diffusion simulation, the results of the chemical concentration diffusion simulation, and the results of the ocean current dynamic parameters simulation are integrated into the node topology feature matrix to construct a three-dimensional environmental data map of the deep-sea cold seep region.

4. The ROV autonomous path planning method for deep-sea cold seep environments according to any one of claims 1-3, characterized in that, Based on preset annotation conditions, high-value areas, low-value areas, obstacle areas, and ocean current interference areas in the three-dimensional environmental data map are annotated.

5. The ROV autonomous path planning method for deep-sea cold seep environments according to claim 4, characterized in that, Based on the aforementioned 3D environmental data map and the real-time status parameters of the ROV, multiple initial candidate paths are obtained through a candidate path generation algorithm, including: Set each grid node in the valueless area of ​​the three-dimensional environmental data map as a disabled path node; Set each grid node in the high-value area of ​​the three-dimensional environmental data map as a mandatory path node; Set all grid nodes in the 3D environment data map, except for disabled path nodes and mandatory path nodes, as optional path nodes; Based on the 3D coordinates of the mandatory path nodes, the 3D coordinates of the optional path nodes, and the real-time status parameters of the ROV, multiple initial candidate paths are obtained through the improved A* algorithm, which are represented as follows: in, Represented as the distance from the starting point to the node Energy consumption; Represented as nodes Distance to the nearest high-value area; The number of initial candidate paths is determined based on the number of high-value regions.

6. The ROV autonomous path planning method for deep-sea cold seep environments according to claim 5, characterized in that, Based on multiple initial candidate paths, the optimal path is selected using the DQN algorithm, and an optimal path instruction is generated, including: Based on the biological dynamics, chemical dynamics and ocean current dynamics of each high-value area, the degree to which each initial candidate path covers the high-value area is evaluated by the DQN algorithm, and the result is used as the value score of the initial candidate path. Based on the real-time status parameters of ROV, the ROV energy consumption score of each initial candidate path is evaluated and selected through a physical energy consumption model. A dynamic adjustment formula for value weight and energy consumption weight is constructed, which is expressed as follows: in, and These are represented as value weight and energy consumption weight, respectively. Represented as initial value weights; Represented as the AI ​​adjustment coefficient; Represented as surplus energy; This represents the maximum battery life. The DQN algorithm iteratively optimizes the preset rounds, selects the initial candidate path with the highest comprehensive score as the optimal path, and generates the optimal path instruction. The comprehensive scoring formula is expressed as follows: in, Represented as a value score; This is represented as an energy consumption score; This is expressed as ocean current resistance; Expressed as thruster energy consumption; Represented as path length; The optimal path instruction includes the three-dimensional coordinates and order of each path node in the optimal path, as well as the maximum speed of the ROV.

7. The ROV autonomous path planning method for deep-sea cold seep environments according to claim 6, characterized in that, Transmitting the optimal path command to the ROV, causing the ROV to navigate according to the optimal path, includes: The optimal path command is transmitted to the ROV. By parsing the optimal path command, the three-dimensional coordinates and order of each path node of the optimal path, as well as the maximum sailing speed, are obtained. Set the three-dimensional coordinates of the necessary path nodes in the path nodes of the optimal path as the three-dimensional coordinates of the target sampling area; The ROV travels towards the target sampling area in the order of each path node of the optimal path, and determines in real time whether the ROV has reached the target sampling area based on the three-dimensional coordinates of the necessary path nodes of the optimal path and according to the preset arrival judgment criteria. During its journey towards the target sampling area, the ROV collects and fuses real-time location information from its multiple positioning devices to obtain the ROV's real-time location, denoted as [missing information]. ; Based on the maximum speed, the future target time is predicted using a trajectory prediction model. The position is represented as: =round(Vmax×5) in, Represented as Real-time location of ROV ; Vmax represents the length of the historical data window; Vmax represents the maximum sailing speed. Based on the aforementioned future target time The position is corrected in advance to ensure that the ROV travels to the target sampling area in the order of each path node of the optimal path; The control torque is calculated based on the thruster dynamics formula to ensure that the ROV resists ocean current interference during its journey towards the target sampling area. This torque is expressed as: in, Represented as control torque; Expressed as moment of inertia; Expressed as thruster angular velocity; Expressed as the damping coefficient; For ocean current interference torque; Expressed as torque coefficient; Real-time ocean current speed; This is expressed as the ROV body length.

8. The ROV autonomous path planning method for deep-sea cold seep environments according to claim 7, characterized in that, The method further includes the ROV collecting and fusing corresponding location information in real time through its multiple positioning devices to obtain the ROV's real-time location, including: The ROV collects a preset number of attitude and 3D acceleration data per second via its IMU. Based on this data, the instantaneous position is calculated using inertial navigation principles and denoted as . ; The ROV measures the velocity vector of the seawater relative to the ROV using DVL (Direct Velocity Flow Measurement), corrects for the accumulated drift error of the IMU (Integrated Mutual Measure), and obtains the corrected position, denoted as . ; The ROV receives acoustic positioning signals from the surface mother ship every 2 seconds via USBL to provide absolute position calibration, obtaining the calibrated position, which is denoted as... ; The real-time location of the ROV was obtained by fusing three sets of location data using a weighted average method. The fusion formula is expressed as: in, , and Let these be IMU weights, DVL weights, and USBL weights, respectively, satisfying... + + =1; The weights of each location data in the fusion formula are dynamically adjusted based on the real-time errors of each positioning device.

9. The ROV autonomous path planning method for deep-sea cold seep environments according to claim 7, characterized in that, Based on the three-dimensional coordinates of the necessary path nodes of the optimal path, the system determines in real time whether the ROV has reached the target sampling area according to a preset arrival determination criterion, including: During navigation towards the target sampling area, the ROV's real-time position is compared in real time with the three-dimensional coordinates of the necessary path nodes of the optimal path: If the deviation between the 3D coordinates of the ROV's real-time location and the 3D coordinates of the necessary path nodes of the optimal path is not greater than a preset threshold, it is determined that the target sampling area has been successfully reached.

10. An autonomous path planning system for ROVs in deep-sea cold seep environments, utilizing the autonomous path planning method for ROVs in deep-sea cold seep environments as described in any one of claims 1-9, characterized in that, include: The deep-sea cold seep data acquisition module is used to acquire multi-source data from the deep-sea cold seep region and perform preprocessing to obtain preprocessed multi-source data. An environmental data map construction module is used to construct a three-dimensional environmental data map of the deep-sea cold seep area based on the preprocessed multi-source data. The initial candidate path generation module is used to obtain multiple initial candidate paths based on the three-dimensional environment data map and the real-time status parameters of the ROV through a candidate path generation algorithm. The optimal path instruction generation module is used to select the optimal path based on multiple initial candidate paths using the DQN algorithm and generate the optimal path instruction. The optimal route autonomous planning module is used to transmit the optimal path command to the ROV, so that the ROV can navigate according to the optimal path.

Images