Underwater propeller control method and system based on tensor recognition and fuzzy control

Abstract

The invention provides an underwater propeller control method and system based on tensor recognition and fuzzy control. The underwater propeller control method and system are suitable for improving the propelling efficiency and adjustment intelligence in a complex flow field. According to the method, a wake flow simulation model is constructed based on geometric parameters and boundary conditions of a propeller, a rotation tensor and a strain tensor are derived after flow field data are obtained, and a tensor field index is extracted through function space mapping so as to determine a target grid region. A high-rotation candidate area is screened through a spectral clustering algorithm, a vortex structure is identified, and multi-dimensional features such as the scale, the strength, the axial direction and the vortex core position of the vortex structure are extracted. And constructing a state vector by combining the current propulsive efficiency and flow field disturbance parameters, inputting an adaptive fuzzy neural network model, reasoning a relationship between a vortex and an operation state, outputting a propeller rotating speed and an attack angle adjusting quantity, and realizing intelligent response and energy efficiency optimization of wake flow disturbance.

Description

Technical Field

[0001] This application relates to the technical field of underwater thrusters, and particularly to an underwater thruster control method and system based on tensor recognition and fuzzy control. Background Art

[0002] As the main power output unit of underwater navigation platforms such as submersibles, the performance of underwater thrusters directly determines the performance of the overall system in terms of endurance, stability, and adaptability to complex working conditions. Most existing thruster systems still mainly rely on static geometric structure configurations and linear rotational speed regulation methods in design and control strategies, lacking in-depth integrated design for the fluid disturbance response mechanism in variable working condition environments.

[0003] During the actual operation of the thruster, due to the complex interaction between the high-speed rotation of the propeller blades and the water body, a series of dynamically changing vortex structures often form in the wake flow field. These vortices not only interfere with the distribution of local propulsion force but may also induce unstable phenomena such as cavitation and enhanced turbulence, causing disorder in the energy transfer path and significantly affecting the propulsion efficiency and the stability of the propulsion system.

[0004] However, traditional control methods usually only collect macroscopic parameters such as the rotational speed, target speed, or water depth of the thruster, lacking a mechanism for identifying and utilizing the vorticity field and disturbance structure inside the wake region. This "perception lack" leads to the control system being unable to effectively respond to the evolution of the flow pattern in the wake, and thus problems such as propulsion performance degradation and system response lag occur under high load or variable flow conditions, becoming one of the key bottlenecks restricting the further improvement of underwater intelligent propulsion systems. Summary of the Invention

[0005] To solve the above technical problems, this application provides an underwater thruster control method and system based on tensor recognition and fuzzy control.

[0006] The technical solutions provided in this application are described below: In the first aspect of this application, an underwater thruster control method based on tensor recognition and fuzzy control is provided, including: Based on the geometric parameters and operating boundary conditions of the thruster, construct a numerical simulation model of the wake region and perform simulation to obtain basic flow field data; Perform tensor field derivation processing based on the basic flow field data to obtain a rotation tensor and a strain tensor describing rotation and deformation characteristics; Map the rotation tensor and the strain tensor to a first function space and a second function space respectively to obtain a local tensor field index group, and determine a target grid region based on the local tensor field index group; Generate a curl feature based on the rotation tensor, and construct a graph structure model based on the curl feature to perform spectral clustering to obtain candidate grid regions, where the curl feature of the candidate grid regions reaches a preset threshold; Perform an aggregation process on each grid cell in the candidate grid regions, and determine a target vortex region, where the target vortex region has rotational dominance in the candidate grid regions; For the identified target vortex regions, extract multi-dimensional vortex structure feature data including at least vortex scale, vortex intensity, rotation axis direction, and vortex core position; Collect the current propulsion efficiency index and flow field disturbance parameters, and construct a multi-dimensional state vector in combination with the multi-dimensional vortex structure feature data; Input the multi-dimensional state vector into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between the vortex intensity and the operating state of the thruster, and output a target rotational speed adjustment amount and a blade angle of attack adjustment amount for adjusting the operating state of the thruster.

[0007] Optionally, generating a curl feature based on the rotation tensor, and constructing a graph structure model based on the curl feature to perform spectral clustering to obtain candidate grid regions, where the curl feature of the candidate grid regions reaches a preset threshold, includes: Calculate the curl feature of each grid cell based on the rotation tensor, where the curl feature is used to characterize the rotation intensity of the local flow field; Construct the grid cells containing the curl feature into a graph structure model, where the nodes in the graph structure model correspond to the grid cells, and the edge weights are set based on the difference in the curl features between adjacent nodes; Apply a spectral clustering algorithm to the graph structure model, and perform unsupervised partitioning on the grid cells to obtain a clustering result; Screen out the clustering clusters with curl features exceeding the preset threshold from the clustering result, and determine them as candidate grid regions.

[0008] Optionally, mapping the rotation tensor and the strain tensor to a first function space and a second function space respectively to obtain a local tensor field index group, and determining the target grid region based on the local tensor field index group includes: Map the rotation tensor to a first function space dominated by rotation features, where the first function space is an eigenmode domain constructed based on tensor eigenvalue decomposition and is used to reflect the local vortex distribution characteristics; Map the strain tensor to a second function space dominated by shear deformation and the direction of the principal stress, where the second function space is used to represent the anisotropic behavior of local deformation; Extract a first tensor response eigenvalue related to the direction of the vortex main axis in the first function space, and extract a second tensor response eigenvalue related to the direction of the maximum principal strain in the second function space; Construct a local tensor field index group based on the first tensor response eigenvalue and the second tensor response eigenvalue to characterize the rotation dominance and shear coupling characteristics of the grid region; Determine the target grid region based on the local tensor field index group.

[0009] Optionally, the determining the target grid region based on the local tensor field index group includes: For each grid cell in the wake region, obtain the corresponding first tensor response eigenvalue and second tensor response eigenvalue respectively; Based on a preset functional relationship, perform a combined operation on the first tensor response eigenvalue and the second tensor response eigenvalue to obtain the combined eigenvalue of each grid cell; According to the relationship between the combined eigenvalue and the first threshold, screen out the grid cells that do not meet the condition of rotation dominance; For the remaining grid cells, calculate the included angle information between the principal direction of the rotation tensor and the principal direction of the strain tensor; Screen out the target grid region that meets the condition of rotation dominance according to the included angle information.

[0010] Optionally, the performing the tensor field derivation process based on the basic flow field data to obtain the rotation tensor and the strain tensor describing the rotation and deformation characteristics includes: Based on the velocity vectors in the basic flow field data, calculate the velocity gradient tensor of each grid cell; Decompose the velocity gradient tensor into an anti-symmetric part and a symmetric part, where the anti-symmetric part corresponds to the rotation tensor for describing the rotation characteristics of the local fluid, and the symmetric part corresponds to the strain tensor for describing the deformation characteristics of the local fluid; Calculate the eigenvalues and eigenvectors of the rotation tensor and the strain tensor respectively.

[0011] Optionally, the condition for meeting the rotation dominance includes that the combined eigenvalue is greater than or equal to the first threshold.

[0012] Optionally, perform an aggregation process on each grid cell in the candidate grid region and determine the target vortex region, where the target vortex region has rotation dominance in the candidate grid region includes: According to the spatial proximity relationship, divide multiple adjacent grid cells that meet the condition of rotation dominance into a connected region as the candidate vortex region; Calculate the comprehensive vortex intensity index within each candidate vortex region, and the comprehensive vortex intensity index includes the coordinates of the vortex core position, the vortex intensity, the scale of the vortex core, and the local energy consumption density; According to a preset vortex intensity threshold and a vortex core scale threshold, a target vortex region that meets the conditions of the vortex intensity threshold and rotational dominance is screened out.

[0013] Optionally, the inputting the multi-dimensional state vector into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between the vortex intensity and the operating state of the thruster, and outputting a target rotational speed adjustment amount and a blade angle of attack adjustment amount for adjusting the operating state of the thruster includes: Fuzzify each variable in the multi-dimensional state vector and map it to the corresponding fuzzy set to form fuzzy input variables; According to a preset fuzzy rule base, determine the activation strength of each fuzzy rule in the fuzzy rule base based on the fuzzy input variables; Determine a target fuzzy rule based on the activation strength, and comprehensively simulate the target fuzzy rule through a fuzzy simulation module in the adaptive fuzzy neural network model to obtain an output fuzzy set; Defuzzify the output fuzzy set to obtain a numerical target rotational speed adjustment amount and a blade angle of attack adjustment amount.

[0014] A second aspect of the present application provides an underwater thruster control system based on tensor recognition and fuzzy control, which is characterized by including: A simulation unit, configured to construct a numerical simulation model of the wake region based on the geometric parameters and operating boundary conditions of the thruster, and perform a simulation to obtain basic flow field data; A tensor field unit, configured to perform tensor field derivation processing based on the basic flow field data to obtain a rotation tensor and a strain tensor describing rotation and deformation characteristics; A function mapping unit, configured to map the rotation tensor and the strain tensor to a first function space and a second function space respectively to obtain a local tensor field index group, and determine a target grid region based on the local tensor field index group; A pre-screening unit, configured to generate a curl feature according to the rotation tensor, and construct a graph structure model based on the curl feature to perform spectral clustering to obtain a candidate grid region, and the curl feature of the candidate grid region reaches a preset threshold; A grid cell aggregation unit, configured to perform an aggregation process on each grid cell in the candidate grid region, and determine a target vortex region, and the target vortex region has rotational dominance in the candidate grid region; A data extraction unit, configured to extract multi-dimensional vortex structure feature data including at least vortex scale, vortex intensity, rotation axis direction, and vortex core position for the identified target vortex region; A data acquisition unit, configured to collect current propulsion efficiency indicators and flow field disturbance parameters, and construct a multi-dimensional state vector in combination with the multi-dimensional vortex structure feature data; Input the multi-dimensional state vector into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between the vortex intensity and the operating state of the thruster, and output the target rotational speed adjustment amount and blade angle of attack adjustment amount for adjusting the operating state of the thruster.

[0015] The third aspect of the present application provides an underwater thruster control system based on tensor recognition and fuzzy control, and the device includes: A processor, a memory, an input / output unit, and a bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, and the processor calls the program to execute the method of the first aspect and any optional method in the first aspect.

[0016] It can be seen from the above technical solutions that the present application has the following advantages: By separately mapping the rotation tensor and the strain tensor to different function spaces, the present invention realizes the effective separation and high-dimensional expression of the rotation characteristics and deformation characteristics in the flow field, enhancing the accuracy and robustness of vortex structure recognition. Using the combination of local tensor field indicators in the two function spaces to discriminate the target grid area can more finely screen out the vortex area dominated by rotation, avoiding the misjudgment problem caused by shear deformation in the traditional method, thereby improving the accuracy and reliability of vortex extraction. This step provides a more accurate and stable input data basis for the subsequent dynamic inference based on the fuzzy neural network, further ensuring the effect of thruster intelligent control and the sensitivity of system response.

[0017] The present invention accurately identifies and locates the vortex structure in the flow field by constructing a numerical simulation model of the wake region based on the geometric parameters of the thruster and the operating boundary conditions, combined with the tensor field analysis of the rotation tensor and the strain tensor. Using the multi-dimensional state vector to comprehensively consider the target multi-dimensional vortex structure characteristic data, propulsion efficiency index, and flow field disturbance parameters, combined with a pre-configured adaptive fuzzy neural network model, dynamically infers the correlation between the vortex intensity and the operating state of the thruster, and effectively outputs accurate target rotational speed adjustment amount and blade angle of attack adjustment amount. This method can intelligently respond to flow field disturbances, realize real-time optimization and adjustment of the thruster operating state, significantly improve the operating efficiency and stability of the underwater thruster, reduce energy consumption and mechanical wear, and improve the overall performance and lifespan of the underwater propulsion system. Description of the Drawings

[0018] To more clearly illustrate the technical solutions in this application, the following will briefly introduce the attached drawings required for description in the embodiments. Obviously, the attached drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other attached drawings can be obtained based on these drawings.

[0019] Figure 1 It is a schematic flowchart of an embodiment of the underwater thruster control method based on tensor recognition and fuzzy control provided in this application; Figure 2 It is a schematic flowchart of a specific embodiment of step S102 in the underwater thruster control method based on tensor recognition and fuzzy control provided in this application; Figure 3 It is a schematic flowchart of a specific embodiment of step S103 in the underwater thruster control method based on tensor recognition and fuzzy control provided in this application; Figure 4 It is a schematic flowchart of a specific embodiment of step S105 in the underwater thruster control method based on tensor recognition and fuzzy control provided in this application; Figure 5 It is a schematic flowchart of a specific embodiment of step S106 in the underwater thruster control method based on tensor recognition and fuzzy control provided in this application; Figure 6 It is a schematic structural diagram of an embodiment of the underwater thruster control system based on tensor recognition and fuzzy control provided in this application; Figure 7 It is a schematic structural diagram of another embodiment of the underwater thruster control system based on tensor recognition and fuzzy control provided in this application. Detailed implementation manners

[0020] In this application, to achieve precise recognition and dynamic regulation of the vortex structure in the underwater thruster flow field, the definition and use of multiple physical parameters and mathematical quantities are involved. The relevant terms and setting standards are described as follows: Vorticity is the rotation rate of velocity in a vector field with respect to space, defined as the curl of the velocity vector field. In this application, the vorticity modulus is used to represent the local rotation intensity in the flow field and is an important basis for judging the existence of vortex structures. For ease of engineering implementation, the threshold of vorticity is usually set according to the statistical distribution of simulation samples. For example, the recognition threshold is determined by the method of adding the average value and the standard deviation, and the typical value range is between 0.1 and 1.5 S -1 Between.

[0021] The tensor response feature refers to the response quantity obtained by tensorizing the velocity field, pressure field, etc. Common ones include the velocity gradient tensor and the stress tensor. In the present invention, two function spaces are introduced to map the tensor features: one is the original tensor space, which retains the complete second-order tensor information of each measurement point; the other is the combined feature space, which maps the tensor into a single scalar or vector form through eigenvalue functions (such as the sum of squares of eigenvalues) for simplified processing and clustering analysis.

[0022] Spectral Clustering is used to classify vortex features and identify typical vortex patterns existing in the flow field. This method is based on the similarity matrix, and completes dimensionality reduction and clustering by constructing the graph Laplacian matrix and extracting the first k eigenvectors. The similarity calculation generally uses the Gaussian kernel function.

[0023] Vortex Features include but are not limited to: the vortex core position (extreme points of vorticity or Q-criterion), rotational intensity, principal axis direction (principal eigenvector of the tensor), and scale (envelope volume of the characteristic region or projected diameter). The extraction of vortex features can be completed by a fluid simulation software (such as Fluent or OpenFOAM) combined with a post-processing tool (such as ParaView), and used as input for subsequent control model training.

[0024] The Fuzzy Neural Network (FNN), as a control model that combines fuzzy rules and neural structures, is used for the dynamic control of the thruster in the present invention. The network input is the multi-dimensional vortex structure features extracted. The fuzzy layer uses Gaussian or triangular membership functions to fuzzify the input, and the rule layer outputs the response result according to preset or self-learning rules, and finally controls the changes in the rotational speed and angle of attack of the thruster. The network structure can adopt a five-layer form of "input layer - fuzzy layer - rule layer - normalization layer - output layer".

[0025] The adjustment quantity in the present invention is the output variable of the control system, mainly including the change in the rotational speed of the thruster (unit: rpm) and the change in the blade angle of attack (unit: °). This adjustment quantity is output in an incremental form.

[0026] A tensor is a mathematical object that can represent multi-dimensional data relationships, and is a natural generalization of scalars (0th-order tensors) and vectors (1st-order tensors). In fluid mechanics, common tensors include the velocity gradient tensor, stress tensor, strain rate tensor, etc. Taking the velocity gradient tensor as an example, its definition is the partial derivative matrix of the velocity field with respect to the spatial coordinates.

[0027] Tensor Field refers to a physical field in which a tensor is associated with each spatial point, which is a further extension of the vector field. For example, the velocity gradient tensor field is a collection of velocity gradient tensors at each point in a three-dimensional flow field. It has rich flow structure information and can be used for vortex identification and flow anomaly judgment.

[0028] Vortex Scale is used to characterize the spatial size of the vortex region, which can be estimated by envelope diameter, area or distance between characteristic points. For example, it is defined as the maximum circumscribed sphere radius of the region with a curl greater than a threshold.

[0029] Propeller geometric parameters: such as pitch ratio, blade twist angle, blade root / tip thickness ratio, inlet angle, etc., have a direct impact on the flow field disturbance characteristics.

[0030] See also Figure 1 The present application first provides an embodiment of an underwater thruster control method based on tensor identification and fuzzy control, the embodiment comprising: S101. Based on the geometric parameters and operating boundary conditions of the propeller, a numerical simulation model of the wake region is constructed, and simulation is performed to obtain basic flow field data; First, a numerical simulation model of the wake region is constructed based on the geometric parameters of the underwater propeller to be controlled (including the diameter, pitch, number of blades, etc. of the propeller) and its operating boundary conditions (such as propulsion speed, density of the fluid medium, viscosity coefficient, boundary inlet velocity distribution and outlet pressure, etc.). The simulation model uses a three-dimensional incompressible flow field model and introduces sliding grid technology in the numerical solution to adapt to the dynamic grid update when the propeller rotates.

[0031] After performing numerical simulation, the basic data of the three-dimensional flow field in the wake area are obtained, including the velocity vector field (Ux, Uy, Uz) and the pressure scalar field (P). The above data serve as the input basis for subsequent tensor processing and structural identification.

[0032] S102, performing tensor field derivation processing based on the flow field basic data to obtain a rotation tensor and a strain tensor describing rotation and deformation characteristics; Based on the velocity vector field data obtained above, the velocity gradient tensor is calculated for each grid cell, which describes the change trend of the velocity at that point in all directions. Furthermore, based on the velocity gradient tensor, two basic descriptions are obtained through tensor transformation: the rotation tensor Ω and the strain tensor S.

[0033] Among them, the rotation tensor Ω represents the rotation of the local fluid, which is generally defined as the anti-symmetric part of the velocity gradient tensor, while the strain tensor S is the symmetric part of the velocity gradient tensor and is used to describe the deformation degree of the fluid element.

[0034] Specifically, referring to Figure 2 , an optional implementation of this step includes: S1021. Calculate the velocity gradient tensor of each grid cell based on the velocity vectors in the basic flow field data; Based on the obtained three-dimensional velocity vector field V = (u, v, w), calculate the velocity gradient tensor for each grid cell , and this tensor is a 3×3 matrix: ; The above partial derivatives are calculated from the velocity data on the simulation grid by the central difference method or other high-order finite difference schemes.

[0035] S1022. Decompose the velocity gradient tensor into an anti-symmetric part and a symmetric part, where the anti-symmetric part corresponds to the rotation tensor and is used to describe the rotation characteristics of the local fluid, and the symmetric part corresponds to the strain tensor and describes the deformation characteristics of the local fluid; Decompose the local velocity gradient tensor into a symmetric part and an anti-symmetric part, that is: The strain tensor S is the symmetric part of the velocity gradient tensor and is expressed as: ; The rotation tensor Ω is the anti-symmetric part of the velocity gradient tensor and is expressed as:

[0036] Among them, represents the transpose of the velocity gradient tensor. The strain tensor S is used to reflect the stretching and compression deformation of local fluid particles, while the rotation tensor Ω describes the rotational motion characteristics of the local fluid.

[0037] S1023. Calculate the eigenvalues and eigenvectors of the rotation tensor and the strain tensor respectively.

[0038] Perform eigenvalue decomposition on the rotation tensor Ω and the strain tensor S respectively, and calculate their eigenvalues and corresponding eigenvectors.

[0039] Specifically, for each tensor T, solve its characteristic equation: det(T - λI) = 0; Among them, λ is the eigenvalue of the tensor, I is the identity matrix, and det represents the determinant. Three eigenvalues λ1, λ2, and λ3 can be obtained through this equation, and then the rotation dominance and deformation trend of the local fluid structure can be analyzed. The above eigenvalues will serve as the input basis for constructing the local tensor field index group and the function space mapping in the following.

[0040] S103. Map the rotation tensor and the strain tensor to the first function space and the second function space respectively to obtain a local tensor field index group, and determine the target grid area based on the local tensor field index group; Map the above rotation tensor and strain tensor to two different function spaces, denoted as the first function space and the second function space. Use eigenvalue decomposition to extract tensor eigenquantities and construct a local tensor field index group. For example, the following method can be adopted: Perform eigenvalue analysis on the tensor obtained by the sum of S and Ω to obtain a set of eigenvalues; In an optional embodiment, refer to Figure 3 , the implementation method of this step S103 includes: S1031. Map the rotation tensor to the first function space dominated by rotation characteristics. The first function space is an eigenmode domain constructed based on tensor eigenvalue decomposition and is used to reflect the characteristics of local vortex distribution; Map the rotation tensor Ω obtained in step S102 to the first function space dominated by rotation characteristics. This first function space is an eigenmode domain constructed based on tensor eigenvalue decomposition. Essentially, it is a structural representation of the tensor eigenvalues and their corresponding eigenvector directions, and can be used to characterize the directionality, rotation intensity, and rotation continuity of the local fluid rotation axis.

[0041] In this embodiment, the response index in this function space is defined as: ;

[0042] Among them, are the three eigenvalues of the rotation tensor, represents the main response intensity of the local vortex axis direction.

[0043] S1032. Map the strain tensor to the second function space dominated by shear deformation and the direction of principal stress. The second function space is used to represent the anisotropic behavior of local deformation; Map the strain tensor S in step S102 to the second function space dominated by shear deformation and the direction of principal stress. This second function space is used to identify the deformation anisotropy of the fluid in different directions and reflect the main directions of local stretching and compression, the degree of shear strain, and other structural behaviors.

[0044] The response index in this space can be defined as: ; where are the eigenvalues of the strain tensor, represents the tensor strength in the direction of the principal strain. The characteristics of the principal strain axis constructed by the second function space can be used to assist in judging whether there are high-shear regions and whether these regions are coupled with the rotation behavior.

[0045] S1033. Extract the first tensor response eigenvalue related to the direction of the vortex main axis in the first function space, and extract the second tensor response eigenvalue related to the direction of the maximum principal strain in the second function space; Extract the first tensor response eigenvalue related to the direction of the vortex main axis in the above first function space , and extract the second tensor response eigenvalue related to the direction of the maximum principal strain in the second function space , as a characterization of the local tensor behavior.

[0046] To enhance the robustness of the structure determination, a rotation-dominant ratio index can be defined, expressed as: ; where is a small constant set to prevent the denominator from being zero. This index is used to evaluate the dominance of the rotation behavior in the local tensor response.

[0047] S1034. Construct a local tensor field index group based on the first tensor response eigenvalue and the second tensor response eigenvalue to characterize the rotation dominance and shear coupling characteristics of the grid region; Based on the above-extracted , , construct a local tensor field index group, which can characterize the multi-dimensional attributes of each grid cell in terms of rotation dominance and shear coupling behavior.

[0048] S1035. Determine the target grid region based on the local tensor field index group.

[0049] Perform an index traversal analysis on the entire simulation grid domain, and select the grid cells that meet the following constraint conditions as the target grid region: > : The local rotation intensity is higher than the vortex detection threshold; > : Rotation dominates in the tensor behavior; Optional: < .

[0050] Among them, , , are preset empirical thresholds. The grid cells that meet the above conditions will form a rotation-dominated region and enter the subsequent spatial aggregation processing steps S104 and S105.

[0051] In another alternative embodiment, the determination of rotation dominance can be completed by the Q-Criterion or the λ2 criterion; The Q-Criterion is used in the pre-screening stage to quickly exclude the shear-dominated regions and initially screen the regions where Q > 0; The λ2 criterion is used in the fine determination stage to further extract the grid cells where λ2 < 0 within the regions where Q > 0, forming a target grid region that conforms to the rotation dominance trend.

[0052] It should be noted that the first function space and the second function space described in step S103 both originate from the tensor decomposition processing of the velocity gradient tensor. However, different from the traditional vortex criterion which is only used for structure recognition, in this embodiment, by constructing a more parameterized and responsive function space and tensor field index group, the subsequent intelligent control model can input richer, continuous, and coupled information, enhancing the analysis and response ability of the control process to the flow field state.

[0053] S104. Generate a curl feature based on the rotation tensor, and construct a graph structure model based on the curl feature to perform spectral clustering to obtain a candidate grid region, where the curl feature of the candidate grid region reaches a preset threshold; In this embodiment, step S104 includes generating a curl feature for the rotation tensor and constructing a graph structure model based on the curl feature to perform spectral clustering, so as to obtain a candidate grid region with a high curl feature. First, based on the obtained rotation tensor, calculate the curl feature of each grid cell, regard all grid cells as nodes in the graph structure, establish edge connections between the nodes according to the physical adjacency relationship or fluid similarity, and calculate the weights of the edges in the graph by the difference in curl.

[0054] Based on the constructed graph structure model, the spectral clustering algorithm is used to perform clustering analysis on grid cells, specifically including steps such as solving the graph Laplacian matrix, extracting eigenvectors, and K-means clustering. Spectral clustering can aggregate grid cells that are spatially discontinuous but have similar vorticities into regions with consistent features while maintaining the consistency of local vorticity. After clustering, the average vorticity features of each clustering region are statistically analyzed, and the clustering results with an average vorticity exceeding a preset threshold are selected as candidate grid regions. These candidate grid regions have obvious rotational dominance and are the key objects for subsequent vortex structure identification and dynamic control reasoning. Through this step, the system can effectively extract representative rotational regions in a complex flow field, improving the efficiency and accuracy of subsequent processing.

[0055] Specifically, the present application provides an implementation manner of step S104, which specifically includes: Calculating the vorticity feature of each grid cell based on the rotation tensor, where the vorticity feature is used to characterize the rotation intensity of the local flow field; constructing a graph structure model with grid cells containing vorticity features, where the nodes in the graph structure correspond to grid cells, and the edge weights are set based on the difference in vorticity features between adjacent nodes; applying the spectral clustering algorithm to the graph structure model and performing unsupervised partitioning on the grid cells to obtain clustering results; screening out the clustering clusters with vorticity features exceeding a preset threshold from the clustering results and determining them as candidate grid regions.

[0056] In this implementation manner, the vorticity feature of each grid cell is calculated based on the rotation tensor, and the vorticity feature is used to characterize the rotation intensity in the local flow field; grid cells containing vorticity features are constructed into a graph structure model, where the nodes in the graph structure correspond to each grid cell, and the weights of the edges are set according to the difference in vorticity features between adjacent nodes to reflect the similarity between nodes; the spectral clustering algorithm is applied to this graph structure model, and unsupervised learning is used to divide the grid cells to obtain multiple clustering results; finally, the clustering clusters with vorticity features exceeding a preset threshold are screened out from the clustering results, and these clustering clusters are determined as candidate grid regions with significant rotation features for subsequent vortex identification and control strategy formulation.

[0057] S105. Aggregate each grid cell in the candidate grid region and determine the target vortex region, where the target vortex region has rotational dominance in the candidate grid region; perform spatial aggregation processing on the target grid region, and aggregate adjacent continuous rotation-dominant grid cells into vortex clusters to form one or more closed or semi-closed vortex regions.

[0058] For each vortex cluster region, further extract its corresponding structural parameters to form target multi-dimensional vortex structure feature data. The structural data may include but is not limited to: vortex center position, radius, vortex axis direction; rotational intensity within the vortex, axial velocity distribution; local pressure gradient and velocity vorticity, etc.

[0059] S106. For the identified target vortex regions, extract multi-dimensional vortex structure feature data including at least vortex scale, vortex intensity, rotational axis, and vortex core position. The data extracted from the target vortex structure, combined with the propulsion efficiency index (such as propulsion distance per unit power consumption) in the current thruster operating state and the detected flow field disturbance parameters (such as oncoming flow changes, wake fluctuation frequency), jointly construct a multi-dimensional state vector.

[0060] This input vector serves as the input dimension of the fuzzy neural network and can dynamically reflect the coupling relationship between the vortex structure and the propulsion environment.

[0061] Specifically, referring to Figure 4 One optional implementation method includes: S1061. Collect target multi-dimensional vortex structure feature data, where the target multi-dimensional vortex structure feature data includes vortex intensity, vortex core position, vortex core scale, and local energy consumption density. For each target vortex region extracted from step S105, obtain its corresponding key structural parameters to form a target multi-dimensional vortex structure feature data set. This data includes but is not limited to the following indicators: Vortex intensity ωs: It can be calculated through the vorticity magnitude, axial velocity gradient, or local rotation tensor norm, and is used to characterize the rotational energy level of the vortex. Vortex core position Xc: It is the spatial coordinate point corresponding to the dominant eigenvalue of rotation (such as the minimum value of λ2 < 0) in the vortex cluster. Vortex core scale Rc: It represents the local size range of the vortex structure and can be estimated by the maximum response radius or characteristic length. Local energy consumption density Ed: It is calculated based on the strain tensor magnitude and the shear power consumption function, and reflects the energy dissipation degree of this region.

[0062] S1062. Collect the current propulsion efficiency index and flow field disturbance parameters. Simultaneously collect the working state parameters of the current thruster, including but not limited to: Unit propulsion efficiency η = PVf, where Vf is the propulsion speed and P is the power consumption per unit time. Wake disturbance frequency fw: Obtained by spectral analysis of the main frequency component of the velocity fluctuation in the flow field. Inflow disturbance angle θin: Record the offset angle of the inflow direction, which reflects the influence of attitude changes on the flow field structure. Fluid Reynolds number Re: Calculated from the propulsion speed, characteristic length, and fluid viscosity parameter to unify the influence of dimensionality.

[0063] S1063. Normalize the collected data; Normalize the collected vortex structure parameters, propulsion efficiency indicators, and perturbation parameters respectively to eliminate the coupling interference between different dimensions and physical quantities. Optional methods include min-max linear normalization or Z-score standardization.

[0064] S1064. Combine the normalized target multi-dimensional vortex structure characteristic data, propulsion efficiency indicators, and flow field perturbation parameters to construct a multi-dimensional state vector.

[0065] The normalized target multi-dimensional vortex structure characteristic data {ωs, Xc, Rc, Ed}, propulsion efficiency indicator η, and flow field perturbation parameters: {fw, θin, Re}; Input as a whole vector according to the preset dimension combination to form the multi-dimensional state vector S_input for the intelligent inference model: S_input = [ωs, Rc, Ed, η, fw, θin, Re].

[0066] Optionally, introduce the feature change rate of the historical frame (such as the change rate of vortex intensity) or the change rate of propulsion efficiency in this input vector to enhance the model's perception ability of the time dynamic trend.

[0067] This input vector is used for the fuzzy neural network dynamic inference in the subsequent step S106, which can reflect the coupling state between the vortex structure and the propulsion environment in real time, so as to drive the intelligent adjustment of the propulsion parameters.

[0068] S107. Collect the current propulsion efficiency indicators and flow field perturbation parameters, and construct a multi-dimensional state vector in combination with the multi-dimensional vortex structure characteristic data; In this embodiment, to realize the intelligent adjustment and control of the propulsion state, after the system completes the identification of the vortex region and the extraction of its structural characteristics, it further collects a number of parameters including the current operating efficiency of the thruster and the wake perturbation characteristics, and fuses them with the multi-dimensional vortex structure characteristic data to construct a multi-dimensional state vector for the input of the intelligent model.

[0069] Specifically, it includes the following steps: Collection of propulsion efficiency indicators: The system obtains the propulsion efficiency data of the current thruster under the target operating conditions in real time. The efficiency can be expressed by the ratio of the net thrust output work per unit time to the system power consumption, and can also include the blade propulsion efficiency, the forward speed under unit input power consumption, and other efficiency parameters related to the operating conditions.

[0070] Obtaining flow field perturbation parameters: By deploying pressure sensors and flow velocity monitoring modules in the wake region or through simulation model deduction, local flow velocity fluctuations, pressure gradient changes, Reynolds number distribution, velocity shear, and other perturbation characteristic quantities are obtained to reflect the flow field stability and vortex interference intensity in the wake.

[0071] Combined with the vortex structure data obtained in the previous step, including vortex scale, vorticity intensity, main axis direction of the vortex, position center of gravity coordinates, principal values of the rotation tensor, etc., these are used as the structural dimension indicators of the state vector.

[0072] The above three types of information are constructed into a state vector according to a preset data splicing strategy. This vector is used as the input of the subsequent fuzzy neural network control model to infer the corresponding relationship between the current state and the target control strategy.

[0073] Through this step, the conversion from physical space observation data to the input of the control space is realized, enabling the control model to maintain accurate response under conditions of strong perturbation and complex wake structures, and improving the operation robustness of the thruster.

[0074] S108: Input the multi-dimensional state vector into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between the vortex intensity and the operation state of the thruster, and output the target rotational speed adjustment amount and blade angle of attack adjustment amount for adjusting the operation state of the thruster.

[0075] The pre-configured adaptive fuzzy neural network model includes multiple input membership functions, a fuzzy rule set, and an output adjustment layer.

[0076] Through the fuzzy inference engine, after inputting the multi-dimensional state vector, fuzzification, rule matching, and defuzzification operations are performed to obtain the control parameters for thruster adjustment.

[0077] The control parameters include: Target rotational speed adjustment amount: The rotational speed value for instructing the thruster to increase or decrease. Blade angle of attack adjustment amount: Adjust the tilt angle of the propeller blade to adapt to the flow field change.

[0078] The system adjusts the operation state of the thruster in real time according to the network output to achieve the purpose of improving propulsion efficiency, reducing energy consumption, and weakening adverse wake interference, thereby realizing intelligent and adaptive control.

[0079] Specifically, it includes: performing fuzzification processing on each variable in the multi-dimensional state vector and mapping it to the corresponding fuzzy set to form fuzzy input variables; calculating the activation strength of each fuzzy rule based on the preset fuzzy rule base and the fuzzy input variables; performing comprehensive simulation on the activated fuzzy rules through the fuzzy simulation module in the adaptive fuzzy neural network model to obtain an output fuzzy set; and performing defuzzification processing on the output fuzzy set to obtain numerical target rotational speed adjustment amount and blade angle of attack adjustment amount.

[0080] For each input variable (such as normalized vortex intensity, energy consumption density, propulsion efficiency, perturbation angle, etc.), define the corresponding membership function set; optional membership functions include triangular function, trapezoidal function, Gaussian function, etc., which are respectively used to describe the membership degree of the variable at different fuzzy linguistic values (such as "low", "medium", "high"); For example, the vortex intensity can be fuzzified into three types of linguistic values: Low, Medium, High, corresponding to different membership function ranges. Finally, each input variable is mapped to one or more fuzzy input variables.

[0081] After fuzzification, the system calls the preset fuzzy rule base for reasoning processing. The rule base consists of several rules in the following form: IF the vortex intensity is High AND the propulsion efficiency is Low AND the energy consumption density is High THEN decrease the rotational speed AND increase the angle of attack; IF the vortex intensity is Medium AND the flow field perturbation is Low THEN maintain the current rotational speed AND slightly adjust the angle of attack; For each fuzzy rule, calculate its activation strength, that is, the combination of the membership degrees of all preconditions (min operation or product operation can be used).

[0082] Input all the activated fuzzy rules and their corresponding activation strengths into the fuzzy simulation module in the neural network. This module has the ability to adaptively adjust weights, and dynamically adjusts the weights between rules and the output offset through the network learning process to improve the reasoning accuracy.

[0083] Integrate the activation situations of multiple rules to construct an output fuzzy set, which forms a fuzzy description of the target output parameters (i.e., rotational speed adjustment amount, blade angle of attack adjustment amount).

[0084] The linguistic values of the output fuzzy quantity can also be set as: Decrease, Maintain, Increase; Each linguistic value corresponds to an output membership function to form the final fuzzy output set.

[0085] Perform defuzzification on the above fuzzy output set to convert the fuzzy language output into a numerical control instruction. Optional defuzzification methods include: Centroid Method: Calculate the output value based on the centroid of the area of the output membership function graph; Max Membership Principle: Select the representative value corresponding to the output linguistic value with the largest membership degree; Weighted average method: Numerically map multiple output linguistic values by weighted average according to the activation intensity.

[0086] Finally, obtain a numerical adjustment instruction: Target rotational speed adjustment amount: Unit is rpm; Blade angle of attack adjustment amount: Unit is degree (°).

[0087] The system transmits these two parameters to the thruster control module in real time to adjust its current operating state to adapt to the changes in vortex characteristics, suppress the interference of unstable wakes, and optimize the propulsion efficiency and energy consumption performance.

[0088] In an optional embodiment, to further improve the accuracy and robustness of vortex structure recognition and enhance the stability of subsequent control instruction generation, step S105 can adopt a more specific implementation manner to complete the spatial aggregation and feature structure extraction of the target grid area. Refer to Figure 5 , the following combines an optional embodiment to illustrate the specific implementation process of this step: S1051. According to the spatial proximity relationship, divide multiple adjacent grid cells that meet the condition of rotational dominance into a connected region as a candidate vortex region; First, for the set of grid cells that have passed the previous-stage rotational dominance criteria (such as Q-criterion, λ2-criterion, etc.), perform connectivity clustering processing based on their adjacency relationship in three-dimensional space. Specifically, the criteria of face adjacency or vertex adjacency can be adopted to merge spatially connected grid cells into a connected subset; each connected subset is defined as a candidate vortex region, representing a local region cluster that may form a vortex; in actual implementation, connected region recognition algorithms such as depth-first search (DFS) or union-find set can be used to implement the grid aggregation process.

[0089] S1052. Calculate the comprehensive vortex intensity index within each candidate vortex region, and the comprehensive vortex intensity index includes vortex core position coordinates, vortex intensity, vortex core scale, and local energy consumption density; For each candidate vortex region, further extract and calculate the representative physical quantities inside it to describe its overall vortex behavior. The comprehensive vortex intensity index includes but is not limited to the following: Vortex core position coordinates: The vortex center position is extracted through the local velocity vorticity extreme point, and the rotation axis coordinates are estimated by combining the flow velocity distribution; Vortex intensity: Calculate the total rotation amount or integral vorticity within the candidate region, and it can be estimated in integral form: ; where is the local vorticity vector.

[0090] The vortex core scale is based on the volume size of the high vorticity region within a certain threshold range from the vortex core center in the candidate region, and the corresponding radius or diameter of the vortex core is determined; The local energy consumption density is obtained by performing a volume integral of the viscous dissipation rate within the region, reflecting the degree of energy consumption of the local flow.

[0091] S1053. According to the preset vortex intensity threshold and vortex core scale threshold, screen out the target vortex regions that meet the conditions of the vortex intensity threshold and rotation dominance.

[0092] According to the set vortex intensity threshold and vortex core scale threshold, perform secondary screening on each candidate vortex region: If the comprehensive vortex intensity of a certain candidate region exceeds the vortex intensity threshold and its vortex core scale is within a reasonable physical range (for example, greater than the minimum numerical resolution unit and less than the maximum scale of the wake region), then this region is determined as the target vortex region; at the same time, this target region needs to meet the verification of the previous rotation dominance condition to ensure that its rotation characteristics are superior to other flow behaviors such as shear and diffusion; Finally, extract all physical characteristic indexes of this target vortex region to form a target multi-dimensional vortex structure characteristic data set, which is used as the input basis for subsequent control modeling and reasoning.

[0093] Refer to Figure 6 , the following describes the embodiments of the system provided in this application, and this embodiment includes: A simulation unit 601, configured to construct a numerical simulation model of the wake region based on the geometric parameters and operating boundary conditions of the thruster, and perform a simulation to obtain basic flow field data; A tensor field unit 602, configured to perform tensor field derivation processing based on the basic flow field data to obtain a rotation tensor and a strain tensor describing rotation and deformation characteristics; A function mapping unit 603, configured to map the rotation tensor and the strain tensor to a first function space and a second function space respectively, obtain a local tensor field index group, and determine a target grid region based on the local tensor field index group; A pre-screening unit 604, configured to generate curl features according to the rotation tensor, and construct a graph structure model based on the curl features to perform spectral clustering to obtain candidate grid regions, where the curl features of the candidate grid regions reach a preset threshold; A grid cell aggregation unit 605, configured to perform an aggregation process on each grid cell in the candidate grid regions, and determine a target vortex region, where the target vortex region has rotational dominance in the candidate grid regions; A data extraction unit 606, configured to extract multi-dimensional vortex structure feature data including at least vortex scale, vortex intensity, rotation axis direction, and vortex core position for the identified target vortex region; A data acquisition unit 607, configured to acquire current propulsion efficiency indicators and flow field disturbance parameters, and construct a multi-dimensional state vector in combination with the multi-dimensional vortex structure feature data; A model input / output unit 608, configured to input the multi-dimensional state vector into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between vortex intensity and the operating state of the thruster, and output a target rotational speed adjustment amount and a blade angle of attack adjustment amount for adjusting the operating state of the thruster.

[0094] Optionally, the function mapping unit 603 is specifically configured to: Map the rotation tensor to a first function space dominated by rotation features, where the first function space is an eigenmode domain constructed based on tensor eigenvalue decomposition and is used to reflect local vortex distribution characteristics; Map the strain tensor to a second function space dominated by shear deformation and principal stress directions, where the second function space is used to represent the anisotropic behavior of local deformation; Extract a first tensor response eigenvalue related to the vortex principal axis direction in the first function space, and extract a second tensor response eigenvalue related to the maximum principal strain direction in the second function space; Construct a local tensor field index group based on the first tensor response eigenvalue and the second tensor response eigenvalue, for characterizing the rotational dominance and shear coupling characteristics of the grid region; Determine a target grid region based on the local tensor field index group.

[0095] Optionally, the function mapping unit 603 is specifically configured to: For each grid cell in the wake region, respectively obtain the corresponding first tensor response eigenvalue and second tensor response eigenvalue; Based on a preset functional relationship, perform a combined operation on the first tensor response eigenvalue and the second tensor response eigenvalue to obtain a combined eigenvalue for each grid cell; Screen out grid cells that do not meet the condition of rotation dominance according to the relationship between the combined eigenvalue and the first threshold; For the retained grid cells, calculate the included angle information between the principal direction of the rotation tensor and the principal direction of the strain tensor; Screen out the target grid region that meets the condition of rotation dominance according to the included angle information.

[0096] Optionally, the tensor field unit 602 is specifically configured to: Based on the velocity vectors in the basic flow field data, calculate the velocity gradient tensor of each grid cell; Decompose the velocity gradient tensor into an anti-symmetric part and a symmetric part, where the anti-symmetric part corresponds to the rotation tensor and is used to describe the rotation characteristics of the local fluid, and the symmetric part corresponds to the strain tensor and describes the deformation characteristics of the local fluid; Calculate the eigenvalues and eigenvectors of the rotation tensor and the strain tensor respectively.

[0097] Optionally, the condition for meeting the rotation dominance includes that the combined eigenvalue is greater than or equal to the first threshold.

[0098] Optionally, the grid cell aggregation unit 606 is specifically configured to: According to the spatial proximity relationship, divide multiple adjacent grid cells that meet the condition of rotation dominance into a connected region as a candidate vortex region; Calculate the comprehensive vortex intensity index within each candidate vortex region, and the comprehensive vortex intensity index includes the coordinates of the vortex core position, the vortex intensity, the vortex core scale, and the local energy consumption density; Screen out the target vortex region that meets the vortex intensity threshold and the condition of rotation dominance according to the preset vortex intensity threshold and vortex core scale threshold.

[0099] Optionally, the model input-output unit 608 is specifically configured to: Perform fuzzification processing on each variable in the multi-dimensional state vector and map it to the corresponding fuzzy set to form fuzzy input variables; According to the preset fuzzy rule base, determine the activation intensity of each fuzzy rule in the fuzzy rule base based on the fuzzy input variables; Determine the target fuzzy rule based on the activation intensity, and perform comprehensive simulation on the target fuzzy rule through the fuzzy simulation module in the adaptive fuzzy neural network model to obtain the output fuzzy set; Perform defuzzification processing on the output fuzzy set to obtain the numerical target rotational speed adjustment amount and blade angle of attack adjustment amount.

[0100] Optionally, the pre-screening unit 604 is specifically configured to: Calculate the curl feature of each grid cell based on the rotation tensor, where the curl feature is used to characterize the rotation intensity of the local flow field; Construct the grid cells containing the curl feature into a graph structure model, where the nodes in the graph structure model correspond to the grid cells, and the edge weights are set based on the difference in the curl features between adjacent nodes; Apply the spectral clustering algorithm to the graph structure model and perform unsupervised partitioning on the grid cells to obtain a clustering result; Screen out the clustering clusters with curl features exceeding a preset threshold from the clustering result and determine them as candidate grid regions.

[0101] Please refer to Figure 7 , this application also provides an underwater thruster control system based on tensor recognition and fuzzy control, including: A processor 701, a memory 702, an input / output unit 703, and a bus 704; The processor 701 is connected to the memory 702, the input / output unit 703, and the bus 704; The memory 702 stores a program, and the processor 701 calls the program to execute any of the above methods.

[0102] This application also relates to a computer-readable storage medium on which a program is stored. When the program runs on a computer, the computer is enabled to execute any of the above methods.

[0103] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0104] In several embodiments provided by this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0105] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0106] In addition, each functional unit in various embodiments of the present application may be integrated into one processing unit, may exist physically alone for each unit, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0107] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

Claims

1. An underwater thruster control method based on tensor recognition and fuzzy control, characterized in that, Including: Based on the geometric parameters and operating boundary conditions of the thruster, a numerical simulation model of the wake region is constructed, and simulations are performed to obtain basic flow field data; Based on the basic flow field data, tensor field derivation processing is performed to obtain a rotation tensor and a strain tensor that describe rotation and deformation characteristics; The rotation tensor and the strain tensor are respectively mapped to a first function space and a second function space to obtain a local tensor field index group, and a target grid region is determined based on the local tensor field index group; A curl feature is generated according to the rotation tensor, and a graph structure model is constructed based on the curl feature to perform spectral clustering to obtain a candidate grid region, and the curl feature of the candidate grid region reaches a preset threshold; Aggregation processing is performed on each grid cell in the candidate grid region, and a target vortex region is determined, and the target vortex region has rotational dominance in the candidate grid region; For the identified target vortex region, multi-dimensional vortex structure characteristic data including at least vortex scale, vortex intensity, rotation axis direction, and vortex core position are extracted; The current propulsion efficiency index and flow field perturbation parameters are collected, and a multi-dimensional state vector is constructed in combination with the multi-dimensional vortex structure characteristic data; The multi-dimensional state vector is input into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between the vortex intensity and the operating state of the thruster, and a target rotational speed adjustment amount and a blade angle of attack adjustment amount for adjusting the operating state of the thruster are output.

2. The underwater thruster control method based on tensor recognition and fuzzy control according to claim 1, characterized in that The generating a curl feature according to the rotation tensor, and constructing a graph structure model based on the curl feature to perform spectral clustering to obtain a candidate grid region, where the curl feature of the candidate grid region reaches a preset threshold, includes: Calculating the curl feature of each grid cell based on the rotation tensor, and the curl feature is used to characterize the rotation intensity of the local flow field; Constructing a grid cell containing the curl feature into a graph structure model, where the nodes in the graph structure model correspond to grid cells, and the edge weights are set based on the difference in curl features between adjacent nodes; Applying a spectral clustering algorithm to the graph structure model, and performing unsupervised partitioning on the grid cells to obtain a clustering result; Selecting clustering clusters with curl features exceeding a preset threshold from the clustering result, and determining them as candidate grid regions.

3. The underwater thruster control method based on tensor recognition and fuzzy control according to claim 1, characterized in that The mapping the rotation tensor and the strain tensor to a first function space and a second function space respectively, obtaining a local tensor field index group, and determining a target grid region based on the local tensor field index group includes: Mapping the rotation tensor to a first function space dominated by rotation features, and the first function space is a characteristic mode domain constructed based on tensor eigenvalue decomposition, and is used to reflect local vortex distribution characteristics; Mapping the strain tensor to a second function space dominated by shear deformation and principal stress direction, and the second function space is used to represent the anisotropic behavior of local deformation; Extracting a first tensor response eigenvalue related to the vortex main axis direction in the first function space, and extracting a second tensor response eigenvalue related to the maximum principal strain direction in the second function space; Construct a local tensor field index group based on the first tensor response eigenvalue and the second tensor response eigenvalue to characterize the rotation dominance and shear coupling characteristics of the grid region; Determine the target grid region based on the local tensor field index group.

4. The underwater thruster control method based on tensor recognition and fuzzy control according to claim 3, wherein, The determining the target grid region based on the local tensor field index group includes: For each grid cell in the wake region, obtain the corresponding first tensor response eigenvalue and second tensor response eigenvalue respectively; Based on a preset functional relationship, perform a combined operation on the first tensor response eigenvalue and the second tensor response eigenvalue to obtain the combined eigenvalue of each grid cell; According to the relationship between the combined eigenvalue and the first threshold, filter out the grid cells that do not meet the condition of rotation dominance; For the remaining grid cells, calculate the included angle information between the main direction of the rotation tensor and the main direction of the strain tensor; Based on the included angle information, screen out the target grid regions that meet the condition of rotation dominance.

5. The underwater thruster control method based on tensor recognition and fuzzy control according to claim 1, wherein The performing tensor field derivation processing based on the basic flow field data to obtain the rotation tensor and the strain tensor describing the rotation and deformation characteristics includes: Based on the velocity vectors in the basic flow field data, calculate the velocity gradient tensor of each grid cell; Decompose the velocity gradient tensor into an anti-symmetric part and a symmetric part, where the anti-symmetric part corresponds to the rotation tensor, which is used to describe the rotation characteristics of the local fluid, and the symmetric part corresponds to the strain tensor, which describes the deformation characteristics of the local fluid; Calculate the eigenvalues and eigenvectors of the rotation tensor and the strain tensor respectively.

6. The underwater thruster control method based on tensor recognition and fuzzy control according to claim 4, characterized in that The condition for meeting the rotation dominance includes that the combined eigenvalue is greater than or equal to the first threshold.

7. The underwater thruster control method based on tensor recognition and fuzzy control according to claim 1, wherein Perform an aggregation process on each grid cell in the candidate grid region and determine the target vortex region. The target vortex region has rotation dominance in the candidate grid region includes: According to the spatial proximity relationship, divide multiple adjacent grid cells that meet the condition of rotation dominance into a connected region as the candidate vortex region; Calculate the comprehensive vortex intensity index within each candidate vortex region. The comprehensive vortex intensity index includes the vortex core position coordinates, the vortex intensity, the vortex core scale, and the local energy consumption density; According to the preset vortex intensity threshold and vortex core scale threshold, screen out the target vortex regions that meet the vortex intensity threshold and the condition of rotation dominance.

8. The underwater thruster control method based on tensor recognition and fuzzy control according to claim 1, characterized in that The inputting the multi-dimensional state vector into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between the vortex intensity and the operating state of the thruster, and output the target rotational speed adjustment amount and blade attack angle adjustment amount for adjusting the operating state of the thruster includes: Perform a fuzzification process on each variable in the multi-dimensional state vector and map it to the corresponding fuzzy set to form fuzzy input variables; According to the preset fuzzy rule base, determine the activation strength of each fuzzy rule in the fuzzy rule base based on the fuzzy input variables; Based on the activation strength, determine the target fuzzy rule, and perform a comprehensive simulation on the target fuzzy rule through the fuzzy simulation module in the adaptive fuzzy neural network model to obtain the output fuzzy set; Defuzzify the output fuzzy set to obtain the numerical target rotational speed adjustment amount and blade angle of attack adjustment amount.

9. An underwater thruster control system based on tensor recognition and fuzzy control, characterized in that, It includes: A simulation unit for constructing a numerical simulation model of the wake region based on the geometric parameters and operating boundary conditions of the thruster, and performing simulations to obtain basic flow field data; A tensor field unit for performing tensor field derivation processing based on the basic flow field data to obtain a rotation tensor and a strain tensor describing rotation and deformation characteristics; A function mapping unit for mapping the rotation tensor and the strain tensor to a first function space and a second function space respectively, obtaining a local tensor field index group, and determining a target grid region based on the local tensor field index group; A pre-screening unit for generating a curl feature according to the rotation tensor, and constructing a graph structure model based on the curl feature to perform spectral clustering to obtain candidate grid regions, where the curl feature of the candidate grid regions reaches a preset threshold; A grid cell aggregation unit for aggregating each grid cell in the candidate grid regions and determining a target vortex region, where the target vortex region has rotational dominance in the candidate grid regions; A data extraction unit for extracting multi-dimensional vortex structure feature data including at least vortex scale, vortex intensity, rotation axis direction, and vortex core position for the identified target vortex region; A data acquisition unit for collecting the current propulsion efficiency index and flow field disturbance parameters, and constructing a multi-dimensional state vector in combination with the multi-dimensional vortex structure feature data; A model input-output unit for inputting the multi-dimensional state vector into a pre-configured adaptive fuzzy neural network model to dynamically infer the correlation between the vortex intensity and the operating state of the thruster, and outputting the target rotational speed adjustment amount and blade angle of attack adjustment amount for adjusting the operating state of the thruster.

10. An underwater thruster control system based on tensor recognition and fuzzy control, characterized in that, It includes: A processor, a memory, an input-output unit, and a bus; The processor is connected to the memory, the input-output unit, and the bus; The memory stores a program, and the processor calls the program to execute the method according to any one of claims 1 to 8.

Images