An adaptive regulation system for multi-anchor cable tension in the height-limited section of a cutter suction dredger

Abstract

The present invention discloses a multi-anchor cable tension adaptive control system for the height-limited section of a cutter suction dredger, which relates to the technical field of intelligent marine engineering equipment and includes a collection module that collects ship environmental situation data, performs timestamp alignment, noise filtering, and coordinate system unification to generate a multi-modal environment data packet; a model construction module that, based on the multi-modal environment data packet, uses a potential field modeling engine to construct a three-dimensional dynamic potential field model, generates a potential field distribution matrix through the finite difference method, and at the same time uses central difference to calculate the partial derivatives of the potential field distribution matrix to generate a three-dimensional potential field gradient tensor; and a path planning module that, based on the three-dimensional potential field gradient tensor, uses the RRT algorithm to iteratively generate candidate path nodes and performs smoothing processing through the cubic spline interpolation algorithm to generate an initial navigation path. The present invention ensures that the anchor cable tension can be dynamically adjusted according to the actual environmental changes by constructing a three-dimensional dynamic potential field model, thereby improving the stability and safety during the operation process.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent marine engineering equipment, and particularly to a multi-anchor cable tension adaptive regulation system for the height-limited section of a cutter suction dredger. Background Art

[0002] With the continuous deepening of marine development and port construction, the cutter suction dredger, as an important dredging tool, plays an important role in seabed excavation, waterway widening and maintenance. In recent years, in order to improve the operation efficiency and safety of the cutter suction dredger in complex environments, various innovative solutions have emerged in the relevant technical fields.

[0003] For example, the real-time environmental monitoring method achieved through sensor fusion can significantly improve the ship's perception ability of the surrounding environment, and the use of machine learning algorithms to optimize the path planning process has achieved a more efficient and safer navigation route design. In traditional methods, the adjustment of the anchor cable tension during the operation of the cutter suction dredger usually depends on the manual adjustment by experienced operators, which is not only inefficient but also difficult to ensure the operation accuracy and safety. Although the development of automation control technology has enabled some automation methods to be applied to assist in the regulation of the anchor cable tension, they often lack flexibility and cannot adapt to the dynamically changing working environment, especially when encountering complex working conditions such as height-limited sections, and the existing automation control mechanisms also have obvious limitations in dealing with the integration of multi-source heterogeneous data and achieving precise path planning. Specifically, the current automation control mechanisms are lacking in the ability to integrate various types of data and generate the optimal navigation path accordingly. In addition, since most of these methods are based on preset rules or static models and lack the real-time response ability to dynamic environmental changes, manual intervention is often required in actual operation, increasing the operation difficulty and risk.

[0004] There are mainly two deficiencies in the existing technology. On the one hand, the traditional automation control mechanism shows obvious inadaptability when facing complex marine engineering environments, especially when it comes to operations in height-limited sections, and it cannot automatically adjust the anchor cable tension according to the actual situation, thus affecting the safety and efficiency of the operation. On the other hand, the existing technology is lacking in the ability to integrate multi-modal environmental data, which limits the accuracy of the subsequent generation of the optimal navigation path. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a multi-anchor cable tension adaptive regulation system for the height-limited section of a cutter suction dredger to solve the problems of insufficient adaptive adjustment ability of the anchor cable tension and insufficient multi-source data fusion.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] The present invention provides a multi-anchor cable tension adaptive control system for a height-limited section of a cutter suction vessel, which comprises a collection module for collecting ship environmental situation data, performing timestamp alignment, noise filtering and coordinate system one, and generating a multi-modal environmental data packet;

[0009] The model building module uses the potential field modeling engine to build a three-dimensional dynamic potential field model based on the multimodal environment data package, and generates the potential field distribution matrix through the finite difference method. At the same time, the partial derivatives of the potential field distribution matrix are calculated using the central difference to generate the three-dimensional potential field gradient tensor;

[0010] The path planning module uses the RRT algorithm to iteratively generate candidate path nodes based on the three-dimensional potential field gradient tensor, and then smoothes them through the cubic spline interpolation algorithm to generate the initial navigation path;

[0011] The instruction generation module inputs the initial navigation path and real-time dredger operation data into the dual-stream neural network, extracts spatial topological features using the spatial stream branch, extracts dynamic correlation features using the temporal stream branch, fuses the spatial topological features and dynamic correlation features, and inputs them into the strategy network to output the adjustment instruction set of the anchor cable tension;

[0012] In the anchor cable tension control module, the potential field modeling engine parses the anchor cable tension adjustment instruction set, generates a new potential field distribution matrix and three-dimensional potential field gradient tensor, and inputs them into the two-stream neural network. The two-stream neural network performs parameter updates and path replanning to generate the final navigation path.

[0013] As a preferred solution of the multi-anchor cable tension adaptive control system in the height-limited section of the dredger suction vessel described in the present invention, the ship environmental situation data includes three-dimensional point cloud data of the height-limited section, sonar data of underwater obstacles and dredger suction vessel operation data.

[0014] As a preferred solution of the multi-anchor cable tension adaptive control system for the height-limited section of the cutter suction vessel of the present invention, wherein: the generation of the multi-modal environmental data packet specifically includes the following steps:

[0015] The ship environment situation data is time-stamped and aligned through the PTP protocol. The three-dimensional point cloud data in the height-restricted section is filtered for outliers through the rectangular information granulation algorithm. The sonar data of underwater obstacles is denoised using the wavelet threshold. The operation data of the cutter suction vessel is clustered using the DBSCAN algorithm to remove outliers.

[0016] The ICP algorithm is used to unify the preprocessed ship environmental situation data into the geographic coordinate system, and a hierarchical fusion strategy is adopted to integrate them to generate a multimodal environmental data package.

[0017] As a preferred solution of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger according to the present invention, wherein: a potential field distribution matrix is generated by the finite difference method, and the specific steps are as follows,

[0018] Based on the multi-modal environment data packet, a standardized three-dimensional voxel map is constructed by using the point cloud voxelization downsampling method, and a spatial index is established through the KD-Tree. At the same time, a three-dimensional dynamic potential field model is constructed by combining the dynamic weight allocation strategy with the hyperbolic secant function;

[0019] Based on the three-dimensional dynamic potential field model, the spatial gradient field is calculated by the finite difference method to obtain the gravitational gradient and the repulsive field strength. At the same time, based on the gravitational gradient and the repulsive field strength, a potential field distribution matrix is generated by the multi-physical field coupling and superposition algorithm.

[0020] As a preferred solution of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger according to the present invention, wherein: the generation of the three-dimensional potential field gradient tensor specifically includes the following steps,

[0021] The central difference in the finite difference method is used to calculate the first-order partial derivative of the potential field distribution matrix, and the initial gradient vector field is obtained through the vector synthesis algorithm based on the first-order partial derivative;

[0022] The initial gradient vector field is reconstructed by the ADMM coupling reconstruction method, the second-order partial derivative is cross-calculated by the central difference method, and the weight distribution ratio of the second-order partial derivative is dynamically adjusted by using the hyperbolic secant function. At the same time, regularization processing is carried out by the alternating direction multiplier method to generate a three-dimensional potential field gradient tensor.

[0023] As a preferred solution of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger according to the present invention, wherein: the generation of the initial navigation path specifically includes the following steps,

[0024] The RRT algorithm is used to expand the nodes of the three-dimensional potential field gradient tensor to obtain the dynamic direction vector, and the dynamic direction vector is converted into the path expansion direction by the elliptical sampling domain conversion method. At the same time, the adaptive step size strategy is adopted to obtain the node expansion step size;

[0025] Candidate path nodes are generated according to the path expansion direction and the node expansion step size through the hierarchical search strategy;

[0026] The candidate path nodes are processed by the cubic spline interpolation algorithm to generate a cubic spline smooth trajectory;

[0027] Based on the cubic spline smooth trajectory, dynamic regularization optimization is carried out by the ADMM algorithm to generate the initial navigation path.

[0028] As a preferred solution of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger described in the present invention, wherein: the steps of extracting spatial topological features by using the spatial flow branch are as follows.

[0029] Input the initial navigation path into the spatial flow branch of the two-stream neural network. The first layer of the spatial flow branch performs spatio-temporal joint convolution and spatial dimensionality reduction to capture spatial structure features. The second layer performs local iterative optimization and cross-layer feature fusion to obtain refined local features.

[0030] The third layer dynamically assigns different scale weights to the spatial structure features and the refined local features through the channel attention mechanism, and fuses them through splicing channels to generate spatial topological features.

[0031] As a preferred solution of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger described in the present invention, wherein: the steps of extracting dynamic correlation features by using the temporal flow branch are as follows.

[0032] Input the real-time operating data of the cutter suction dredger into the temporal flow branch of the two-stream neural network. The forward TCN of the bidirectional long short-term memory network captures long-range trends through dilated convolution, and the backward TCN models reverse degradation patterns through causal convolution.

[0033] The long-range trends and reverse degradation patterns are bidirectionally spliced through the temporal attention mechanism to generate dynamic correlation features.

[0034] As a preferred solution of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger described in the present invention, wherein: the steps of outputting the adjustment instruction set of the anchor cable tension are as follows.

[0035] The spatial topological features and the dynamic temporal features are weighted and fused and then input into the DDPG policy network. The Actor-Critic collaborative optimization framework is used for policy gradient update to generate the anchor cable tension adjustment parameter vector. Based on the anchor cable tension adjustment parameter vector, the anchor cable tension adjustment instruction set is generated through the finite element algorithm.

[0036] As a preferred solution of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger described in the present invention, wherein: the steps of generating the final navigation path are as follows.

[0037] Based on the adjustment instruction set of the anchor cable tension, a new potential field distribution matrix is generated through the finite element tension field coupling algorithm, and the implicit Euler method is used for nonlinear calculation to generate a new three-dimensional potential field gradient tensor.

[0038] The new potential field distribution matrix and the three-dimensional potential field gradient tensor are input into the two-stream neural network, and the elastic weight consolidation regularization method is used to update the weight parameters to generate an optimized two-stream neural network.

[0039] Based on the optimized two-stream neural network, the initial navigation path deviation is corrected by the sliding mode control law to generate the final navigation path.

[0040] The beneficial effects of the present invention are as follows: The potential field modeling engine is used to generate a three-dimensional dynamic potential field model, and an accurate three-dimensional potential field gradient tensor is generated based on this, ensuring that the anchor cable tension can be dynamically adjusted according to the actual environment changes, thereby improving the stability and safety during the operation. Secondly, through efficient preprocessing steps such as timestamp alignment, noise filtering, and coordinate system unification, various types of environmental data (such as three-dimensional point cloud data, underwater obstacle sonar data, and cutter suction dredger operation data) are seamlessly integrated to generate a high-quality multi-modal environmental data packet. These high-precision data provide a solid foundation for subsequent path planning, making the generated navigation path more accurate and reliable. Brief Description of the Drawings

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 It is a schematic diagram of the multi-anchor cable tension adaptive regulation system for the height-limited section of the cutter suction dredger.

[0043] Figure 2 It is a flowchart for generating the initial navigation path based on the three-dimensional potential field gradient tensor.

[0044] Figure 3 It is a flowchart of the working principle of the path planning module.

[0045] Figure 4 It is a schematic diagram of the architecture of the multi-anchor cable tension adaptive regulation system. Detailed Embodiments

[0046] To make the above objects, features, and advantages of the present invention more obvious and understandable, the detailed embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0047] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0048] Second, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or selectively mutually exclusive embodiment with other embodiments.

[0049] Referring to Figures 1 to 4 , this embodiment provides a multi-anchor cable tension adaptive regulation system for the height-limited section of a cutter suction dredger, including the following steps:

[0050] An acquisition module that acquires ship environmental situation data, performs timestamp alignment, noise filtering, and coordinate system unification, and generates a multi-modal environmental data packet.

[0051] Specifically, the following operations are included.

[0052] First, acquire the ship environmental situation data. The ship environmental situation data specifically includes three-dimensional point cloud data of the height-limited section, sonar data of underwater obstacles, and cutter suction dredger operation data;

[0053] Install a lidar scanning device on the cutter suction dredger to scan the surrounding environment in real time. By emitting high-frequency laser beams and receiving the reflected echo signals, acquire the three-dimensional point cloud data of the height-limited section;

[0054] Install a multi-beam sonar unit at the bottom of the cutter suction dredger. By simultaneously emitting multiple sound beams underwater and receiving the echo signals returned from underwater objects, acquire the sonar data of underwater obstacles;

[0055] The cutter suction dredger operation data includes data on engine speed, fuel consumption, wheel speed, ship speed, and position. Among them, the engine speed and fuel consumption are acquired through an on-board diagnostic unit (OBD), the wheel speed and ship speed are acquired through wheel speed sensors, and the ship position is acquired through a GPS device;

[0056] After the acquisition is completed, use the PTP protocol to perform timestamp alignment on the ship environmental situation data. In specific operations, deploy a PTP hardware clock synchronization unit in the multi-sensor unit, and adopt a master-slave time reference architecture. Based on the PTP protocol, use message interaction and round-trip delay measurement to synchronize the multi-sensor unit. After the synchronization is completed, use linear interpolation algorithm and cubic spline interpolation algorithm to achieve cross-frequency domain timestamp alignment. At the same time, integrate a clock offset compensation mechanism and an abnormal clock jump detection algorithm to perform jump detection and dynamic time window sliding verification to ensure the timestamp synchronization of the ship environmental situation data;

[0057] Next, perform noise filtering on the ship environmental situation data. In specific operations, for the three-dimensional point cloud data in the height-limited section, use the rectangular information granulation algorithm to divide the three-dimensional point cloud data into multiple small rectangular regions, calculate the statistical characteristics of the points in each region, and based on the statistical characteristics, identify outliers through the Grubbs criterion and use information granule similarity analysis to eliminate them, thereby improving the data quality;

[0058] For the sonar data of underwater obstacles, use the wavelet threshold denoising method. Analyze the signal characteristics of the sonar data at different scales through wavelet decomposition, and set the denoising threshold through an adaptive threshold adjustment mechanism. Remove the high-frequency noise exceeding the denoising threshold through the soft threshold function to retain useful information;

[0059] For the operation data of the cutter suction dredger, use the DBSCAN clustering algorithm. Generate the main data clusters through density reachability analysis, effectively identify the outliers far from the main data clusters through outlier quantification analysis, and remove them through the dynamic sliding window filtering mechanism to ensure the accuracy and reliability of the data;

[0060] After noise filtering, unify the preprocessed ship environmental situation data to the geographic coordinate system for subsequent fusion and analysis. In specific operations, based on the ICP algorithm, use the FPFH feature descriptor to construct the local feature space of the point cloud, and use the Tukey weight function to remove the wrong corresponding point sets caused by occlusion or noise in the local feature space of the point cloud to find the best corresponding relationship between the two corresponding point sets. According to the best corresponding relationship, adjust the position of one corresponding point set through SVD decomposition to make it coincide with the other corresponding point set. After adjustment, unify the ship environmental situation data to the WGS-84 geographic coordinate system through seven-parameter coordinate transformation;

[0061] Next, fuse the processed ship environmental situation data to generate a multi-modal environment data packet. In specific operations, use a hierarchical fusion strategy for integration: construct the spatial index of the octree for the three-dimensional point cloud data in the height-limited section, and obtain the compressed point cloud data through the aggregation of in-voxel features of the point cloud; use the wavelet packet reconstruction method to enhance the signal of the sonar data of underwater obstacles, and extract the echo delay and intensity features through time-domain peak detection and energy integration calculation; apply the Kalman filter to the ship operation data for state estimation to obtain the optimal estimated value of the ship state; encapsulate the compressed point cloud data, echo delay and intensity features, and the optimal estimated value of the ship state through the JSON-LD format, and use the Zstd compression algorithm to generate a standardized multi-modal environment data packet, and perform verification through SHA-256 hashing to ensure data integrity.

[0062] The model construction module constructs a three-dimensional dynamic potential field model based on the multimodal environment data packet using the potential field modeling engine, generates a potential field distribution matrix through the finite difference method, and simultaneously calculates the partial derivatives of the potential field distribution matrix using central differences to generate a three-dimensional potential field gradient tensor.

[0063] Specifically, the following operations are included:

[0064] Based on the multimodal environment data packet, a three-dimensional voxel map is constructed using the point cloud voxelization downsampling method, and a spatial index is established through KD-Tree. In the specific operation, first, high-density point cloud data is extracted from the multimodal environment data packet through a JSON-LD parser, and the point cloud voxelization downsampling method is used. Through voxel grid division and centroid method feature retention, the high-density point cloud data is converted into a uniformly distributed voxel grid, thereby reducing the data volume and improving the efficiency of subsequent processing.

[0065] Based on the uniformly distributed voxel grid, through octree recursive segmentation, null voxel pruning and multi-resolution LOD generation are performed to construct a three-dimensional voxel map, and based on KD-Tree, through median axis selection segmentation and BBF search optimization, a spatial index is established for the three-dimensional voxel map to quickly query and access data in specific regions. KD-Tree can quickly locate target points in large-scale data sets, greatly improving the speed of data retrieval.

[0066] Next, based on the three-dimensional voxel map, a three-dimensional dynamic potential field model is constructed using the dynamic weight allocation strategy combined with the hyperbolic secant function. In the specific operation, first, multi-scale spatial convolution is performed on the three-dimensional voxel map through the Euclidean distance transformation algorithm to obtain the obstacle distance; secondly, the three-dimensional voxel map is subjected to hybrid clustering through the Bayesian probability inference algorithm to obtain the channel probability distribution; finally, the three-dimensional voxel map is subjected to optical flow field dynamic tracking through the feature extraction algorithm to extract the dynamic target clustering features.

[0067] The hyperbolic secant function is used to quantify the obstacle distance through the adaptive attenuation coefficient to generate the obstacle repulsive potential field; the gradient field mapping is used to normalize the channel probability distribution through the probability density function to generate the channel potential field; the time sliding window method is used to exponentially weight the dynamic target clustering features to obtain the dynamic target influence domain.

[0068] It should be noted that the adaptive attenuation coefficient is defined based on the obstacle distance and combined with the hyperbolic secant function, and its value range is usually [0.01, 1.5].

[0069] Based on the obstacle repulsive potential field, the channel potential field, and the influence domain of dynamic targets, an initial environmental feature set is constructed by splicing feature tensors, and the inverse weighted calculation is performed using the entropy weight method to obtain the static weights of the initial environmental feature set. At the same time, the dynamic weights of the initial environmental feature set are obtained using the BPTT algorithm through temporal gradient backpropagation;

[0070] Using the improved Log-Sum-Exp function, the static weights and dynamic weights are fused in multiple potential fields to generate a continuously differentiable three-dimensional dynamic potential field, and based on the Lyapunov exponent, the stability of the three-dimensional dynamic potential field is verified by calculating the divergence rate of the phase space trajectory;

[0071] Based on the three-dimensional dynamic potential field, a real-time maintenance mechanism based on ROS nodes is established, and the incremental KD-Tree is used to quickly respond to environmental changes. At the same time, OpenGL visualization rendering is used for verification to form a three-dimensional dynamic potential field model with both distance sensitivity and environmental adaptability. The method of combining the dynamic weight allocation strategy and the hyperbolic secant function not only improves the adaptability of the three-dimensional dynamic potential field model but also enhances the stability of its performance in complex environments;

[0072] Based on the constructed three-dimensional dynamic potential field model, a potential field distribution matrix is generated by the finite difference method. In specific operations, based on the finite difference method, grid points are established by the equidistant voxel division method, including the central point and 26 neighborhood points, and based on the central difference format, the three-direction synchronous gradient is solved for each grid point to obtain the potential field gradient;

[0073] Tikhonov regularization is used to handle the boundary singularities in the potential field gradient, and the Jacobi iterative method is used to separately solve the Poisson equation for the potential field gradient to generate the gravitational gradient. At the same time, through Gaussian kernel density estimation, the potential field gradient is convolved and filtered to obtain the repulsive field strength;

[0074] The gravitational gradient and the repulsive field strength are fused by the Log-Sum-Exp weighted fusion algorithm for multiple physical fields to generate the potential field distribution matrix;

[0075] Based on the potential field distribution matrix, a three-dimensional potential field gradient tensor is generated. In specific operations, central differences in the finite difference method are used for three-direction synchronous calculations, and after 2 calculations, weighted averaging is performed to generate the initial difference result. Based on the initial difference result, error correction is performed through Richardson extrapolation compensation to generate the first-order partial derivative of the potential field distribution matrix;

[0076] Using the quaternion spherical linear interpolation processing method, the direction continuity of the first-order partial derivatives is corrected to generate direction gradient components, and through Gaussian-Seidel smoothing filtering, noise suppression and outlier removal are performed on the direction gradient components. At the same time, the direction gradient components are superimposed through a vector synthesis algorithm to obtain an initial gradient vector field, and the initial gradient vector field describes the potential field change trend at each position in the environment;

[0077] To further improve the quality of the initial gradient vector field, the ADMM coupling reconstruction method is used to reconstruct it. In specific operations, through the ADMM coupling reconstruction method, noise in the initial gradient vector field is removed by alternating direction multiplier iteration, and the irregular parts in the initial gradient vector field are removed based on the hybrid regularization of the Huber function to make it smoother and more accurate, so as to generate the reconstructed gradient vector field;

[0078] Based on the reconstructed gradient vector field, the central difference method is continued to cross-calculate the second-order partial derivatives. In specific operations, based on the central difference method, the elements of the Hessian matrix are calculated for the reconstructed gradient vector field to generate the original second-order differential result. Based on the original second-order differential result, Richardson extrapolation compensation is used to eliminate the truncation error to generate the second-order partial derivatives. The use of the central difference method not only improves the accuracy of gradient calculation but also enhances the ability to capture complex terrain features;

[0079] Next, the hyperbolic secant function is used to dynamically adjust the weight distribution ratio of the second-order partial derivatives. In specific operations, the hyperbolic secant function is used to re-weight the original second-order differential result to generate an adaptive second-order differential field, and based on the adaptive second-order differential field, principal component analysis is used to project the feature space to re-allocate the ratio of the weights of the second-order partial derivatives to achieve dynamic weight adjustment based on terrain features;

[0080] Finally, regularization processing is performed through the alternating direction multiplier method to generate the final three-dimensional potential field gradient tensor. In specific operations, the enhanced ADMM algorithm is used to perform 50 iterations, and the residual noise is eliminated through the mixed norm constraint. At the same time, the Schur complement decomposition is used to accelerate the inverse operation of the second-order partial derivatives. The final three-dimensional potential field gradient tensor is generated, and the three-dimensional potential field gradient tensor is processed through the elastic net regularization method to prevent overfitting and improve the generalization ability, ensuring that the generated three-dimensional potential field gradient tensor has high reliability and stability in practical applications.

[0081] The path planning module, based on the three-dimensional potential field gradient tensor, uses the RRT algorithm to iteratively generate candidate path nodes and performs smoothing processing through the cubic spline interpolation algorithm to generate the initial navigation path.

[0082] Specifically, it includes the following steps,

[0083] The RRT algorithm is used to expand nodes of the three-dimensional potential field gradient tensor to obtain a dynamic direction vector. In specific operations, first, the RRT algorithm is used to randomly select target points in the three-dimensional potential field gradient tensor through Gaussian distribution sampling, and the target points are gradually expanded using an adaptive step size control to generate an initial path tree. Based on the initial path tree, a dynamic direction vector is obtained through the vector field integral method;

[0084] Next, the dynamic direction vector is converted into a path expansion direction through the elliptical sampling domain conversion method. In specific operations, based on the dynamic direction vector, equidistant spherical projection is performed through unit sphere mapping to construct an initial unit circle sampling domain, and the distribution characteristics of the dynamic direction vector are extracted through principal component analysis (PCA). According to the distribution characteristics, the axial variability is calculated through eigenvalue decomposition to obtain the ratio of the major and minor axes of the ellipse. The specific mathematical formula is as follows:

[0085] ;

[0086] Where, represents the ratio of the major and minor axes of the ellipse, represents the maximum distribution eigenvalue extracted through principal component analysis, represents the second-largest distribution eigenvalue extracted through principal component analysis;

[0087] It should be noted that the axial variability refers to the distribution dispersion degree of the dynamic direction vector in three-dimensional space, and its core role is to guide the construction of the shape of the elliptical sampling domain by quantifying the spatial distribution characteristics of the dynamic direction vector;

[0088] According to the ratio of the major and minor axes of the ellipse, the unit circle sampling domain is converted into an elliptical sampling domain through the affine transformation method. Then, quasi-Monte Carlo sampling is used to perform stratified importance sampling within the elliptical sampling domain to generate a candidate direction set. According to the candidate direction set, third-order control point optimization is performed through Bézier curve interpolation to generate a continuous path segment. Based on the continuous path segment, a path expansion direction is generated through Frenet frame projection. Among them, the application of the elliptical sampling domain conversion method makes the path expansion direction more reasonable and flexible;

[0089] At the same time, an adaptive step size strategy is used to obtain the node expansion step size. In specific operations, based on the dynamic direction vector, the potential field intensity is normalized through the hyperbolic tangent function to generate a reference step size. Based on the reference step size, dynamic attenuation is performed through a local curvature feedback mechanism to generate the node expansion step size, and a PID controller is used to adjust the increment error in real time to smooth the node expansion step size, ensuring that the expansion process satisfies both dynamic constraints and maintains the exploration efficiency. Among them, the application of the adaptive step size strategy ensures that the path expansion is both efficient and safe;

[0090] Then, according to the path expansion direction and the node expansion step size, candidate path nodes are generated through a hierarchical search strategy. In specific operations, the hierarchical search strategy performs dynamic hierarchical division through a spatial multi-resolution decomposition algorithm to construct a three-level search architecture, including a coarse-grained search layer, a medium-grained optimization layer, and a fine-grained verification layer;

[0091] Among them, the coarse-grained search layer screens the feasibility of the path expansion direction through an octree space division algorithm, and performs binary marking through a potential field intensity threshold segmentation method to obtain high-risk areas of obstacle collisions. At the same time, the AABB collision detection algorithm is used to eliminate high-risk areas of obstacle collisions to generate a set of primary feasible regions; based on the set of primary feasible regions, the Dijkstra algorithm is used to search for the minimum potential field integral path to generate an initial candidate node set;

[0092] The medium-grained optimization layer performs probability re-selection based on the initial candidate node set through the Metropolis-Hastings sampling method guided by the potential field gradient to generate an optimized node distribution; based on the optimized node distribution, B-spline interpolation is used for local smoothing optimization to obtain a curvature-constrained path segment;

[0093] The fine-grained verification layer verifies the node reachability based on the curvature-constrained path segment through multi-resolution voxel penetration testing using a parallelized ray detection method, and evaluates the node dynamic feasibility through a time-domain trajectory simulation method in combination with the dynamic window method. Finally, a set of candidate path nodes is output. Among them, the hierarchical search strategy can quickly exclude obviously infeasible path options at the macroscopic scale and avoid potential obstacles and optimize the smoothness of the path at the microscopic scale;

[0094] After generating the candidate path nodes, they are processed through a cubic spline interpolation algorithm to generate a cubic spline smooth trajectory. In specific operations, first, based on the set of candidate path nodes, through least squares fitting, a parametric mapping relationship between candidate path nodes is constructed using the regularized least squares method, and the Thomas calculation is used through the tridiagonal matrix solution method to obtain the spline coefficient matrix; then, through a piecewise continuity constraint algorithm, the C² continuity boundary condition is used to make the first and second partial derivatives continuous to optimize the spline coefficient matrix;

[0095] Based on the optimized spline coefficient matrix, the L-BFGS optimizer is used to adjust the positions of the candidate path nodes; after the adjustment is completed, the Fritsch-Carlson monotonicity-preserving algorithm is used to generate a cubic spline smooth trajectory through piecewise cubic Hermite interpolation; at the same time, the adaptive node encryption method is used to ensure the accuracy of the cubic spline smooth trajectory through RANSAC outlier rejection;

[0096] Based on the cubic spline smooth trajectory, dynamic regularization optimization is carried out through the ADMM algorithm to generate the initial navigation path. In specific operations, first, the cubic spline trajectory is discretized into a dense path point set by equal arc length sampling, and an optimization objective function is constructed by the multi-objective weighted fusion method. Based on the optimization objective function, iterative solution is carried out by the alternating direction multiplier method:

[0097] First, the projection gradient method is used to accelerate the curvature calculation through Chebyshev polynomial approximation to generate the curvature optimization path segment;

[0098] Secondly, based on the curvature optimization path segment, the proximal gradient method is used to handle the non-smooth term, and combined with the Nesterov acceleration method to improve the convergence speed of the potential field adaptor to obtain the potential field compatible path;

[0099] Finally, based on the potential field compatible path, quadratic programming is used to solve the dynamic constraints to obtain the dynamic feasible solution;

[0100] In each iteration of the alternating direction multiplier method, the condition number of the dynamic feasible solution is optimized by the Schur complement decomposition method to generate the coordinated global solution; based on the coordinated global solution, the initial navigation path is generated by the B-spline reparameterization method. The application of the ADMM algorithm can effectively solve large-scale optimization problems and ensure that the generated initial navigation path has high feasibility and safety.

[0101] The instruction generation module inputs the initial navigation path and the real-time operation data of the trailing suction hopper dredger into the two-stream neural network, extracts the spatial topological features by using the spatial stream branch, extracts the dynamic correlation features by using the temporal stream branch, fuses the spatial topological features and the dynamic correlation features and inputs them into the policy network to output the adjustment instruction set of the anchor cable tension.

[0102] Specifically, it includes the following steps:

[0103] First, the construction and training of the two-stream neural network are carried out. In specific operations, a heterogeneous two-branch structure is constructed: the spatial stream branch uses ResNet-50 as the backbone network, is initialized by loading the ImageNet pre-trained weights, and is specially used to process the static potential field features; the temporal stream branch adopts a 3D ConvNet structure, and the 10-frame optical flow sequence generated by the TV-L1 algorithm is used as the input to capture the dynamic potential field features;

[0104] The training process adopts a three-stage alternating training strategy: In the first stage, the spatial stream branch is trained. The cross-entropy loss function is used for multi-class potential field feature classification, and the Adam optimizer is used for fine-tuning on the static potential field dataset with multiple samples. During this period, the statistics of the batch normalization layer are frozen to maintain the stability of the feature distribution. Here, the statistics of the batch normalization layer refer to the mean and variance dynamically calculated by the moving average method based on the input data of each mini-batch during the training process;

[0105] In the second stage, the temporal stream branch is trained. The triplet loss is used for temporal feature embedding learning, and the SGD optimizer with momentum is used to train on the dynamic potential field features of multiple samples. At the same time, 3-level temporal pyramid pooling is used to enhance the temporal feature expression ability;

[0106] In the third stage, the two-stream features are concatenated through a learnable adaptive fusion layer, and the end-to-end fine-tuning is carried out using a hierarchical learning rate strategy to output the fused feature classification result. At the same time, an early stopping mechanism is set to prevent overfitting of the two-stream neural network;

[0107] After the two-stream neural network training is completed, the initial navigation path is input into the two-stream neural network, and the spatial topological features are extracted using the spatial stream branch. In specific operations, the initial navigation path is transformed into a three-dimensional terrain tensor through the multi-source data fusion method, and the three-dimensional terrain tensor is input into the spatial stream branch of the two-stream neural network through three-dimensional convolution;

[0108] The spatial stream branch includes three levels of feature extraction hierarchies, a cross-scale feature fusion layer, and a channel attention enhancement layer. Among them, the first-level feature extraction hierarchy uses a three-dimensional convolutional layer for spatial context-aware convolution to generate a primary feature map; based on the primary feature map, multi-scale spatial downsampling is performed through a max-pooling layer to capture spatial structure features; the spatio-temporal joint convolution performs convolution in the temporal dimension on the spatial structure features, enabling the spatial stream branch to better understand the spatial distribution of the path and its change trend over time; the spatial dimensionality reduction is achieved through 1×1×1 convolution, reducing the dimension of the spatial structure features and improving the computational efficiency while retaining key information;

[0109] The second-level feature extraction hierarchy uses a three-dimensional convolutional layer for local feature refinement convolution to generate an intermediate feature map. Based on the intermediate feature map, cross-layer feature enhancement is performed through a residual connection layer to obtain refined local features; the local iterative optimization method optimizes the refined local features multiple times through gradient accumulation to improve the quality of the features; and the cross-layer feature fusion adaptively weights and fuses the refined local features through a channel attention gating mechanism to generate a richer feature representation;

[0110] The third-level feature extraction layer uses a global pooling layer to compress and aggregate features, generating a global context descriptor; based on the global context descriptor, high-order semantic encoding is performed through a fully connected layer to obtain a compact feature representation; based on the compact feature representation, different scale weights are dynamically assigned to the spatial structure features and refined local features through a channel attention mechanism. Specifically, in the operation, first, global average pooling is used for the spatial structure features to generate global channel statistics; 3×3 depthwise separable convolution is used for the refined local features to generate local channel statistics; based on the global channel statistics and local channel statistics, different scale weights are assigned to the spatial structure features and refined local features through a 2-layer fully connected layer;

[0111] Based on the different scale weights, through channel splicing and feature recalibration techniques, the spatial structure features and refined local features are fused to generate spatial topology features. Among them, the channel attention mechanism can automatically adjust the weights of each channel according to the importance of the features, thereby enhancing important features and suppressing irrelevant features;

[0112] Meanwhile, the real-time operation data of the trailing suction hopper dredger is input into the time stream branch to extract dynamic correlation features. Specifically, in the operation, based on the real-time operation data of the trailing suction hopper dredger, a time-frequency synchronization coding method is used to construct a time series tensor;

[0113] Next, based on the time series tensor, a bidirectional long short-term memory network is constructed. Specifically, in the operation, using the time series tensor, a forward TCN (temporal convolutional network) and a backward TCN are constructed through dilated causal convolution. Each TCN layer contains 32 LSTM units; each LSTM unit contains a gated mechanism with peephole connections inside; the forward layer captures long-range trends through multi-layer dilated convolution, and the backward layer lags the correlation pattern through reverse-time convolution;

[0114] In the training stage, a three-stage optimization strategy is adopted: in the pre-training stage, based on the time series tensor, a Wasserstein generative adversarial network is used to generate training samples;

[0115] In the main training stage, a temporal focal loss function is used to dynamically weight difficult samples for the training samples, and the training samples are iterated multiple times through an Adam optimizer. In each iteration, the gradient clipping method is applied to prevent gradient explosion;

[0116] In the fine-tuning stage, a fully connected layer is used to optimize the hybrid attention mechanism, and the trained bidirectional long short-term memory network is output.

[0117] Furthermore, the trained bidirectional long short-term memory network is used to extract dynamic correlation features: among them, the forward TCN includes a dilated convolutional layer and a gated activation layer. The dilated convolutional layer generates multi-scale temporal features through dilated convolution; based on the multi-scale temporal features, long-range trends are captured through residual connections to reduce the risk of gradient vanishing;

[0118] The backward TCN includes a causal convolutional layer and a reverse gating layer. The causal convolutional layer generates reverse temporal features through time-reversed convolution. Based on the reverse temporal features, a reverse degradation pattern is modeled by a degradation mode detector to identify potential ship performance decline. Through the combination of the above forward TCN and backward TCN, not only can the long-term trend in the data be identified, but also its potential reverse change pattern can be analyzed, ensuring that the provided time series features are more comprehensive and accurate.

[0119] Next, the long-term trend and reverse degradation pattern are input into the spatio-temporal fusion layer, and a time attention mechanism is used for feature fusion to generate dynamic correlation features. In specific operations, the spatio-temporal fusion layer performs time alignment and bidirectional splicing on the long-term trend features and reverse degradation pattern features, and inputs them into the time attention mechanism for processing. The time attention mechanism calculates the interaction relationship of the query, key, and value matrices through the multi-head attention algorithm to obtain temporal dependence features. After the temporal dependence features are subjected to residual connection through the dilated time convolutional layer, dynamic correlation features are output. The entire process is optimized using a pre-layer normalization structure and learnable fusion weights, which not only ensures the stability of training but also significantly improves the feature similarity. Among them, the application of the time attention mechanism highlights the data in important time periods, making the generated dynamic correlation features more representative.

[0120] After the extraction of the spatial topological features and dynamic correlation features, the two are weighted and fused and then input into the DDPG policy network to output an anchor cable tension adjustment instruction set. In specific operations, the spatial topological features and dynamic correlation features are fused through an adaptive feature gating fusion mechanism to generate a 1152-dimensional joint feature vector, which is input into the DDPG policy network through feature normalization.

[0121] Based on the DDPG policy network, the Actor-Critic collaborative optimization framework is called through the deep deterministic policy gradient algorithm. In the Actor part of the Actor-Critic collaborative optimization framework, Gaussian policy parameterization is performed through a three-layer fully connected network to generate an initial action distribution. Using the Tanh activation function, the initial action distribution is scaled through a dynamic action boundary mapping to generate an anchor cable tension adjustment strategy. The Critic part evaluates the quality of the tension adjustment strategy through a double four-layer fully connected network combined with the Q-value function. For example, the Critic part calculates a Q-value according to the action distribution generated by the Actor through the Bellman optimal equation. The Q-value reflects the expected return of taking this action in the current state. If the Q-value exceeds 0.8, it means that the tension adjustment strategy performs well in the current environment; otherwise, if the Q-value is lower than 0.5, it means that the tension adjustment strategy needs to be improved.

[0122] Based on the anchor cable tension adjustment strategy after Critic evaluation, the Ornstein-Uhlenbeck process method is used to perform action exploration and strategy optimization to generate the anchor cable tension adjustment parameter vector. The ADDPG strategy network combines the advantages of the value function method and the policy gradient method, and can efficiently learn the optimal strategy in the continuous action space.

[0123] Based on the anchor cable tension adjustment parameter vector, the anchor cable tension adjustment instruction set is generated through the finite element algorithm. In the specific operation, the nonlinear finite element solver (ANSYS APDL) is used to perform transient dynamic simulation on the anchor cable tension adjustment parameter vector to calculate the force distribution of each node; based on the force distribution, the anchor cable tension adjustment parameter vector is iteratively adjusted through the constrained optimization algorithm, and the stability of the anchor cable tension adjustment parameters is verified through Lyapunov exponent calculation; based on the iteratively adjusted anchor cable tension adjustment parameters, the anchor cable tension adjustment instruction is output through cubic spline interpolation and PID controller.

[0124] In the anchor cable tension control module, the potential field modeling engine parses the anchor cable tension adjustment instruction set, generates a new potential field distribution matrix and three-dimensional potential field gradient tensor, and inputs them into the two-stream neural network. The two-stream neural network performs parameter updates and path replanning to generate the final navigation path.

[0125] The specific steps include:

[0126] The potential field modeling engine parses the adjustment instruction set of the anchor cable tension to generate a new potential field distribution matrix and a three-dimensional potential field gradient tensor. In the specific operation, based on the adjustment instruction set of the anchor cable tension, the anchor cable tension is discretized into 50 Timoshenko beam units through a nonlinear finite element solver, and the finite element tension field coupling algorithm is used to perform bidirectional coupling of the structural field and the fluid field on the Timoshenko beam unit. At the same time, the Fixed-Stress splitting algorithm is used to ensure the coupling stability. After the coupling is completed, the displacement-tension coupling field data is extracted through the finite element algorithm, and the displacement-tension coupling field data is visualized and verified through the Paraview post-processing software.

[0127] Based on the displacement-tension coupling field data after verification, the Helmholtz equation solver based on the FEniCS framework is used to perform potential energy redistribution calculations: first, Gmsh is used to generate an unstructured tetrahedral mesh through the Delaunay triangulation algorithm; secondly, the GMRES iterative solver of the PETSc linear algebra library is used to solve the Helmholtz equation on the unstructured tetrahedral mesh to generate a node potential energy distribution field; then, based on the node potential energy distribution field, the potential energy gradient is automatically extracted through a Python script; finally, based on the potential energy gradient, the potential field distribution matrix is ​​generated through a bicubic spline interpolation algorithm;

[0128] Next, the implicit Euler method is adopted to generate a new three-dimensional potential field gradient tensor. In the specific operation, based on the adjustment instruction set of the anchor cable tension, the implicit Euler method is used to numerically solve the nonlinear potential field evolution equation: First, in the FEniCS framework, the discrete weak form equation is established through the variational form conversion method; based on the discrete weak form equation, the SNES nonlinear solver of PETSc is used to handle the nonlinear terms to obtain the numerical solution; based on the numerical solution, the ILU(3) preprocessing technology is used to accelerate the Krylov subspace iterative calculation, and the transient potential field distribution is output.

[0129] Based on the transient potential field distribution, a new three-dimensional potential field gradient tensor is calculated by the structured central difference method: First, a difference template is constructed through the grid step parameterization tool; then, the Scharr operator is used to perform isotropic gradient calculation on the difference template to obtain the preliminary gradient field data; secondly, based on the preliminary gradient field data, through CUDA parallel computing, grid point-level parallel processing is realized to generate a new three-dimensional potential field gradient tensor, which is stored in the HDF5 format for subsequent analysis.

[0130] After the new potential field distribution matrix and the three-dimensional potential field gradient tensor are input into the two-stream neural network, the elastic weight consolidation regularization method is used to update the weight parameters to generate an optimized two-stream neural network. In the specific operation, the new potential field distribution matrix and the three-dimensional potential field gradient tensor are input into the two-stream neural network through the data preprocessing pipeline.

[0131] The two-stream neural network uses the elastic weight consolidation (EWC) regularization method to update the weight parameters. The specific process is as follows: First, the potential field distribution matrix and the three-dimensional potential field gradient tensor are respectively input into the spatial stream branch and the temporal stream branch; through the feature fusion layer of the two-stream neural network, they are fused through the cross-modal attention mechanism to obtain the joint feature vector; then, based on the real-time ship environmental situation data, the loss is calculated through the cross-entropy loss function and the triplet loss function to generate the comprehensive loss value; then, based on the comprehensive loss value, the EWC regularization term method is used to constrain the weight update of the joint feature vector to construct a constrained optimization objective; based on the constrained optimization objective, the weight of the spatial stream branch is updated through hierarchical gradient clipping, and the weight of the temporal stream branch is updated through temporal gradient normalization.

[0132] Finally, the Adam optimizer is used for backpropagation update: the spatial flow branch adopts hierarchical weight solidification for differential parameter freezing; the temporal flow branch implements temporal-sensitive solidification for motion feature protection; the entire process of weight parameter update is implemented under the PyTorch framework, and automatic mixed precision (AMP) is used to accelerate training. The finally generated optimized two-stream neural network can not only improve the spatial perception ability in complex environments, but also enhance the capture and response efficiency of dynamic temporal features;

[0133] Based on the optimized two-stream neural network, the final navigation path is generated. In specific operations, first, the initial navigation path is input into the spatial flow branch to extract 128-dimensional spatial topological features. At the same time, the operation data of the cutter suction dredger is input into the temporal flow branch to generate 64-dimensional dynamic correlation features. Through the gated attention mechanism, the two types of features are fused into a 192-dimensional joint feature vector. Based on the joint feature vector, the improved RRT* algorithm is used to generate a candidate path set through dynamic constraint sampling. The MLP layer is used to verify the safety of the candidate path set through obstacle distance field analysis, and the Sigmoid output layer is used to perform non-linear probability weighting on the candidate path set through cubic polynomial regression to verify path safety. Finally, the MPC framework is used for path tracking optimization, and a B-spline curve path and a control instruction sequence are generated through an adaptive weight adjustment mechanism. Based on the B-spline curve path and the control instruction sequence, path tracking and adjustment are performed through a model control law to generate the final navigation path.

[0134] In summary, the present invention: the potential field modeling engine generates a three-dimensional dynamic potential field model and generates an accurate three-dimensional potential field gradient tensor based on this, ensuring that the anchor cable tension can be dynamically adjusted according to actual environmental changes, thereby improving the stability and safety during the operation. Secondly, through efficient preprocessing steps such as timestamp alignment, noise filtering, and coordinate system unification, various types of environmental data (such as three-dimensional point cloud data, underwater obstacle sonar data, and cutter suction dredger operation data) are seamlessly integrated to generate a high-quality multi-modal environmental data packet. These high-precision data provide a solid foundation for subsequent path planning, making the generated navigation path more accurate and reliable.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A multi-anchor cable tension adaptive control system for the height-limited section of a cutter suction dredger, characterized in that: including a data acquisition module that acquires ship environmental situation data, performs timestamp alignment, noise filtering, and coordinate system unification, and generates a multi-modal environmental data packet; a model construction module that constructs a three-dimensional dynamic potential field model based on the multi-modal environmental data packet using a potential field modeling engine, generates a potential field distribution matrix through the finite difference method, and simultaneously calculates the partial derivatives of the potential field distribution matrix using central difference to generate a three-dimensional potential field gradient tensor; a path planning module that iteratively generates candidate path nodes based on the three-dimensional potential field gradient tensor using the RRT algorithm and performs smoothing processing through the cubic spline interpolation algorithm to generate an initial navigation path; an instruction generation module that inputs the initial navigation path and real-time cutter suction dredger operation data into a two-stream neural network, extracts spatial topological features using the spatial stream branch, extracts dynamic correlation features using the temporal stream branch, fuses the spatial topological features and dynamic correlation features and inputs them into a policy network to output an adjustment instruction set for the anchor cable tension; an anchor cable tension control module that parses the adjustment instruction set for the anchor cable tension by the potential field modeling engine, generates a new potential field distribution matrix and a three-dimensional potential field gradient tensor, and inputs them into the two-stream neural network. The two-stream neural network performs parameter update and path replanning to generate a final navigation path.

2. The multi-anchor cable tension adaptive control system for the height-limited section of a cutter suction dredger according to claim 1, characterized in that: The ship environmental situation data includes three-dimensional point cloud data of height-limited sections, sonar data of underwater obstacles, and cutter suction dredger operation data.

3. The multi-anchor cable tension adaptive control system for the height-limited section of the cutter suction dredger according to claim 1, wherein: The generation of the multi-modal environmental data packet specifically includes the following steps performing timestamp alignment on the ship environmental situation data through the PTP protocol, performing outlier filtering on the three-dimensional point cloud data of the height-limited section through the rectangular information granulation algorithm, performing wavelet threshold denoising on the sonar data of underwater obstacles, and performing outlier removal on the cutter suction dredger operation data using the DBSCAN clustering algorithm; unifying the preprocessed ship environmental situation data to the geographic coordinate system using the ICP algorithm and integrating it using a hierarchical fusion strategy to generate a multi-modal environmental data packet.

4. The multi-anchor cable tension adaptive control system for the height-limited section of a trailing suction hopper dredger according to claim 1, wherein: The generation of the potential field distribution matrix through the finite difference method specifically includes the following steps based on the multi-modal environmental data packet, constructing a three-dimensional voxel map using the point cloud voxelization downsampling method, establishing a spatial index through the KD-Tree, and simultaneously constructing a three-dimensional dynamic potential field model through a dynamic weight allocation strategy combined with the hyperbolic secant function; based on the three-dimensional dynamic potential field model, performing spatial gradient field calculation through the finite difference method to obtain the gravitational gradient and the repulsive field strength, and simultaneously generating a potential field distribution matrix through a multi-physics field coupling and superposition algorithm based on the gravitational gradient and the repulsive field strength.

5. The multi-anchor cable tension adaptive control system for the height-limited section of a trailing suction hopper dredger according to claim 1, wherein: The generation of the three-dimensional potential field gradient tensor specifically includes the following steps calculating the first-order partial derivatives of the potential field distribution matrix using central difference in the finite difference method, and obtaining the initial gradient vector field based on the first-order partial derivatives through the vector synthesis algorithm; reconstructing the initial gradient vector field through the ADMM coupling reconstruction method, cross-calculating the second-order partial derivatives through the central difference method, dynamically adjusting the weight allocation ratio of the second-order partial derivatives using the hyperbolic secant function, and simultaneously performing regularization processing through the alternating direction multiplier method to generate a three-dimensional potential field gradient tensor.

6. The multi-anchor cable tension adaptive control system for the height-limited section of the cutter suction dredger according to claim 5, characterized in that: The generation of the initial navigation path specifically includes the following steps The RRT algorithm is used to expand nodes of the three-dimensional potential field gradient tensor, obtain a dynamic direction vector, and convert the dynamic direction vector into a path expansion direction through the elliptical sampling domain conversion method. Meanwhile, an adaptive step size strategy is adopted to obtain the node expansion step size; Candidate path nodes are generated according to the path expansion direction and the node expansion step size through a hierarchical search strategy; The candidate path nodes are processed by a cubic spline interpolation algorithm to generate a cubic spline smooth trajectory; Based on the cubic spline smooth trajectory, dynamic regularization optimization is carried out through the ADMM algorithm to generate an initial navigation path.

7. The self-adaptive tension regulation system for multiple anchor cables in the height-limited section of a trailing suction hopper dredger according to claim 6, characterized in that: The extraction of spatial topological features using the spatial flow branch is as follows. The initial navigation path is input into the spatial flow branch of the two-stream neural network. The first layer of the spatial flow branch performs spatio-temporal joint convolution and spatial dimensionality reduction to capture spatial structure features, and the second layer performs local iterative optimization and cross-layer feature fusion to obtain refined local features; The third layer dynamically assigns different scale weights to the spatial structure features and the refined local features through a channel attention mechanism, and fuses them through concatenated channels to generate spatial topological features.

8. The multi-anchor cable tension adaptive control system for the height-limited section of a trailing suction hopper dredger according to claim 7, wherein: The extraction of dynamic correlation features using the temporal flow branch is as follows. The real-time cutter suction dredger operation data is input into the temporal flow branch of the two-stream neural network. The forward TCN of the bidirectional long short-term memory network captures long-range trends through dilated convolution, and the backward TCN models the reverse degradation pattern through causal convolution; The long-range trend and the reverse degradation pattern are bidirectionally concatenated through a temporal attention mechanism to generate dynamic correlation features.

9. The multi-anchor cable tension adaptive control system for the height-limited section of a cutter suction dredger according to claim 1, characterized in that: The output of the adjustment instruction set for the anchor cable tension is as follows. The spatial topological features and the dynamic temporal features are weighted and fused and then input into the DDPG policy network. The Actor-Critic collaborative optimization framework is used for policy gradient update to generate an anchor cable tension adjustment parameter vector; Based on the anchor cable tension adjustment parameter vector, an adjustment instruction set for the anchor cable tension is generated through the finite element algorithm.

10. The height-limited section multi-anchor cable tension adaptive control system for a cutter suction dredger according to claim 1, characterized in that: The generation of the final navigation path is as follows. Based on the adjustment instruction set for the anchor cable tension, a new potential field distribution matrix is generated through the finite element tension field coupling algorithm, and the implicit Euler method is used for nonlinear calculation to generate a new three-dimensional potential field gradient tensor; The new potential field distribution matrix and the three-dimensional potential field gradient tensor are input into the two-stream neural network, and the elastic weight consolidation regularization method is used to update the weight parameters to generate an optimized two-stream neural network; Based on the optimized two-stream neural network, the initial navigation path deviation is corrected through a sliding mode control law to generate the final navigation path.

Images