An intelligent path planning system for submarine pipeline laying based on deep learning

Abstract

This application relates to an intelligent path planning system for submarine pipeline laying based on deep learning. The system includes: a sea area terrain feature extraction subsystem, which is used to obtain the terrain data of the target submarine area and generate a sea area terrain feature map of the target submarine area based on the terrain data; a global laying path planning subsystem, which is used to generate an initial pipeline network layout plan between a preset pipeline laying starting point and an ending point according to the sea area terrain feature map by using a graph search algorithm; a local laying path planning subsystem, which is used to calculate the laying complexity parameters of the connection paths in the initial pipeline network layout plan based on a deep learning model, and use a multi-objective optimization algorithm to iteratively adjust the node coordinate information and path topology information to generate an intelligent path for submarine pipeline laying. Using this system can achieve the full-process automation, multi-objective optimal solution and dynamic adaptation of submarine pipeline laying path planning.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and particularly relates to an intelligent path planning system for submarine pipeline laying based on deep learning. Background Art

[0002] In the field of submarine pipeline laying projects, pipeline laying path planning is a key link to ensure the smooth implementation of the project, improve project efficiency and safety. However, when laying pipelines in complex submarine environments, various technical challenges are often faced.

[0003] With the continuous deepening of ocean resource development, the demand for submarine pipeline laying is increasing day by day. However, the submarine environment is extremely complex, with diverse topographies and landforms, including various slope changes and obstacles. Accurately obtaining and analyzing submarine terrain information has become a major problem in path planning. In the past, it was difficult to obtain a feature map that could accurately reflect the actual situation of the submarine terrain based on the acquired terrain data, and the ability to identify submarine obstacles was limited, seriously affecting the accuracy and reliability of path planning.

[0004] In terms of path planning algorithms, traditional methods also have limitations. When dealing with complex submarine terrains, previous path planning algorithms are difficult to make full use of terrain feature information, and the generated initial pipeline network layout plan is often not reasonable enough to effectively avoid complex terrain areas, increasing construction difficulty and costs. Moreover, in traditional technologies, there is a lack of effective algorithms and models, and the data analysis ability is low, making it difficult to generate a pipeline laying path that meets the actual engineering requirements. Summary of the Invention

[0005] Based on this, it is necessary to provide an intelligent path planning system for submarine pipeline laying based on deep learning that can achieve automated, multi-objective, and dynamically adaptable path planning for submarine pipeline laying in view of the above technical problems.

[0006] The present application provides an intelligent path planning system for submarine pipeline laying based on deep learning, including:

[0007] A sea area terrain feature extraction subsystem, configured to obtain terrain data of a target submarine area and generate a sea area terrain feature map of the target submarine area based on the terrain data, where the terrain data includes slope features and obstacle distribution features;

[0008] A global laying path planning subsystem, configured to generate an initial pipeline network layout plan between a preset pipeline laying start point and an end point according to the sea area terrain feature map by using a graph search algorithm, where the initial pipeline network layout plan includes node coordinate information of connection nodes and path topology information of connection paths;

[0009] The local laying path planning subsystem is used to calculate the laying complexity parameters of the connection paths in the initial pipeline network layout plan based on a deep learning model, and use a multi-objective optimization algorithm to iteratively adjust the node coordinate information and path topology information to generate an intelligent path for submarine pipeline laying. The optimization objectives of the multi-objective optimization algorithm include the laying complexity parameters.

[0010] The above intelligent path planning system for submarine pipeline laying based on deep learning obtains terrain data including slope characteristics and obstacle distribution characteristics through the sea area terrain feature extraction subsystem, enabling the system to comprehensively and meticulously grasp the terrain conditions of the target submarine area, accurately present the terrain complexity and obstacle distribution, and provide accurate and crucial basic information for subsequent path planning.

[0011] Through the global laying path planning subsystem, using a graph search algorithm to plan the initial pipeline network layout plan can enable the submarine pipeline laying path planning process to fully analyze the submarine terrain features, effectively avoid terrain complex areas and obstacle areas, and improve the rationality and feasibility of the initial plan.

[0012] By using the local laying path planning subsystem to calculate the laying complexity parameters with the help of a deep learning model, various factors affecting the laying difficulty of submarine pipelines can be integrated, improving the practicality of path planning. And by using a multi-objective optimization algorithm to iteratively adjust the node coordinate information and path topology information, multiple preset objectives can be taken into account simultaneously, and then an intelligent path more in line with the actual engineering requirements can be generated, improving the overall quality and comprehensive benefits of the submarine pipeline laying path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0014] Figure 1 It is a schematic structural diagram of an intelligent path planning system for submarine pipeline laying based on deep learning provided by an embodiment of the present application.

[0015] Figure 2 It is a schematic structural diagram of a local laying path planning subsystem provided by an embodiment of the present application.

[0016] Figure 3 It is a schematic structural diagram of another local laying path planning subsystem provided by an embodiment of the present application.

[0017] Figure 4Schematic diagram of a sea area terrain feature extraction subsystem provided by an embodiment of the present application;

[0018] Figure 5 Schematic diagram of another sea area terrain feature extraction subsystem provided by an embodiment of the present application;

[0019] Figure 6 Schematic diagram of another intelligent path planning system for submarine pipeline laying based on deep learning provided by an embodiment of the present application. Detailed implementation manners

[0020] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0021] In an exemplary embodiment, as Figure 1 shown, an intelligent path planning system 100 for submarine pipeline laying based on deep learning is provided, including:

[0022] A sea area terrain feature extraction subsystem 101, which can be used to obtain the terrain data of the target sea area and can generate a sea area terrain feature map of the target sea area based on the terrain data. The terrain data can include slope features and obstacle distribution features.

[0023] Optionally, the sea area terrain feature extraction subsystem 101 can obtain the acoustic detection terrain data of the target sea area based on the acoustic detection device, can obtain the remote sensing detection terrain data of the target sea area based on the remote sensing detection device, and can perform data matching and data fusion after preprocessing the acoustic detection terrain data and the remote sensing detection terrain data to generate the terrain data of the target sea area. Among them, the acoustic detection device can include, but is not limited to, a single-beam echosounder, a multi-beam echosounder, a coherent multi-beam system, and a side-scan sonar device; the remote sensing detection device can include, but is not limited to, a ship remote sensing device, an aerial remote sensing device, and a space remote sensing device; the preprocessing can include, but is not limited to, radiometric correction, geometric correction, filtering and denoising, signal enhancement, image enhancement, and format conversion.

[0024] Optionally, the sea area terrain feature extraction subsystem 101 can generate a three-dimensional terrain model of the target sea area based on the terrain data including slope features and obstacle distribution features of the target sea area, and construct a sea area terrain feature map based on the three-dimensional terrain model of the target sea area.

[0025] Further, the sea area terrain feature extraction subsystem 101 can reconstruct the surface of the point cloud data through locally weighted least square fitting based on the moving least square method. The fitting function of the point cloud data within a local neighborhood can be expressed as:

[0026] ;

[0027] In the formula, is the fitting function, is the target point, is the order of the fitting function, are the undetermined coefficients, is the basis function. The sea area terrain feature extraction subsystem 101 can solve the undetermined coefficients by minimizing the locally weighted error function, and the minimization of the locally weighted error function can be expressed as:

[0028] ;

[0029] In the formula, is the minimization of the locally weighted error function, is the target point is the number of points in the neighborhood of the target point is the weight function, is the target point in the neighborhood of the th point, is the th point in the neighborhood of the target point, and

[0030] Optionally, the sea area terrain feature extraction subsystem 101 can project the slope feature and the obstacle distribution feature onto the two-dimensional terrain model of the target seabed area based on the terrain data including the slope feature and the obstacle distribution feature of the target seabed area, and can construct a sea area terrain feature map based on the two-dimensional terrain model of the target seabed area.

[0031] Exemplarily, the terrain data of the seabed area can include elevation data. The sea area terrain feature extraction subsystem 101 can calculate the slope based on the elevation data, and the slope calculation formula can be:

[0032] ;

[0033] In the formula, is the slope, and are the elevations in the direction and The gradient in the

[0034] Schematically, the format of the sea area terrain feature map matches the input format of the graph search algorithm.

[0035] The global laying path planning subsystem 102 can be used to generate an initial pipeline network layout plan between a preset pipeline laying start point and an end point according to the sea area terrain feature map by using a graph search algorithm. The initial pipeline network layout plan can include the node coordinate information of the connection nodes and the path topology information of the connection paths.

[0036] Schematically, the global laying path planning subsystem 102 can be connected to the sea area terrain feature extraction subsystem 101 to obtain the sea area terrain feature map of the target sea area generated by the sea area terrain feature extraction subsystem 101 and the terrain data including the slope feature and the obstacle distribution feature.

[0037] Optionally, the global laying path planning subsystem 102 can convert the sea area terrain feature map into the input format of the graph search algorithm; set the start point and the end point using the graph search algorithm to initialize the graph model; calculate the shortest path using the graph search algorithm; save the node coordinate information and the path topology information of the path to generate the initial pipeline network layout. Among them, the graph search algorithm can include but is not limited to the Dijkstra algorithm, the A* algorithm, the breadth-first search algorithm, and the depth-first search algorithm. The input format of the graph search algorithm can be an adjacency list, and the path topology information includes the connection relationship between nodes.

[0038] The local laying path planning subsystem 103 can be used to calculate the laying complexity parameter of the connection path in the initial pipeline network layout plan based on a deep learning model and use a multi-objective optimization algorithm to iteratively adjust the node coordinate information and the path topology information to generate an intelligent path for submarine pipeline laying. The optimization objectives of the multi-objective optimization algorithm include the laying complexity parameter.

[0039] Schematically, the local laying path planning subsystem 103 can be connected to the sea area terrain feature extraction subsystem 101 and the global laying path planning subsystem 102 to obtain the sea area terrain feature map of the target sea area generated by the sea area terrain feature extraction subsystem 101 and the terrain data including the slope feature and the obstacle distribution feature, and obtain the initial pipeline network layout plan including the node coordinate information of the connection nodes and the path topology information of the connection paths generated by the global laying path planning subsystem 102.

[0040] Further, the local laying path planning subsystem 103 can use the laying complexity parameter as the input of a multi-objective optimization algorithm to iteratively adjust the node coordinate information and path topology information, generate a multi-objective optimization iterative output path, and perform secondary optimization on the multi-objective optimization iterative output path to reduce the number of turns and the turning amplitude, thereby generating an intelligent path for submarine pipeline laying. Among them, the input of the multi-objective optimization algorithm can be one or more of the laying complexity parameter, pipeline length parameter, obstacle distance parameter, slope change parameter, cost target parameter, path smoothness parameter, and risk assessment parameter. The multi-objective optimization algorithm can include, but is not limited to, multi-objective particle swarm optimization, multi-objective genetic algorithm, multi-objective ant colony optimization algorithm, multi-objective non-dominated sorting genetic algorithm, and hybrid multi-objective optimization algorithm.

[0041] Optionally, the local laying path planning subsystem 103 can input the sea area terrain feature map, terrain data, and initial pipeline network layout scheme of the target sea area into a deep learning model to calculate the laying complexity parameter of the connection path in the initial pipeline network layout scheme. The deep learning model can be constructed based on one or more of the convolutional neural network model, transformer model, and long short-term memory network model.

[0042] Exemplarily, the input of the deep learning model can be the sea area terrain feature map and the initial pipeline network layout scheme. The deep learning model can include a convolutional neural network module. The fully connected layer of the convolutional neural network module can be built based on a BP neural network, and the output of the fully connected layer of the convolutional neural network module built based on the BP neural network can be used as the output of the deep learning model.

[0043] In the above intelligent path planning system for submarine pipeline laying based on deep learning, by obtaining terrain data containing slope features and obstacle distribution features and generating the corresponding sea area terrain feature map, the accuracy and feasibility of path planning can be improved; by using the graph search algorithm, possible connection paths can be quickly found under complex terrain conditions, which improves the planning efficiency and reduces the construction difficulty and cost; by leveraging the deep learning model and using the multi-objective optimization algorithm, various influencing factors can be comprehensively considered to achieve the intelligence and automation of path planning, reduce manual intervention, improve work efficiency, and thus enhance the quality of submarine pipeline laying path planning.

[0044] In an optional embodiment of the present application, please refer to Figure 2 , the local laying path planning subsystem 103 can include a deep learning feature calculation module 201, a laying complexity parameter calculation module 202, and a path iterative adjustment module 203.

[0045] The deep learning feature calculation module 201 can be used to analyze the initial pipeline network layout scheme based on a deep learning model according to the slope feature and the obstacle distribution feature, and generate the terrain distribution parameter and the laying cost parameter. The terrain distribution parameter can be used to characterize the terrain undulation degree of the connection path, and the laying cost parameter can be used to reflect the laying cost of the connection path.

[0046] Schematically, the deep learning feature calculation module 201 can be connected to the sea area terrain feature extraction subsystem and the global laying path planning subsystem, obtain the sea area terrain feature map of the target sea area generated by the sea area terrain feature extraction subsystem and the terrain data including the slope feature and the obstacle distribution feature, and obtain the initial pipeline network layout scheme including the node coordinate information of the connection nodes and the path topology information of the connection path generated by the global laying path planning subsystem. The deep learning feature calculation module 201 can also be connected to the path iterative adjustment module 203, and obtain the iterative path generated by the path iterative adjustment module 203 based on the multi-objective optimization algorithm for iterative adjustment of the node coordinate information and the path topology information.

[0047] Optionally, the deep learning model can be trained from a training data set constructed based on the pipeline network layout scheme of the sample of the historical submarine pipeline laying sea area, the terrain data including the slope feature and the obstacle distribution feature, the terrain distribution parameter and the laying cost parameter. The input of the deep learning model can be the initial pipeline network layout scheme and the terrain data including the slope feature and the obstacle distribution feature, and the output of the deep learning model can be the terrain distribution parameter and the laying cost parameter of the initial pipeline network layout scheme. Among them, the deep learning model can include a multi-layer perceptron module (MLP), and the output of the multi-layer perceptron module can be the output terrain distribution parameter and the laying cost parameter.

[0048] Furthermore, the input of the deep learning model can be the iterative path generated by the path iterative adjustment module 203 based on the multi-objective optimization algorithm for iterative adjustment of the node coordinate information and the path topology information and the terrain data including the slope feature and the obstacle distribution feature, and the output of the deep learning model can be the terrain distribution parameter and the laying cost parameter of the iterative path generated by the path iterative adjustment module 203 based on the multi-objective optimization algorithm for iterative adjustment of the node coordinate information and the path topology information.

[0049] Schematically, the initial pipeline network layout scheme and the terrain data can achieve position matching.

[0050] The laying complexity parameter calculation module 202 can be used to calculate the laying complexity parameter in combination with the fuzzy comprehensive laying difficulty evaluation model based on the terrain distribution feature and the obstacle influence factor. The laying complexity parameter can be used to characterize the influence of the pipe section length and the terrain complexity on the laying path.

[0051] Exemplarily, the laying complexity parameter calculation module 202 may be connected to the deep learning feature calculation module 201 to obtain the terrain distribution features generated by the deep learning feature calculation module 201; the laying complexity parameter calculation module 202 may be connected to the global laying path planning subsystem to obtain the initial pipeline network layout scheme generated by the global laying path planning subsystem; the laying complexity parameter calculation module 202 may be connected to the path iteration adjustment module 203 to obtain the iterative path generated by the path iteration adjustment module 203.

[0052] Optionally, the laying complexity parameter calculation module 202 may input the preset obstacle influence factor and the terrain distribution features generated by the deep learning feature calculation module 201 into the fuzzy comprehensive laying difficulty evaluation model to calculate the laying complexity parameters.

[0053] Optionally, the factor set of the fuzzy comprehensive laying difficulty evaluation model may include but is not limited to terrain distribution and obstacle density, and the comment set of the fuzzy comprehensive laying difficulty evaluation model may include but is not limited to low, relatively low, medium, relatively high, and high. The fuzzy relation matrix of the fuzzy comprehensive laying difficulty evaluation model may include the membership degree reflecting the th factor to the th level comment membership degree. The fuzzy comprehensive laying difficulty evaluation model may obtain the corresponding weight vector based on the factor set , and calculate the evaluation result vector based on the weight vector and the fuzzy relation matrix. Among them, the calculation formula of the evaluation result vector may be:

[0054] ;

[0055] In the formula, is the th item of the result vector, is the dimension of the factor set, is the th item of the weight vector. The fuzzy comprehensive laying difficulty evaluation model may obtain the score vector corresponding to each level based on the comment set . The calculation formula of the laying complexity parameter may be:

[0056] ;

[0057] In the formula, is the laying complexity parameter, is the number of levels of the comment set, ​For the th item of the score vector.

[0058] The path iterative adjustment module 203 is used to iteratively adjust the node coordinate information and the path topology information based on a multi-objective optimization algorithm to generate an intelligent path for submarine pipeline laying. The optimization objectives of the multi-objective optimization algorithm further include laying cost parameters.

[0059] Exemplarily, the path iterative adjustment module 203 can be connected to the laying complexity parameter calculation module 202 to obtain the laying complexity parameters generated by the laying complexity parameter calculation module 202; the path iterative adjustment module 203 can be connected to the deep learning feature calculation module 201 to obtain the laying cost parameters generated by the deep learning feature calculation module 201. The path iterative adjustment module 203 can use the laying cost parameters and the laying complexity parameters as the optimization objectives of the multi-objective optimization algorithm.

[0060] In the above intelligent path planning system for submarine pipeline laying based on deep learning, it is possible to deeply analyze the slope feature and the obstacle distribution feature based on the deep learning model. Compared with the traditional technical solution, it can more accurately grasp the complex relationship between the initial pipeline network layout plan and the terrain and obstacles, so as to generate terrain distribution parameters and laying cost parameters that are more in line with the actual terrain conditions, providing a more reliable data basis for subsequent path planning; by adopting a multi-objective optimization algorithm to iteratively adjust the node coordinate information and the path topology information, it is possible to take into account multiple relevant factors to generate an intelligent path for submarine pipeline laying with higher comprehensive benefits, effectively reducing the total cost of pipeline laying, improving resource utilization efficiency, and at the same time ensuring the quality and safety of pipeline laying; by adopting deep learning technology and fuzzy control technology, it is possible to make the formulation of the pipeline laying plan more intelligent and automated, reduce the errors of manual intervention and subjective judgment, and improve the work efficiency and decision-making scientificity of submarine pipeline laying path planning.

[0061] In an optional embodiment of the present application, please refer to Figure 3 , the local laying path planning subsystem 103 further includes a submarine terrain commonality migration module 310. The submarine terrain commonality migration module 310 includes a common feature extraction component 311, a feature label generation component 312, and a deep learning model adjustment component 313, where:

[0062] The common feature extraction component 311 can be used to extract the common features of the sea area terrain feature map according to the slope feature and the obstacle distribution feature. The common features are used to characterize the terrain commonality between the target submarine area and the historical submarine pipeline laying sea area.

[0063] Exemplarily, the common feature extraction component 311 can be connected to the sea area terrain feature extraction subsystem to obtain the terrain data of the target submarine area obtained by the sea area terrain feature extraction subsystem.

[0064] Optionally, the common feature extraction component 311 can convert the terrain data of the target seabed area and the historical seabed pipeline laying areas into a unified data format, and perform data cleaning and normalization processing on the terrain data including slope features and obstacle distribution features. The common feature extraction component 311 can extract common features from the slope features and obstacle distribution features of the processed terrain data of the target seabed area and the historical seabed pipeline laying areas to obtain common features. The common feature extraction can be one or more of correlation analysis, variance selection analysis, principal component analysis, and Z-value analysis.

[0065] The feature label generation component 312 can be used to input the common features into a clustering analysis model for classification to generate semantic feature labels. The semantic feature labels include continental shelf labels, continental slope labels, ocean basin labels, and trench labels.

[0066] Exemplarily, the feature label generation component 312 can be connected to the common feature extraction component 311 to obtain the common features extracted by the common feature extraction component 311. The clustering analysis model can be, but is not limited to, a K-Means clustering analysis model, a hierarchical clustering analysis model, a density-based spatial clustering of applications with noise (DBSCAN) model, and a Gaussian mixture probability-based clustering analysis model. The feature label generation component 312 can input the standardized common features into the selected clustering algorithm for classification, analyze the generated clustering results, and determine the semantic feature labels corresponding to each cluster.

[0067] Optionally, taking the K-Means clustering analysis model that can be adopted by the clustering analysis model as an example for illustration, the feature label generation component 312 can: in the initialization step, randomly initialize K clustering centers; in the assignment step, assign each sample to the cluster to which the nearest clustering center belongs; in the update step, update the clustering centers to the means of all samples in each cluster; repeat the assignment step and the update step until the clustering centers no longer change or reach the maximum number of iterations.

[0068] The deep learning model adjustment component 313 can be used to identify the target approximate historical seabed pipeline laying area in the historical seabed pipeline laying area based on the semantic feature labels, and update the parameters of the deep learning model by improving the feature alignment loss function in combination with the terrain data of the target seabed area and the terrain data of the target approximate historical seabed pipeline laying area.

[0069] Exemplarily, the deep learning model adjustment component 313 can be connected to the feature label generation component 312 and the deep learning feature calculation module 201, and adjust the deep learning model in the deep learning feature calculation module 201 based on the semantic feature labels generated by the feature label generation component 312.

[0070] Optionally, the expression of the improved feature alignment loss function can be:

[0071] ;

[0072] In the formula, is the improved feature alignment loss function, is the number of samples of the target approximate historical seabed pipeline laying area, is the number of samples of the target seabed area, is the feature mapping function, is the -th sample of the target approximate historical seabed pipeline laying area, is the -th sample of the target seabed area, is the square of the norm in the Reproducing Kernel Hilbert Space (RKHS).

[0073] In the above intelligent path planning system for seabed pipeline laying based on deep learning, the common features of terrain data can be extracted according to the slope features and obstacle distribution features, and the similarities between the current sea area terrain and the historical seabed pipeline laying areas can be effectively identified, providing a more accurate terrain information basis for subsequent path planning. By adopting corresponding laying strategies according to the characteristics of different terrain types in path planning, the pertinence and reliability of the laying path planning can be improved.

[0074] Furthermore, in the above intelligent path planning system for seabed pipeline laying based on deep learning, by improving the feature alignment loss function and combining the terrain data of the target seabed area and the terrain data of the target approximate historical seabed pipeline laying area to update the parameters of the deep learning model, the deep learning model can better adapt to the pipeline laying path planning tasks under different seabed terrain conditions, improving the generalization ability and accuracy of the deep learning model, and thus further enhancing the performance and quality of the path planning.

[0075] In an optional embodiment of the present application, the input of the improved feature alignment loss function is the feature mapping function of the target approximate historical seabed pipeline laying area and the feature mapping function of the target seabed area, and the feature mapping function is a function that maps the sea area terrain feature map to the terrain distribution feature and the laying cost feature.

[0076] Among them, the expression of the improved feature alignment loss function can be:

[0077] ;

[0078] In the formula, is the improved feature alignment loss function, is the coefficient of the maximum mean discrepancy term of the terrain distribution, is the number of samples in the sea area for approximating the historical seabed pipeline laying target, is the number of samples in the target seabed area, is the terrain distribution sub-function of the feature mapping function, is the th sample in the sea area for approximating the historical seabed pipeline laying target, is the th sample in the target seabed area, is the square of the norm in the reproducing kernel Hilbert space (RKHS), is the coefficient of the laying cost cross-constraint term, is the th gradient of the cost feature with respect to the terrain feature of the sample in the target seabed area, is the target approximate historical cost terrain gradient hyperparameter, is the expectation of the gradient of the cost feature with respect to the terrain feature of the sample in the sea area for approximating the historical seabed pipeline laying target.

[0079] In the above intelligent path planning system for seabed pipeline laying based on deep learning, by analyzing the sea area for approximating the historical seabed pipeline laying target and the samples in the target seabed area, the differences and commonalities in terrain features and cost features in different regions can be effectively captured, which helps to apply historical experience to the path planning of the current target area. It can maximize the economy and safety of seabed pipeline laying while ensuring the project quality.

[0080] In an optional embodiment of the present application, please refer to Figure 3 , the path iterative adjustment module 203 includes an improved sparrow search algorithm component 341.

[0081] The improved sparrow search algorithm component 341 can be used to adjust the node coordinate information and path topology information by using the improved sparrow search algorithm based on the laying complexity parameter and the laying cost parameter. The sparrow individuals of the improved sparrow search algorithm include the node coordinate information and the path topology information.

[0082] Among them, the position update formula of the discoverer in the improved sparrow search algorithm can be:

[0083] ;

[0084] In the formula, is the th iteration of the st discoverer's th node coordinate in the order from the starting point to the ending point, is the th best th node coordinate in the order from the starting point to the ending point in the previous For the th iteration, the th node coordinates of the discoverers in the order from the starting point to the ending point, where is the random weight coefficient, is a random number within is a random number within is the maximum number of iterations, is a random number subject to the normal distribution, is a matrix of the same type as the node coordinate matrix with all elements being 1.

[0085] In the above intelligent path planning system for submarine pipeline laying based on deep learning, by using an improved sparrow search algorithm to adjust the node coordinates and path topology, compared with traditional search algorithms, the global search ability and convergence speed can be improved.

[0086] In an optional embodiment of the present application, please refer to Figure 3 , the local laying path planning subsystem 103 may include a submarine terrain commonality migration module 310, a deep learning feature calculation module 201, a laying complexity parameter calculation module 202, a path iteration adjustment module 203, and a submarine pipeline digital twin display module 350. Through the above-mentioned modules, the local laying path planning subsystem 103 can significantly improve the intelligent level of submarine pipeline laying, optimize the laying path, reduce construction risks and costs, and improve the overall quality of the project.

[0087] In an optional embodiment of the present application, please refer to Figure 4 , the sea area terrain feature extraction subsystem 101 includes a three-dimensional terrain model generation module 410 and a sea area terrain feature map generation module 420.

[0088] The three-dimensional terrain model generation module 410 can be used to construct a three-dimensional terrain model with differential obstacle annotation based on terrain data. The three-dimensional terrain model with differential obstacle annotation is used to annotate obstacle blocks in differential blocks adapted to the terrain complexity.

[0089] Optionally, the three-dimensional terrain model generation module 410 can input the terrain data into a three-dimensional modeling model to construct an initial three-dimensional terrain model, and perform block differential simplification on the generated initial three-dimensional model based on the slope feature and obstacle distribution feature of the terrain data to generate a differential three-dimensional terrain model. The three-dimensional terrain model generation module 410 can perform obstacle annotation on the differential three-dimensional terrain model to obtain a three-dimensional terrain model with differential obstacle annotation. Among them, the three-dimensional modeling model can be, but is not limited to, a regular grid three-dimensional modeling model, an irregular triangular network three-dimensional modeling model, and a point cloud three-dimensional modeling model.

[0090] The sea area terrain feature map generation module 420 can be used to map the three-dimensional terrain model with differential obstacle markings to a two-dimensional plane, generate a two-dimensional grid map with obstacle markings, and construct a sea area terrain feature map based on the two-dimensional grid map with obstacle markings.

[0091] Exemplarily, the sea area terrain feature map generation module 420 can be connected to the three-dimensional terrain model generation module 410, obtain the three-dimensional terrain model with differential obstacle markings constructed by the three-dimensional terrain model generation module 410, map the three-dimensional terrain model with differential obstacle markings to a two-dimensional plane, generate a two-dimensional grid map with obstacle markings, and then construct a sea area terrain feature map based on the two-dimensional grid map with obstacle markings.

[0092] In the above intelligent path planning system for submarine pipeline laying based on deep learning, by constructing a three-dimensional terrain model with differential obstacle markings, it is possible to clearly mark the obstacle blocks in the differential blocks adapting to the terrain complexity, and visually reflect the three-dimensional structural characteristics of the seabed terrain and the distribution characteristics of the obstacles; by simplifying the three-dimensional space information into two-dimensional plane data, it is convenient to simplify the complexity of data processing and analysis, and improve the data storage, retrieval, and calculation performance of the system; the sea area terrain feature map constructed based on the two-dimensional grid map with obstacle markings can integrate terrain and obstacle information and improve the performance of subsequent graph search algorithms.

[0093] In an optional embodiment of the present application, please refer to Figure 4 , the three-dimensional terrain model generation module 410 includes a point cloud processing component 411, a block division component 412, and an obstacle marking component 413, where:

[0094] The point cloud processing component 411 can be used to construct a three-dimensional terrain model of the target seabed area based on terrain data and in combination with point cloud processing technology.

[0095] Optionally, the point cloud processing component 411 can collect terrain data of the target seabed area using professional marine exploration equipment such as a multibeam echosounder system and a sidescan sonar, and construct an original point cloud data set. The point cloud processing component 411 can perform filtering, noise reduction, and data registration on the original point cloud data set to obtain optimized point cloud data. The point cloud processing component 411 can construct a three-dimensional terrain model of the target seabed area based on the optimized point cloud data using a surface reconstruction algorithm.

[0096] The block division component 412 can be used to calculate a terrain complexity parameter matrix based on the three-dimensional terrain model in combination with slope characteristics and obstacle distribution characteristics, and use an adaptive grid division algorithm to divide the three-dimensional terrain model into grid blocks whose sizes are proportional to the terrain complexity parameters based on the terrain complexity parameter matrix, and construct a differential three-dimensional terrain model.

[0097] Optionally, the block division component 412 can be connected to the point cloud processing component 411 to obtain the three-dimensional terrain model generated by the point cloud processing component 411. By combining the slope characteristics and obstacle distribution characteristics of the terrain data of the target seabed area, the complexity parameter of the terrain data of the target seabed area can be calculated, and a terrain complexity parameter matrix can be formed. The block division component 412 can use an adaptive grid division algorithm to divide the three-dimensional terrain model into grid blocks whose sizes are proportional to the terrain complexity parameters, and construct a differential three-dimensional terrain model.

[0098] The obstacle annotation component 413 can be used to input the slope characteristics and obstacle distribution characteristics of the grid blocks of the differential three-dimensional terrain model into the three-dimensional obstacle recognition deep learning model to identify the obstacle grid blocks in the differential three-dimensional terrain model and construct a differential obstacle annotation three-dimensional terrain model.

[0099] Optionally, the three-dimensional obstacle recognition deep learning model can be a three-dimensional convolutional neural model. The three-dimensional obstacle recognition deep learning model can identify the obstacle grid blocks in the differential three-dimensional terrain model, and the obstacle annotation component 413 can perform obstacle annotation on the identified obstacle grid blocks to construct a differential obstacle annotation three-dimensional terrain model.

[0100] In the above intelligent path planning system for submarine pipeline laying based on deep learning, a large number of terrain data points can be accurately collected and processed through point cloud processing technology, so that the true situation of the seabed terrain can be highly restored; by using an adaptive grid division algorithm to divide the three-dimensional terrain model into grid blocks whose sizes are proportional to the terrain complexity parameters, unnecessary calculation amount can be reduced while ensuring the accuracy and reliability of the system, and the processing efficiency of the system can be improved; through the powerful learning and recognition ability of the three-dimensional obstacle recognition deep learning model, the obstacle grid blocks in the differential three-dimensional terrain model can be automatically and accurately identified, and the efficiency and accuracy of obstacle recognition can be improved.

[0101] In an optional embodiment of the present application, please refer to Figure 5 , the sea area terrain feature map can be an obstacle annotation two-dimensional grid map.

[0102] The sea area terrain feature extraction subsystem 101 can include a grid map initialization module 501, a risk assessment module 502, and a two-dimensional obstacle annotation module 503, where:

[0103] The grid map initialization module 501 can be used to map the terrain data to a two-dimensional plane to generate a two-dimensional initial grid map, and the grids of the two-dimensional initial grid map include slope feature attribute information and obstacle distribution feature attribute information.

[0104] The risk assessment module 502 can be used to calculate the risk assessment value of the grid based on the slope feature attribute information and the obstacle distribution feature attribute information.

[0105] The two-dimensional obstacle annotation module 503 can be used to input the risk assessment value of the grid into the two-dimensional obstacle recognition deep learning model, identify the obstacle grids in the two-dimensional initial grid map, and construct an obstacle-annotated two-dimensional grid map.

[0106] Optionally, the two-dimensional obstacle recognition deep learning model can be a convolutional neural network model.

[0107] In the above intelligent path planning system for submarine pipeline laying based on deep learning, by mapping the terrain data to a two-dimensional plane to generate a two-dimensional initial grid map, the planar expression of complex three-dimensional terrain data can be realized, making the data easier to process and analyze; by converting the terrain data into a two-dimensional grid map with specific attribute information, a unified data format can be provided for subsequent processing, facilitating the standardized processing of subsequent modules and improving the compatibility and generality of the system; by using the powerful pattern recognition ability of the deep learning model to automatically and accurately identify the obstacle grids in the two-dimensional initial grid map, compared with traditional manual recognition or simple rule-based judgment methods, it can handle complex terrain features and risk relationships, greatly improving the accuracy and efficiency of obstacle recognition.

[0108] In an optional embodiment of the present application, the graph search algorithm is the ant colony algorithm.

[0109] Optionally, the sea area terrain feature map is a grid map. The global laying path planning subsystem can set the pheromone importance factor, heuristic information importance factor, pheromone evaporation rate, and pheromone update constant, and initialize the pheromone matrix of the sea area terrain feature map; the global laying path planning subsystem can use the preset pipeline laying start point and end point as the start node and target node of the sea area terrain feature map; the global laying path planning subsystem can control each ant to select the next node according to the calculated probability until the end node is reached; the global laying path planning subsystem can update the pheromone concentration according to the evaporation rate and the path length of the ant, and repeat the path selection and pheromone update until the termination condition is met to generate an initial pipeline network layout plan.

[0110] In an optional embodiment of the present application, please refer to Figure 3 , the local laying path planning subsystem may further include a submarine pipeline digital twin display module 350.

[0111] The digital twin display module 350 of the subsea pipeline can be connected to the path iterative adjustment module 203 and the sea area terrain feature extraction subsystem, and generate and display the digital twin model of the subsea pipeline based on the intelligent subsea pipeline laying path generated by the path iterative adjustment module 203 and the sea area terrain feature map generated by the sea area terrain feature extraction subsystem.

[0112] In an exemplary embodiment, as Figure 6 shown, an intelligent path planning system 100 for subsea pipeline laying based on deep learning is provided, including: a sea area terrain feature extraction subsystem 101, a global laying path planning subsystem 102, and a local laying path planning subsystem 103, where:

[0113] The sea area terrain feature extraction subsystem 101 may include a three-dimensional terrain model generation module 410 and a sea area terrain feature map generation module 420. The sea area terrain feature map generation module 420 can be connected to the three-dimensional terrain model generation module 410 and the global laying path planning subsystem 102.

[0114] The local laying path planning subsystem 103 may include a seabed terrain commonality migration module 310, a deep learning feature calculation module 201, a laying complexity parameter calculation module 202, a path iterative adjustment module 203, and a digital twin display module 350 of the subsea pipeline. The seabed terrain commonality migration module 310 can be connected to the sea area terrain feature extraction subsystem 101 and the deep learning feature calculation module 201; the deep learning feature calculation module 201 can be connected to the global laying path planning subsystem 102, the seabed terrain commonality migration module 310, and the laying complexity parameter calculation module 202; the digital twin display module 350 of the subsea pipeline can be connected to the path iterative adjustment module 203 and the sea area terrain feature extraction subsystem 101.

[0115] In the above intelligent path planning system for subsea pipeline laying based on deep learning, through the collaborative work of each subsystem and module, accurate processing of terrain data and scientific planning of the subsea pipeline laying path can be achieved, thereby improving the efficiency, safety, and economy of the subsea pipeline laying project.

[0116] The above embodiments only represent several implementation manners of the embodiments of the present application, and the description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the embodiments of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. A deep learning-based intelligent path planning system for submarine pipeline laying, characterized in that: include: A sea area terrain feature extraction subsystem is used to obtain terrain data of a target seabed area and generate a sea area terrain feature map of the target seabed area based on the terrain data, wherein the terrain data includes slope characteristics and obstacle distribution characteristics; A global laying path planning subsystem, for generating an initial pipeline network layout plan between a preset pipeline laying starting point and an end point according to the sea area terrain feature map by using a graph search algorithm, wherein the initial pipeline network layout plan includes node coordinate information of connection nodes and path topology information of connection paths; A local laying path planning subsystem, which is used to calculate the laying complexity parameters of the connection paths in the initial pipeline network layout solution based on a deep learning model, and iteratively adjust the node coordinate information and the path topology information using a multi-objective optimization algorithm to generate an intelligent path for laying submarine pipelines, wherein the optimization target of the multi-objective optimization algorithm includes the laying complexity parameters; The local paving path planning subsystem includes a seabed terrain commonality migration module and a deep learning feature calculation module, and the seabed terrain commonality migration module includes a commonality feature extraction component, a feature label generation component and a deep learning model adjustment component; The deep learning feature calculation module is used to analyze the initial pipeline network layout plan according to the slope characteristics and the obstacle distribution characteristics based on the deep learning model, and generate terrain distribution parameters and laying cost parameters, wherein the terrain distribution parameters are used to characterize the terrain undulation of the connection path, and the laying cost parameters are used to reflect the laying cost of the connection path; The common feature extraction component is used to extract the common features of the terrain data according to the slope features and the obstacle distribution features, wherein the common features are used to characterize the commonalities of the terrain between the target seabed area and the sea area where the historical submarine pipeline was laid; The feature label generation component is used to input the common features into the cluster analysis model, classify them, and generate semantic feature labels, wherein the semantic feature labels include continental shelf labels, continental slope labels, ocean basin labels, and trench labels; The deep learning model adjustment component is used to identify the target approximate historical submarine pipeline laying sea area in the historical submarine pipeline laying sea area based on the semantic feature label, and update the parameters of the deep learning model by improving the feature alignment loss function, combining the terrain data of the target seabed area and the terrain data of the target approximate historical submarine pipeline laying sea area.

2. The system according to claim 1, characterized in that: The input of the improved feature alignment loss function is the feature mapping function of the target approximate historical submarine pipeline laying sea area and the feature mapping function of the target seabed area, and the feature mapping function is a function mapping the terrain data to the terrain distribution characteristics and the laying cost characteristics; Among them, the expression of the improved feature alignment loss function is: ； In the formula, To improve the feature alignment loss function, is the coefficient of the maximum mean difference term of terrain distribution, The number of samples of the sea area where the target approximates the historical submarine pipeline laying, is the number of samples in the target seafloor area, is the terrain distribution subfunction of the feature mapping function, For the The target is similar to the sample of the historical submarine pipeline laying area. For the samples from the target seafloor area, is the square of the norm in the reproducing kernel Hilbert space (RKHS), is the cross-constraint coefficient of the laying cost, For the The cost characteristics of the samples in the target seafloor area are compared with the gradient of the terrain characteristics, is the target approximation history cost terrain gradient hyperparameter, The goal is to approximate the expected gradient of the cost characteristics of a sample of historical submarine pipeline laying areas with respect to the terrain characteristics.

3. The system according to claim 1, characterized in that: The local paving path planning subsystem includes a paving complexity parameter calculation module and a path iteration adjustment module; The laying complexity parameter calculation module is used to calculate the laying complexity parameter based on the terrain distribution characteristics and obstacle influence factors in combination with the fuzzy comprehensive laying difficulty evaluation model, and the laying complexity parameter is used to characterize the influence of the pipe section length and terrain complexity on the laying path; The path iteration adjustment module is used to iteratively adjust the node coordinate information and the path topology information based on the multi-objective optimization algorithm to generate the submarine pipeline laying intelligent path. The optimization target of the multi-objective optimization algorithm also includes the laying cost parameter.

4. The system according to claim 3, characterized in that: The path iteration adjustment module includes an improved sparrow search algorithm component, which is used to: adjust the node coordinate information and the path topology information using an improved sparrow search algorithm based on the paving complexity parameter and the paving cost parameter, and the sparrow individual of the improved sparrow search algorithm includes the node coordinate information and the path topology information; Among them, the finder position update formula of the improved sparrow search algorithm is: ； In the formula, For the The iteration Only the finder's order from the starting point to the end point The node coordinates of the nodes, For the front The best one in the iteration is the first one in the order from the start to the end. The node coordinates of the nodes, For the The iteration The discoverer's order from the starting point to the end point is The node coordinates, is the random weight coefficient, for The random number inside for The random number inside is the maximum number of iterations, is a random number that follows a normal distribution, is a matrix of the same type as the node coordinate matrix whose elements are all 1.

5. The system according to claim 1, characterized in that: The sea area terrain feature extraction subsystem includes a three-dimensional terrain model generation module and a sea area terrain feature map generation module: The three-dimensional terrain model generation module is used to construct a differentiated obstacle-annotated three-dimensional terrain model based on the terrain data, wherein the differentiated obstacle-annotated three-dimensional terrain model is used to annotate obstacle blocks in differentiated blocks whose sizes are adapted to the complexity of the terrain; The sea area terrain feature map generation module is used to map the differentiated obstacle-annotated three-dimensional terrain model to a two-dimensional plane, generate an obstacle-annotated two-dimensional grid map, and construct the sea area terrain feature map based on the obstacle-annotated two-dimensional grid map.

6. The system according to claim 5, characterized in that: The three-dimensional terrain model generation module includes a point cloud processing component, a block division component and an obstacle marking component: The point cloud processing component is used to construct a three-dimensional terrain model of the target seabed area based on the terrain data in combination with point cloud processing technology; The block division component is used to calculate a terrain complexity parameter matrix based on the three-dimensional terrain model in combination with the slope characteristics and the obstacle distribution characteristics, and to divide the three-dimensional terrain model into grid blocks whose sizes are proportional to the terrain complexity parameters based on the terrain complexity parameter matrix using an adaptive grid division algorithm to construct a differentiated three-dimensional terrain model; The obstacle annotation component is used to input the slope characteristics and the obstacle distribution characteristics of the grid blocks of the differentiated three-dimensional terrain model into the three-dimensional obstacle recognition deep learning model, identify the obstacle grid blocks in the differentiated three-dimensional terrain model, and construct a differentiated obstacle annotated three-dimensional terrain model.

7. The system according to claim 1, characterized in that: The sea area topographic feature map is a two-dimensional grid map with obstacle markings; The sea area terrain feature extraction subsystem includes a grid map initialization module, a risk assessment module and a two-dimensional obstacle marking module: A grid map initialization module, used for mapping the terrain data to a two-dimensional plane to generate a two-dimensional initial grid map, wherein the grids of the two-dimensional initial grid map include slope characteristic attribute information and obstacle distribution characteristic attribute information; The risk assessment module is used to calculate the risk assessment value of the grid based on the slope characteristic attribute information and the obstacle distribution characteristic attribute information; The two-dimensional obstacle annotation module is used to input the risk assessment value of the grid into the two-dimensional obstacle recognition deep learning model, identify the obstacle grid in the two-dimensional initial grid map, and construct the obstacle annotated two-dimensional grid map.

8. The system according to any one of claims 5 to 7, characterized in that: The graph search algorithm is an ant colony algorithm.

9. The system according to any one of claims 1 to 7, characterized in that: The local laying path planning subsystem also includes a submarine pipeline digital twin display module; The submarine pipeline digital twin display module is used to generate and display a submarine pipeline digital twin model based on the submarine pipeline laying intelligent path and the sea area terrain feature map.

Images