Method for monitoring local stress of marine flexible pipeline framework layer structure under radial extrusion in real time
By combining the Attention-LSTM neural network with the attention mechanism, the problem of real-time monitoring of local stress in the marine flexible riser skeleton layer was solved, achieving fast and accurate stress prediction, reducing computing costs, improving prediction accuracy, and ensuring structural safety.
Patent Information
- Application Number
- CN202510784703.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-12
- Publication Date
- 2025-09-16
AI Technical Summary
Existing methods are unable to accurately predict the local stress of the "S"-shaped interlocking wrapped skeleton layer structure of offshore flexible risers in real time, especially under radial compression. The traditional method simplifies the model, resulting in long calculation time and prone to failure.
The Attention-LSTM neural network combined with the attention mechanism is used to input the key physical characteristic parameters of the skeleton layer (such as ellipticity and inner diameter) and radial extrusion load-displacement data to achieve fast and accurate prediction of the local stress of the skeleton layer.
It achieves high-precision real-time monitoring of the local stress of the skeleton layer structure, reduces the calculation cost, improves the prediction accuracy, can timely detect the risk of structural damage, and provide reliable preventive measures.
Smart Images
Figure CN120654569A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of neural networks and attention mechanisms, and relates to a method for real-time monitoring of the local stress of a marine flexible pipeline skeleton layer structure under radial extrusion. Specifically, the present invention is a method for real-time monitoring of the local stress of a special-section skeleton layer structure under radial extrusion based on an Attention-LSTM neural network. Background Art
[0002] With the growing global demand for energy, particularly for resources like oil and natural gas, the development of offshore oil and gas resources has become a crucial means of addressing this demand. Offshore flexible risers are crucial transportation equipment for subsea oil and gas production. Compared to traditional steel risers, flexible risers utilize a non-bonded assembly of metal and polymer layers. This not only maintains excellent mechanical properties but also exhibits reduced bending stiffness, making them more adaptable to the dynamic loads and deformations experienced in complex deepwater environments. As flexible risers are deployed at increasing depths, they must withstand higher deepwater pressures. Their increased length also significantly increases their overall weight. This change necessitates greater compressive force applied by the tensioning system during installation to effectively balance the pipeline's deadweight and ensure safe and reliable installation. As a key piece of engineering equipment on installation vessels, the tensioner's core function is to apply radial loads during flexible riser installation to offset the axial tension caused by the pipeline's deadweight. However, excessive compressive force during installation can damage the flexible riser's internal structure. Therefore, real-time monitoring of stress changes in the flexible riser structure during installation is essential to prevent potential structural failure.
[0003] As shown in Figure 1, the skeleton layer structure within the flexible riser is an S-shaped, interlocking, high-angle spiral winding structure. Its primary function is to bear the radial extrusion loads during installation and the hydrostatic pressure in deepwater environments, ensuring the structural stability and integrity of the flexible riser during installation and service. Excessive installation extrusion loads cause the skeleton layer structure to undergo elliptical deformation, extrusion slip deformation of the cross section, and induce yielding of the structural material, as shown in Figure 2. This in turn reduces the radial load-bearing capacity of the flexible riser under service conditions, ultimately significantly increasing the risk of crush failure.
[0004] This study investigates the problem of predicting the mechanical properties of S-shaped interlocking wound carcass structures. Currently, various equivalent models have been established for theoretical analysis. However, existing methods simplify the spirally wound structure of the S-shaped interlocking cross-section into a homogeneous circular ring model, making it impossible to predict the local stresses in the carcass structure. Finite element models can represent the stress-strain state of complex structures, but due to the large number of mesh elements and the complex nonlinear contact problems involved, they are computationally intensive and prone to convergence failure during implicit analysis. This study innovatively proposes an intelligent prediction model that integrates an LSTM neural network with an attention mechanism. By inputting key physical parameters of the carcass (including ellipticity, inner diameter, and radial extrusion load-displacement data), this model achieves rapid and accurate prediction of the local stresses in the carcass under radial extrusion conditions. This model has significant application value in pipeline health monitoring, providing reliable technical support for timely identification of structural damage risks and the implementation of preventive measures. Summary of the Invention
[0005] In response to the problem that existing methods are unable to accurately predict the local stress of the skeleton layer of the "S"-shaped interlocking structure in real time, the purpose of the present invention is to use a method that integrates the attention mechanism with the LSTM (Long Short-Term Memory) neural network to input the structure's ovality, inner diameter, and load-displacement data under radial extrusion to quickly and accurately predict the local stress of the skeleton layer structure.
[0006] The above invention objectives are achieved through the following technical solutions:
[0007] A method for real-time monitoring of local stress in complex, irregular-section skeleton layers under radial extrusion based on an Attention-LSTM neural network is proposed. The method combines the Attention-LSTM neural network with an attention mechanism to address the uneven stress distribution caused by complex cross-sectional shapes. The method includes the following steps:
[0008] Step 1: Record the ovality and inner diameter parameters of the skeleton layer structure and collect the load-displacement data of the skeleton layer structure during the installation process. The present invention uses numerical software to simulate the skeleton layer structure installation process and extract the skeleton layer structure size information and load-displacement data.
[0009] Step 2: First, use the Attention mechanism to convert the key physical features of the skeleton layer (ellipticity, inner diameter) into vector representations through the embedding layer. Then perform a linear transformation on the input vector to generate the query (Q) matrix, key (K) matrix, and value (V) matrix.
[0010] Step 3: Calculate the weights of the ellipticity and inner diameter of the skeleton layer structure by taking the dot product of the query matrix and the key matrix. Then, use the Softmax function to normalize these weights. Use the normalized weights to perform weighted summation on the value matrix to obtain the attention vector that integrates the physical features of the skeleton layer structure (skeleton layer ellipticity and skeleton layer inner diameter).
[0011] Step 4: Expand the dimension of the vector output by the attention mechanism to match the dimension of the skeleton layer load-displacement data under radial compression. Concatenate the attention output with the skeleton layer load-displacement data under radial compression of the same data dimension and import it into the LSTM neural network model.
[0012] Step 5. At the output end of the LSTM neural network model, a fully connected layer using the ReLU activation function is connected to enhance the nonlinear modeling capability of the model. Through the fully connected layer of the linear activation function, the high-dimensional features output by the upper layer are mapped to the target dimension so that it matches the feature space of the local stress vector of the skeleton layer, and finally the local stress vector of the skeleton layer that meets the prediction requirements is output.
[0013] Furthermore, the LSTM (Long Short-Term Memory) neural network model in step 4 is specifically used to process sequence data in machine learning. Its main goal is to solve the gradient vanishing problem in data sets with long-term dependencies. The LSTM neural network consists of an "input gate," an "output gate," a "forget gate," and a "memory unit," as follows:
[0014] The input gate and memory unit mentioned above: The input gate is used to control how new input information enters the memory unit and determines the degree to which it is integrated with the existing information. Its formula is as follows:
[0015]
[0016] in, is the output of the input gate at the current time step; and Represent the memory state of the previous time step and the input of the current time step respectively. and Respectively and The weight of represents the bias term; Is an activation function that maps input to the range of 0 to 1. When the value of the input gate is close to 1, the LSTM neural network model is more inclined to absorb new information. When the value of the input gate is close to 0, the LSTM neural network model tends to retain old information. Represents a memory unit, which is mainly used to store new and old information, as shown in formula (2):
[0017]
[0018] in, and Indicates the memory unit and The weight of .
[0019] The forget gate determines the information that needs to be discarded in the memory cell. When the output value of the forget gate is 1, it retains the information; when the value is 0, it deletes the information. Its formula is expressed as follows:
[0020]
[0021] in, represents the forget gate; and Indicates the forgotten gate and The weight of represents the bias term.
[0022] The output gate The function of the output gate is to pass relevant information from the memory unit to the next memory state and output this information. When the output of the output gate is close to 0, the relevant information is ignored; when the output is close to 1, the corresponding information is output. Its formula is as follows:
[0023]
[0024] in, represents the output gate; and Indicates the output gate and The weight of represents the bias term.
[0025] Furthermore, the attention mechanism in step 1 selectively focuses on the key physical features of the skeleton layer structure (skeleton layer structure ellipticity and skeleton layer structure inner diameter) when processing sequence data. (Query), (key) and A mapping of (values) to outputs, where the weights are calculated by and application In this method, the attention mechanism is used to convert the ellipticity and inner diameter of the skeleton layer structure into vectors and input them into the LSTM network model.
[0026]
[0027] To address the inability of using either the Attention mechanism or LSTM neural networks alone to achieve the high-precision requirements for stress state prediction, this paper introduces an attention mechanism to map the ellipticity and inner diameter of the skeleton structure into a feature vector. This vector is then dimensionally concatenated with the load-displacement data of the skeleton layer under radial compression, and then fed into the LSTM network model. This fusion strategy improves the model's stress state prediction accuracy.
[0028] A system for real-time monitoring of local stress under radial extrusion of complex special-shaped cross-section skeleton layers based on an Attention-LSTM neural network is provided. The steps for implementing the above method are as follows:
[0029] 1) Data acquisition module, used to collect skeleton layer structural characteristic information and load-displacement data;
[0030] 2) Data preprocessing module, used for preprocessing input data;
[0031] 3) Building the Attention-LSTM Model
[0032] 4) Attention-LSTM model training input data for stress prediction;
[0033] 5) Output module, used to generate and display the stress results of the predicted skeleton layer structure.
[0034] Furthermore, the system is suitable for real-time monitoring of the mechanical properties of marine flexible pipelines during installation.
[0035] A computer program product comprising computer readable instructions which, when executed, implement the method described.
[0036] In summary, the present invention includes at least one of the following beneficial technical effects:
[0037] The present invention combines the attention mechanism with the LSTM neural network to achieve rapid and accurate prediction of the local stress of the skeleton layer under radial compression. By taking macroscopic key physical characteristics (such as ovality, inner diameter and load-displacement) as input, accurate prediction of the local stress of the complex cross-section of the skeleton layer is achieved. This model is very beneficial in monitoring the health of the pipeline and can take timely measures to prevent damage or failure. In terms of local structural response, the key physical characteristics of the skeleton layer structure are combined with the attention mechanism, and the LSTM network is used to accurately predict the local stress of the skeleton layer cross-section by inputting the load-displacement under radial compression. When the data-driven method predicts the local stress of the skeleton layer, compared with the traditional LSTM neural network model, the Attention-LSTM neural network that considers structural characteristics (ovality, inner diameter) is very important for improving the accuracy of the model. The prediction results of the data-driven model have an error of less than 5% compared with the finite element results, which effectively improves the accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a skeleton layer structure with an "S" shaped interlocking section for marine flexible pipelines;
[0039] Figure 2 The cross section of the skeleton layer undergoes extrusion and slip deformation during installation;
[0040] Figure 3 This is a flow chart for predicting local stress of skeleton layer structure based on LSTM-Attention network model;
[0041] Figure 4 Loss function curves of the two neural network models on the training set and test set; Figure 4 (a) is the loss function curve of the Attention-LSTM model; Figure 4 (b) is the loss function curve of the LSTM model;
[0042] Figure 5 The correlation and absolute error between the Attention-LSTM network and the finite element model in predicting the radial compression of the skeleton layer of size a are shown. The local stress prediction results of the two network models for skeleton layer structures of three sizes are compared. Figure 5 (a) Cross-sectional stress prediction of the skeleton layer structure (inner diameter 140 mm, ellipticity 1.4%) using the Attention-LSTM model. Figure 5 (b) Cross-sectional stress of the skeleton layer structure (inner diameter 140 mm, ellipticity 1.4%) predicted by the LSTM model; Figure 5 (c) Cross-sectional stress prediction of the skeleton layer structure (inner diameter 210 mm, ellipticity 1.9%) by the Attention-LSTM model; Figure 5 (d) Cross-sectional stress of the skeleton layer structure (inner diameter 210 mm, ellipticity 1.9%) predicted by the LSTM model; Figure 5 (e) Cross-sectional stress prediction of the skeleton layer structure (inner diameter 270 mm and ellipticity 2.6%) by the Attention-LSTM model; Figure 5 (f) Cross-sectional stress of the skeleton layer structure (inner diameter 270 mm, ellipticity 2.6%) predicted by the LSTM model;
[0043] Figure 6 This is the correlation between the Attention-LSTM network and the finite element model in predicting the radial compression of the skeleton layer of size b, that is, the comparison of the determination coefficients of the two models.
[0044] Figure 7 Comparison between the cross-sectional stress prediction of the skeleton layer structure by the Attention-LSTM model in the present invention and the results of the refined finite element model. Figure 7 (a) Attention-LSTM model predicts the cross-sectional stress of three inner diameter skeleton layer structures; Figure 7 (b) Attention-LSTM model predicts the cross-sectional stress of three types of ellipticity skeleton layer structures.
[0045] Figure 8 The results of the Attention-LSTM model in this invention predicting the hoop stress of the skeleton layer structure are compared with those of the refined finite element model. Figure 8 (a) Attention-LSTM model predicts the hoop stress of the skeleton layer structure at three inner diameters; Figure 8 (b) The Attention-LSTM model predicts the hoop stress of the skeleton layer structure under different radial deformations. DETAILED DESCRIPTION
[0046] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application; it is obvious that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0047] In the description of this application, it should be noted that the terms "upper," "lower," "inner," "outer," "top / bottom," and the like, indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate the description of this application and simplify the description. They do not indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0048] In the description of this application, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "provided with," "mounted / connected," and "connected" should be understood in a broad sense. For example, "connected" can mean a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in this application based on the specific circumstances.
[0049] Implementation model: The implementation model should be explained in steps first and then explained in diagrams.
[0050] A method for real-time monitoring of local stress of complex special-shaped cross-section skeleton layers under radial extrusion based on an Attention-LSTM neural network comprises the following steps:
[0051] Step 1: Record the ovality and inner diameter parameters of the skeleton layer structure and collect the load-displacement data of the skeleton layer structure during the installation process.
[0052] Step 2: Construct a feature encoder based on the attention mechanism, convert the collected ellipticity and inner diameter parameters into feature vectors through the embedding layer; perform linear transformation on the feature vectors to generate query matrix, key matrix and value matrix.
[0053] Step 3: Calculate the attention weight by taking the dot product of the query matrix and the key matrix, normalize the weight using the Softmax function, and perform weighted summation on the value matrix using the attention weight to obtain the attention output of the inner diameter and ellipticity of the skeleton layer structure.
[0054] Step 4: Expand the dimension of the key physical feature vector (ellipticity, inner diameter) of the skeleton layer after the output of the Attention mechanism to match the same dimension as the load-displacement data of the skeleton layer. Then, import it into the LSTM neural network model to carry out data training and output the local stress information of the skeleton layer.
[0055] Step 5: Based on the output training of the LSTM model of the first fully connected neural network layer, the nonlinear characteristics are improved. The second fully connected neural network layer changes the dimension of the output stress value of the LSTM neural network model to match the dimension of the local stress data of the skeleton layer structure, and finally outputs the local stress of the skeleton layer structure.
[0056] Reference Figure 3 The neural network model built based on the Attention-LSTM in this invention uses ovality, inner diameter, and load-displacement data under radial compression as input, and outputs the average von Mises stress of the skeleton layer cross section at 0° and 90°. First, the key physical characteristics of the skeleton layer structure (ovality and inner diameter) and the load-displacement data during installation are extracted. This dataset is then divided into a training set and a validation set at a ratio of 7:1. These sets are then imported into the constructed Attention-LSTM model to predict the local stress of the skeleton layer structure.
[0057] The inner diameter and ovality of the skeleton layer are key structural features that affect its radial compression performance. To improve the neural network model's ability to recognize these key features, this paper introduces an attention mechanism. By calculating the feature weights of the inner diameter and ovality, the model's focus on important parameters is enhanced. Because the load-displacement data of the skeleton layer during radial compression has a time-dependent nature, this solution employs a long short-term memory (LSTM) network for time-series modeling to effectively capture the dynamic mechanical behavior of the skeleton layer. The output of the LSTM network is processed by a two-layer fully connected neural network: the first layer extracts high-order nonlinear features, and the second layer performs dimensionality transformation, ultimately outputting information about the local stresses in the skeleton layer structure during installation.
[0058] In the Attention-LSTM model, the LSTM layer is configured with 64 units to extract temporal features. L2 regularization is introduced in the network weight calculation to effectively prevent overfitting during model training. The fully connected layer uses 128 neurons and employs the ReLU activation function for nonlinear transformations. The output layer flattens the data, with the number of neurons set to the product of the sample data size and the output dimension. The stochastic gradient descent (SGD) optimizer was used for testing in the fully connected layer, with the mean squared error (MSE) selected as the loss function.
[0059] Results and Discussion
[0060] Figure 4 The loss function curves for the Attention-LSTM model and the LSTM model on the training and test sets are shown. The figure shows that the loss functions of both models converged after 60 iterations, and the mean squared error (MSE) values for both models on the training and test sets were both below 0.01, validating the feasibility and effectiveness of these two network models.
[0061] This paper compared the prediction performance of the Attention-LSTM model and the LSTM model by comparing three sets of skeleton-layer structural parameters. As shown in Figures 5 and 6, the Attention-LSTM model constructed in this paper demonstrates significant advantages over the traditional LSTM model (which achieved a minimum goodness-of-fit of only 0.841): its predicted stress-displacement curves are more consistent with the refined finite element model results, with a minimum goodness-of-fit of 0.994, close to the ideal value of 1. This result not only validates the effectiveness of incorporating an attention mechanism into neural network modeling to focus on key structural features of the skeleton layer, but also demonstrates that expanding the data dimension through feature enhancement can significantly improve model prediction accuracy, providing a more precise solution for predicting the mechanical properties of skeleton-layer structures.
[0062] Figure 7 The results of the comparison between the Attention-LSTM model constructed by the present invention and the fine finite element model for the prediction of the end face stress of the skeleton layer with different structural parameters during radial compression are presented. The following findings can be drawn from the model analysis: (1) For the influence of the change in inner diameter, the Attention-LSTM model accurately captures the mechanical property that the radial stiffness decreases due to the increase in inner diameter; (2) For the influence of the change in ellipticity, the Attention-LSTM model accurately predicts the trend of the cross-sectional stress decreasing with the increase in ellipticity. The Attention-LSTM model proposed by the present invention can accurately characterize the cross-sectional stress variation law of the skeleton layer structure with different inner diameters and different ellipticities. Its prediction accuracy is comparable to that of the fine finite element model with high computational cost, which fully verifies the reliability and engineering applicability of the model under complex structural parameter conditions.
[0063] Figure 8 The stress variation trend of the skeleton layer structure in the circumferential direction predicted by the Attention-LSTM model is shown. The following findings can be drawn from the model analysis: (1) The model accurately captures the symmetrical distribution of hoop stress and successfully predicts the maximum stress concentration phenomenon at the 0° and 90° positions and the minimum stress characteristics in the 45° area; (2) The model accurately characterizes the characteristics of the small-sized skeleton layer reaching the yield stress earlier at the 0° and 90° positions, and accurately quantifies the size effect of the elastic region of the skeleton layer structure during installation. These prediction results are highly consistent with the calculation results of the fine finite element model, fully demonstrating the high-precision prediction ability of the Attention-LSTM model of the present invention for the hoop stress field of the skeleton layer complex structure, and provides a reliable technical means for judging the failure of the skeleton layer structure during installation.
[0064] Summarize:
[0065] The present invention proposes a stress monitoring method for the skeleton layer of marine flexible pipelines based on the fusion of attention mechanism and long short-term memory network. The method realizes high-precision stress prediction of complex special-shaped cross-section structures under radial extrusion conditions through deep learning technology. The innovation of the present invention is mainly reflected in the following aspects: (1) the attention mechanism is introduced into the field of pipeline skeleton layer stress prediction, and the key structural parameters such as ovality and inner diameter are focused by feature weight distribution, which significantly improves the prediction accuracy of the model; (2) a method is constructed to combine the key physical information of the skeleton layer structure (ovality and inner diameter) with the time series information (load-displacement data) during the installation process. Combining the time series modeling capability of the LSTM network and the feature selection capability of the attention mechanism, the variation law of the local stress of the skeleton layer structure during the installation process is monitored in real time with high precision. Compared with traditional theoretical models and equivalent methods, the model proposed in the present invention has higher prediction accuracy. By introducing the attention mechanism, the model can better focus on key physical features, thereby improving the prediction performance. In addition, this method reduces the dependence on complex finite element models and reduces the computational cost, which is conducive to the real-time monitoring of pipeline health status in practical applications.
[0066] The technical principle of this invention is to establish a nonlinear mapping relationship between structural parameters (ovality, inner diameter) and dynamic loads through an Attention-LSTM hybrid neural network. The attention module automatically identifies key features, the LSTM network extracts temporal dependencies, and dual fully connected layers implement feature space transformation, ultimately outputting highly accurate cross-sectional and hoop stress distributions. In particular, the model automatically learns material hardening effects through end-to-end training, avoiding errors caused by theoretical model simplification in traditional methods.
[0067] The above-described embodiments merely express the implementation methods of the present invention, but should not be understood as limiting the scope of the patent of the present invention. It should be pointed out that for those skilled in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.
Claims
1. A method for real-time monitoring of local stress of a marine flexible pipeline skeleton layer structure under radial extrusion, characterized in that: The method described combines the Attention-LSTM neural network with the attention mechanism, including the following steps: Step 1: Use numerical software to simulate the skeleton layer structure installation process and extract the skeleton layer structure size information and load-displacement data; Step 2: Use the Attention mechanism to convert the key physical features of the skeleton layer into vector representations through the embedding layer, and perform linear transformation on the input vector to generate the query matrix, key matrix, and value matrix. The key physical features are ellipticity and inner diameter. Step 3: Calculate the weights of the ellipticity and inner diameter of the skeleton layer structure by taking the dot product of the query matrix and the key matrix, normalize the weights, and use the normalized weights to perform weighted summation on the value matrix to obtain the attention vector that integrates the physical features of the skeleton layer structure. Step 4: Expand the dimension of the vector output by the attention mechanism to match the dimension of the skeleton layer load-displacement data under radial compression. Concatenate the attention output with the skeleton layer load-displacement data under radial compression of the same data dimension and import it into the LSTM neural network model. Step 5. At the output end of the LSTM neural network model, a fully connected layer using the ReLU activation function is connected to map the high-dimensional features output by the upper layer to the target dimension so that it matches the feature space of the local stress vector of the skeleton layer, and finally outputs the local stress vector of the skeleton layer that meets the prediction requirements.
2. The method for real-time monitoring of local stress of a marine flexible pipeline skeleton layer structure under radial extrusion according to claim 1 is characterized in that: The attention mechanism is to selectively focus on the key physical features of the skeleton layer structure when processing sequence data, that is, to focus on the ellipticity and inner diameter of the skeleton layer structure; the attention mechanism defines the mapping of query Q, key K, value V to output, where the weight is obtained by applying the query, key After activation function, we get: The attention mechanism is used to convert the ellipticity and inner diameter of the skeleton layer structure into vectors and input them into the LSTM neural network model; ， By introducing the attention mechanism, the ellipticity and inner diameter of the skeleton layer structure are mapped into a feature vector, and this vector is dimensionally parallelized with the skeleton layer load-displacement data under radial extrusion and then input into the LSTM network model.
3. The method for real-time monitoring of local stress of a marine flexible pipeline skeleton layer structure under radial extrusion according to claim 1 is characterized in that: In step 3, the Softmax function is used to normalize the weights.
4. A system for real-time monitoring of local stress of marine flexible pipeline skeleton layer structure under radial extrusion, characterized by: The method described in claim 1, 2 or 3 is implemented by a system, and the steps are as follows: 1) Data acquisition module, used to collect skeleton layer structural characteristic information and load-displacement data; 2) Data preprocessing module, used for preprocessing input data; 3) Build the Attention-LSTM model; 4) Attention-LSTM model training input data for stress prediction; 5) Output module, used to generate and display the stress results of the predicted skeleton layer structure.
5. The system for real-time monitoring of local stress of a marine flexible pipeline skeleton layer structure under radial extrusion according to claim 4 is characterized in that: The system is suitable for real-time monitoring of the mechanical properties of marine flexible pipelines during installation.
6. A computer program product, characterized in that Contains computer-readable instructions that, when executed, implement the method of claim 1, 2, or 3.