A method for predicting deformation and load of marine cables based on CNN model

Abstract

The invention discloses a method for predicting deformation and load of marine cable based on CNN model. The prediction method measures and collects the vertical cross-sectional flow field of the inflow direction of the marine cable and the local displacement and local strain data of the cable through acoustic Doppler current profiler (ADCP), fiber grating sensor and displacement meter; establishes a computational fluid dynamics model through open source software OpenFOAM, takes the vertical cross-sectional flow field collected by the acoustic Doppler current profiler as input, and obtains the three-dimensional flow field information of the cable under different forms; further, the three-dimensional flow field information, the cable form and the measured local strain data are combined into a three-dimensional information matrix; establishes a convolutional neural network (CNN) model, takes the three-dimensional information matrix as input data, trains the convolutional neural network model, compares the measured local displacement and strain data of the cable, finds the optimal model, and finally realizes the prediction of cable deformation and local stress.

Description

Technical Field

[0001] The invention relates to a method for predicting deformation and load of marine cables based on a CNN model, in particular to predicting local deformation and stress of cables with different shapes under the action of ocean currents. Background Art

[0002] As global marine oil and gas resource exploitation develops towards the deep sea, offshore drilling and oil production floating platforms such as semi-submersible platforms and floating production storage and offloading units (FPSO) have been widely used. Marine pipes and cables are key components for deepwater oil and gas exploitation, and are also the weakest and most vulnerable link in the oil and gas exploitation process. Marine pipes and cables are connected to the operating platform through flexible joints, and the bottom end is connected to the seabed wellhead or underwater manifold. When the fluid flows through the pipes and cables, vortices that alternately fall off will appear on both sides of the pipes and cables, causing periodic changes in the riser lift. At the same time, due to the large movement of the upper platform under the action of waves and currents, the position of the bottoming point of the pipes and cables will continue to change, resulting in a drastic change in the bending curvature of the pipes and cables near the bottoming area, increased local stress, and fatigue damage that is very likely to occur, affecting the service life. Traditional methods for studying the deformation and local stress of marine pipes and cables include theoretical research, physical model experimental research, and numerical simulation research. Due to the complexity of the actual marine environment, theoretical research is usually based on certain assumptions, and the results obtained will be different from the actual project and difficult to apply; physical model test research usually scales the actual structure to the laboratory scale, which has higher reliability, but the refined test has extremely high requirements for equipment, and it is impossible to predict the deformation and stress of pipes and cables under different marine environments; in numerical simulation research, in order to ensure the accuracy of the calculation, a large amount of computing resources are usually required. At the same time, the deformation and strain of pipes and cables are affected by the marine environment, and it is necessary to combine computational fluid dynamics and finite element simulation software, and the implementation process is relatively complicated. Compared with traditional research methods, with the rapid development of sensors and machine learning, based on the actual measurable results, machine learning can more efficiently predict the deformation and strain of marine pipes and cables.

[0003] In order to obtain the deformation and load prediction model of marine cables, a series of training is required to the model first, so that for given inflow conditions and cable shapes, the model can accurately predict the required results. However, there are still great limitations in the prediction of traditional cable deformation and load: in order to obtain a prediction model that can be applied to actual projects, it is necessary to consider the coupling process of multiple conditions such as ocean currents, cable shapes, etc., while the traditional prediction model only considers the influence of one of them; since marine cables are flexible structures, the vortices generated when fluid flows through them can easily cause deformation of the structure, and the flow field information near the cable needs to be used as a characteristic condition for prediction; traditional prediction models are usually only based on the results of numerical simulation or physical model tests, ignoring the relevance and applicability to actual projects. Based on the above considerations, there is a need for a prediction method that can obtain flow field conditions and accurately predict the deformation and stress of cables. Summary of the invention

[0004] In view of the problems that the traditional models for predicting deformation and stress of pipes and cables do not consider the coupling effects of flow field conditions and pipe and cable morphology, and ignore the relevance and applicability to actual projects, the present invention relates to a method for predicting deformation and load of marine pipes and cables based on a CNN model. The method obtains actual engineering data through measurements such as acoustic Doppler profilers, fiber grating sensors, displacement meters, etc., reconstructs the flow field near the marine pipe and cable in three dimensions through OpenFOAM, obtains the three-dimensional flow field information near the pipe and cable, and uses the advantages of convolutional neural networks to establish a marine pipe and cable deformation and load prediction model that can be applied to actual projects.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is:

[0006] A method for predicting deformation and load of marine cable based on CNN model is proposed. In order to obtain in-situ data, the vertical cross-sectional flow field in the inflow direction of the cable is measured by an acoustic Doppler profiler, and the local displacement and strain of the cable are measured by fiber grating sensors and displacement meters arranged on the marine cable.

[0007] Based on the measured vertical cross-sectional flow field in the inflow direction of the cable, a computational fluid dynamics model was established using the open source software OpenFOAM to obtain the three-dimensional flow field information of the cable in different forms, which included velocity vector and pressure. The three-dimensional flow field information, cable form and the measured local strain data of the cable were combined into a three-dimensional information matrix as the initial training data for the neural network.

[0008] A convolutional neural network (CNN) deformation and load prediction model for pipes and cables is established. The initial training data is used as input to train the convolutional neural network model. The output of the model is the deformation and local stress of pipes and cables. By comparing the measured local displacement and strain data of pipes and cables, the optimal prediction model is obtained to realize the prediction of deformation and local stress of pipes and cables.

[0009] In the above technical solution, further, the acoustic Doppler profiler measures the flow field of five vertical sections upstream of the cable, wherein the middle section is located directly in front of the cable, and the intervals between the other sections are the cable diameter D.

[0010] Furthermore, the fiber grating sensors and displacement meters are arranged along the cable, starting from the junction of the vertical section and the curved section of the cable to the bottom section of the cable, and 100 fiber grating sensors and displacement meters are arranged at equal intervals.

[0011] Furthermore, the computational fluid dynamics model adopts the finite volume method, establishes the overall grid through BlockMesh in OpenFOAM, imports the model of the pipe and cable, and then divides the grid near the pipe and cable through SnappyHexMesh, and deletes the grid inside the pipe and cable.

[0012] Furthermore, the meshes near the pipes and cables are divided by SnappyHexMesh, and the specific implementation method is as follows: the meshes near the pipes and cables are cut by castellatedMesh, the meshes are fitted by snap, and finally the boundary layer meshes are added by addLayers to improve the calculation accuracy.

[0013] Furthermore, in the computational fluid dynamics model, the input is the vertical cross-sectional flow field obtained by actual measurement, and the output is the three-dimensional flow field information and the shape of the pipe and cable near the pipe and cable.

[0014] Furthermore, the three-dimensional information matrix has four layers, including flow velocity vector U, pressure P, cable morphology X and local strain ε, wherein the information matrix is ​​divided into a training set, a validation set and a test set, and the data set division ratio is 8:1:1.

[0015] Furthermore, the convolutional neural network cable deformation and load prediction model can set different network structures, numbers of convolutional layers, number of pooling layers, number of deconvolution layers, number of fully connected layers, number of convolutional kernels, number of neurons, initial learning rate and number of training times according to the complexity of the structure.

[0016] Furthermore, the deformation and local stress of the pipe and cable are predicted by a time program sequence. During the training process, the error analysis function MAE, average percentage error MAPE, root mean square error RMSE, correlation coefficient R 2Train the prediction model to obtain the optimal model. The calculation formula of the error analysis function is as follows:

[0017]

[0018]

[0019]

[0020]

[0021] Where n is the number of data in the test set, is the predicted value, y i is the actual value, is the average of the predicted values, is the average of the actual values.

[0022] Furthermore, after the marine cable deformation and load prediction model is trained, the deformation and load of the submarine cable can be predicted according to the marine environment near the marine cable and the shape of the submarine cable.

[0023] The present invention is beneficial in that:

[0024] The present invention can measure the vertical cross-sectional flow field information, the local displacement and strain of the pipe and cable by means of an acoustic Doppler profiler, a fiber grating sensor and a displacement meter arranged in the actual marine pipe and cable. The present invention uses the open source software OpenFOAM, takes the measured vertical cross-sectional flow field as an input condition, simulates and obtains the three-dimensional flow field information near the pipe and cable under different forms, especially the vortex shedding at the rear; further, the three-dimensional flow field information near the pipe and cable, the pipe and cable form and the local strain data are formed into a three-dimensional information matrix as the initial training data of the neural network; further establish a CNN prediction model, by changing the network structure, the number of convolution layers, the number of pooling layers, the number of deconvolution layers, the number of fully connected layers, the number of convolution kernels of the convolution layer, the number of neurons, the initial learning rate and the number of training times, the prediction results are compared with the measured values, and the optimal model suitable for prediction is found; finally, a marine pipe and cable deformation and load prediction neural network model suitable for actual engineering is established, and this prediction model can comprehensively consider the influence of various factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the forecasting method of the present invention;

[0026] Figure 2 is a schematic diagram of a three-dimensional reconstruction model of a pipe and cable according to the present invention;

[0027] Figure 3 It is a structural diagram of a convolutional neural network model of the present invention;

[0028] Figure 4It is a schematic diagram of the mean absolute error convergence of the present invention;

[0029] Figure 5 It is the instantaneous three-dimensional vorticity diagram behind the cable tube of the present invention;

[0030] Figure 6 It is a schematic diagram comparing the prediction results of the local stress of the cable of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific implementation examples, but the protection scope of the present invention is not limited to the implementation examples.

[0032] Figure 1 The flowchart of the prediction method of the present invention is shown in Figure 1. First, based on a given actual project, the vertical cross-sectional flow field of the inflow direction of the marine cable is measured by an acoustic Doppler profiler, and the local displacement and strain of the cable are measured by the fiber grating sensor and displacement meter arranged on the marine cable. Further, the given marine cable is reconstructed in three dimensions through the open source software OpenFOAM, and the measured vertical cross-sectional flow field is used as the input condition. The three-dimensional flow field information near the marine cable is obtained through numerical simulation, especially the vortex shedding behind the cable and the cable morphology. The three-dimensional flow field information, cable morphology and measured local displacement and strain data obtained from the computational fluid dynamics model are further combined into a three-dimensional information matrix as the initial training data of the neural network to establish a marine cable deformation and load prediction model. The three-dimensional information matrix is ​​divided into a training set, a validation set and a test set. By changing the network structure, the number of convolution layers, the number of pooling layers, the number of deconvolution layers, the number of fully connected layers, the number of convolution kernels, the number of neurons, the initial learning rate and the number of training times, the prediction results are compared with the measured values ​​to find the optimal model suitable for prediction. Finally, the deformation and local stress of the marine cable are predicted.

[0033] Figure 2 It is a schematic diagram of the three-dimensional reconstruction model of the pipe and cable of the present invention. According to the displacement of the pipe and cable at different water depths measured by the displacement meter, the pipe and cable model is established by the open source software OpenFOAM, and the grid is generated by BlockMesh and SnappyHexMesh. The finite volume method is used to establish the overall grid through BlockMesh in OpenFOAM, import the model of the pipe and cable, and then use SnappyHexMesh to divide the grid near the pipe and cable, and delete the internal grid of the pipe and cable, so as to establish a computational fluid dynamics model. The boundary conditions of the computational fluid dynamics model are: symmetrical boundary conditions are used on both sides and the top of the model, wall boundary conditions are used at the bottom, and free outflow boundary conditions are used at the outlet. The input of the model is the measured vertical section flow field, and the output is the three-dimensional flow field information near the pipe and cable and the shape of the pipe and cable.

[0034] Figure 3 This is a structural diagram of the convolutional neural network model of the present invention. The function of the convolution layer is to extract the characteristic parameters of the flow field and the displacement of the pipe and cable, and the number of extracted features is equal to the number of convolution kernels; the function of the pooling layer is to compress the information, and each time it passes through the pooling layer, the size of the three-dimensional information matrix is ​​reduced to half of the original size. This process can filter useful information and improve the accuracy of the prediction results; the function of the fully connected layer is to capture the characteristics of the input information matrix in detail, and establish a mapping with the target result after a series of nonlinear transformations; finally, in order to avoid overfitting, a random disconnection layer is added to the network at the end, which ensures the generalization ability of the model by randomly disconnecting a part of the neurons.

[0035] Figure 4 It is a schematic diagram of the mean absolute error convergence of the present invention. As can be seen from the figure, with the increase of the number of training times, the mean absolute error of the training set and the validation set gradually decreases and tends to be stable, and the final error is reduced to a smaller value.

[0036] Figure 5 This is the instantaneous three-dimensional vorticity diagram behind the cable of the present invention. As shown in the figure, a streamwise vortex appears in the upper right of the curved section of the cable, the wake shows obvious three-dimensional characteristics, and a shear layer is attached to the horizontal section of the cable, proving that the present invention accurately captures the flow field characteristics near the cable through the computational fluid dynamics model.

[0037] Figure 6 It is a schematic diagram for comparing the prediction results of the local stress of the pipe and cable of the present invention. As can be seen from the figure, the model proposed by the present invention can better predict the local stress of the pipe and cable.

[0038] Of course, the above are only specific application examples of the present invention. The present invention has other implementation modes. Any technical solutions formed by equivalent replacement or equivalent transformation fall within the protection scope required by the present invention.

Claims

1. A method for predicting deformation and load of marine cables based on CNN model, characterized by: Here are the steps: Obtaining in-situ field data: Using an acoustic Doppler profiler to measure the vertical cross-sectional flow field in the direction of the inflow of the duct and cable, and using fiber grating sensors and displacement meters arranged on the marine duct and cable to measure the local displacement and strain of the duct and cable; Acquisition of initial training data: Based on the measured vertical cross-sectional flow field of the cable inflow direction, the three-dimensional flow field information of the cable in different forms is obtained through the computational fluid dynamics model. The three-dimensional flow field information includes the flow velocity vector and the pressure. The three-dimensional flow field information, the cable form and the measured local strain data of the cable are combined into a three-dimensional information matrix as the initial training data of the neural network. A convolutional neural network deformation and load prediction model for cables and pipes is established: the initial training data is used as input to train the convolutional neural network model, and the model outputs the deformation and local stress of cables and pipes. Then, by comparing the measured local displacement and strain data of cables and pipes, the optimal prediction model is obtained to realize the prediction of deformation and local stress of cables and pipes.

2. The method for predicting deformation and load of marine cables based on CNN model according to claim 1, characterized in that: The acoustic Doppler profiler measures the flow field of five vertical sections upstream of the cable, wherein the middle section is located directly in front of the cable, and the intervals between the other sections are the cable diameter D.

3. The method for predicting deformation and load of marine cables based on CNN model according to claim 1, characterized in that: The fiber grating sensors and displacement meters are arranged along the pipe cable. Starting from the junction of the vertical section and the curved section of the pipe cable to the bottom section of the pipe cable, several fiber grating sensors and displacement meters are arranged at equal intervals.

4. The method for predicting deformation and load of marine cables based on CNN model according to claim 1, characterized in that: The computational fluid dynamics model adopts the finite volume method, and the overall grid is established by BlockMesh in OpenFOAM, the model of the pipe and cable is imported, and then the grid near the pipe and cable is divided by SnappyHexMesh, and the internal grid of the pipe and cable is deleted.

5. The method for predicting deformation and load of marine cables based on CNN model according to claim 4 is characterized in that: The method of dividing the mesh near the pipe and cable by SnappyHexMesh is as follows: cutting the mesh near the pipe and cable by castellatedMesh, fitting the mesh by snap, and finally adding the boundary layer mesh by addLayers.

6. The method for predicting deformation and load of marine cables based on CNN model according to claim 4 is characterized in that: In the computational fluid dynamics model, the input is the measured vertical cross-sectional flow field, and the output is the three-dimensional flow field information and the shape of the pipe and cable near the pipe and cable.

7. The method for predicting deformation and load of marine cables based on CNN model according to claim 1, characterized in that: The three-dimensional information matrix has four layers, including flow velocity vector U, pressure P, pipe and cable shape X and local strain ε.

8. The method for predicting deformation and load of marine cables based on CNN model according to claim 1, characterized in that: The deformation and local stress of the cable are predicted by a time program. During the training process, the error analysis function MAE, average percentage error MAPE, root mean square error RMSE, correlation coefficient R 2 Train the prediction model to obtain the optimal model. The calculation formula of the error analysis function is as follows: Where n is the number of data in the test set, is the predicted value, y i is the actual value, is the mean of the predicted values, and y is the mean of the actual values.

Images