A method for predicting large buoy motion response and mooring tension based on deep learning

Abstract

The present invention provides a method for predicting the motion response of a large buoy and the tension of an anchoring system combined with deep learning. The method is based on the computational fluid dynamics open source code OpenFOAM, establishes a hydrodynamic mathematical model of a large buoy system, studies the motion response of a large buoy, the motion response of an auxiliary buoy and the tension characteristics of an anchoring system under different wind, wave and current effects; and uses measured data to compare and verify the mathematical model. At the same time, a deep learning framework LSTM model is established, and the motion response of the buoy and the auxiliary buoy and the tension data of the anchoring system calculated by the verified mathematical model are used to train the LSTM model and optimize it using measured values. The optimized LSTM model can be used to achieve accurate prediction of the motion response of large buoys, auxiliary buoys and the tension of the anchoring system. This method couples the mathematical model constructed based on the computational fluid dynamics open source code OpenFOAM with the LSTM model, which can greatly improve the accuracy and prediction efficiency of the prediction of the motion response of the buoy and the tension of the anchoring system.

Description

Technical Field

[0001] The present invention relates to the field of deep learning, and in particular to a method for predicting large buoy motion response and mooring tension combined with deep learning. Background Art

[0002] How to accurately detect the marine environment and rationally develop marine resources is a major problem facing the world. Large buoys are an important observation platform that can detect the marine environment in a long-term and stable manner. The movement of buoys in the anchored sea area and the stress of the anchor system are crucial to their stability and reliability. Therefore, it is necessary to study and analyze the motion response of large buoys and the anchor tension.

[0003] The working environment of the ocean buoy system is complex and harsh. It is also affected by marine environmental factors such as wind, waves, and currents. In extreme marine environments, the mooring system is prone to damage, breakage, and displacement. For a long time, researchers have often used physical experiments and numerical simulation methods to study the motion response of buoys and the tension characteristics of mooring systems. However, in physical experiments, it is difficult to simulate the motion response of buoys and the tension of mooring systems under the coupling of complex marine environments such as wind, waves, and currents, which requires a lot of materials, time, and manpower costs; in numerical simulations, the simulation of buoy motion response and mooring system tension is usually based on potential flow theory, without considering the influence of strong nonlinear wave and current loads. The prediction results often have certain deviations, and it is urgent to improve its calculation accuracy. Summary of the invention

[0004] The present invention relates to a method for predicting the motion response and mooring tension of a large buoy combined with deep learning. Based on the computational fluid dynamics open source code OpenFOAM, the method establishes a hydrodynamic mathematical model of a large buoy system, and studies the motion response characteristics of a large buoy, the motion response of an auxiliary buoy, and the tension characteristics of a mooring system under different wind, wave, and current conditions. The motion response of a buoy system and an auxiliary buoy and the tension of the mooring system are obtained by using a six-degree-of-freedom motion instrument and an underwater dynamometer, and the information is compared and verified with numerical results. At the same time, a deep learning framework LSTM model is established, and the motion response of a large buoy and an auxiliary buoy and the tension data of the mooring system calculated by using the optimized hydrodynamic mathematical model of a large buoy system are used to train the LSTM model, and the measured data are used to optimize the LSTM model. Based on the optimized LSTM model, the accurate prediction of the motion response of a large buoy and an auxiliary buoy and the tension of the mooring system can be achieved.

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

[0006] A method for predicting the motion response and mooring tension of large buoys combined with deep learning is proposed. For a given large buoy system, a hydrodynamic mathematical model of the large buoy system under the action of wind, waves and currents is established based on the computational fluid dynamics open source code OpenFOAM. The mooring system tension and the motion response of large buoys and auxiliary buoys under different wave heights and wave periods are obtained, and the hydrodynamic characteristics of buoys and auxiliary buoys and the multi-structure coupled motion response characteristics are studied.

[0007] The motion response of large buoys and auxiliary buoys and the tension information of the mooring system are obtained by using a six-degree-of-freedom motion instrument and an underwater dynamometer, and compared with the calculation results of the hydrodynamic mathematical model of the large buoy system to obtain the optimized hydrodynamic mathematical model of the large buoy system. The motion response of large buoys and auxiliary buoys and the tension data of the mooring system are calculated using the optimized hydrodynamic mathematical model of the large buoy system, and the obtained motion response data series of large buoys and auxiliary buoys and the tension data of the mooring system are normalized and divided into a training set and a validation set; an LSTM model is constructed, and the model is trained using the training set data to preliminarily obtain a prediction model; the model parameters are adjusted using the validation set data to minimize the error between the prediction results and the measured values, and the optimal LSTM model is determined.

[0008] For the determined LSTM optimal model, the wind field, wave height, wave period, and current velocity measured in the project are used as input data to predict the tension of the mooring system and the motion response of large buoys and auxiliary buoys.

[0009] In the above technical solution, further, based on the computational fluid dynamics open source code OpenFOAM, a hydrodynamic mathematical model of a large buoy system under the action of wind, waves and current is established, specifically:

[0010] Initial boundary conditions and input physical parameters, including wind field, wave height, wave period and ocean current velocity, were set in the OpenFOAM numerical calculation model.

[0011] Based on the OpenFOAM standard solver olaFlow, the anchor chain solving module MOODY is implanted in the motion solution calculation of the large buoy system to solve the hydrodynamic mathematical model of the large buoy system, and the motion response of the large buoy and auxiliary buoy and the tension of the mooring system are obtained, so as to study the hydrodynamic characteristics of the large buoy and auxiliary buoy and the multi-structure coupled motion response characteristics.

[0012] The tension analysis of the mooring system is carried out based on the dynamic equilibrium equation and is calculated using the discontinuous finite element numerical method.

[0013] Furthermore, the buoy motion response equation obtained based on the hydrodynamic mathematical model of the large buoy system is as follows:

[0014]

[0015] Among them, K(t-τ) is the system delay function matrix; t is time; τ is the delay time; F i (t) is the environmental load acting on the buoy structure; F m (t) is the tension of the mooring system; X is the motion response matrix of the buoy in six degrees of freedom, including surge, sway, heave, roll, pitch and pitch; They represent the second-order and first-order derivatives of X respectively; M represents the mass matrix of the buoy; μ represents the additional mass matrix of the buoy; and C is the damping matrix.

[0016] Furthermore, the motion response of the large buoy and auxiliary buoy in the test was measured by a six-degree-of-freedom motion instrument, and the tension of the mooring system was measured by an underwater dynamometer. The data obtained were used to verify the calculation results of the hydrodynamic mathematical model of the large buoy system.

[0017] Furthermore, before inputting into the LSTM model, the motion response of the large buoy and auxiliary buoy and the tension data of the mooring system need to be normalized. The normalization equation is as follows:

[0018] x norm =(xx min ) / (x max -x min )

[0019] Similarly, after the normalized data has been trained on the network, it must go through a denormalization process to generate the actual output data. The calculation equation for the actual output data is as follows:

[0020] x=x norm (x max -x min )+x min

[0021] Among them, x is the actual value of the training data; x norm is the normalized value; x max and x min are the maximum and minimum values ​​of the training data respectively.

[0022] Furthermore, the normalized data set is divided into a training set and a validation set in a ratio of 7:3. The training set and validation set data are used to train and adjust the model parameters so that the error between the predicted result and the actual value is minimized and the optimal LSTM model is determined.

[0023] Furthermore, the prediction model sets the LSTM model according to the complexity of the data: the time window value is 200, the number of LSTM layers is 3, the number of neurons in each layer is 256, and the optimizer is Adam.

[0024] Furthermore, during the training process, the prediction model is trained according to the error analysis function mean absolute error MAE, mean square error MSE, and maximum percentage of prediction error E max Train the neural network and update the parameters to obtain the optimal model. The calculation formula of the error analysis function is as follows:

[0025]

[0026] Where n is the number of data in the test set, is the predicted value, y i is the actual value.

[0027] Furthermore, based on actual projects, the measured wind field, wave height and wave period, and current velocity are used as input data, and the optimal model is used to predict the tension of the mooring system and the motion response of the buoy and auxiliary buoy.

[0028] The present invention is beneficial in that:

[0029] Based on the computational fluid dynamics open source code OpenFOAM, a hydrodynamic mathematical model of a large buoy system is established to study the motion response characteristics of large buoys, auxiliary buoy motion response and mooring system tension characteristics under different wind, wave and current conditions. The motion response of large buoys and auxiliary buoys and the tension of the mooring system are obtained using a six-degree-of-freedom motion instrument and an underwater dynamometer, and the information is compared and verified with the numerical results. At the same time, a deep learning framework LSTM model is established. The LSTM model is trained and optimized using the verified motion response of buoys and auxiliary buoys and the tension data of the mooring system, so as to achieve accurate prediction of the motion response of large buoys and auxiliary buoys and the tension of the mooring system. This method is based on the coupling of the open source platform OpenFOAM and the LSTM model, and completes the prediction of the motion response of large buoys and auxiliary buoys and the tension of the mooring system under wind, wave and current fields. Among them, optimizing the hydrodynamic mathematical model of the large buoy system through measured data can improve the accuracy of numerical calculations. The optimized hydrodynamic mathematical model of the large buoy system can also make up for the defect that it is difficult to simulate the movement of the buoy system under complex working conditions. At the same time, the LSTM prediction model of the present invention uses the optimized numerical model calculation data for training and uses measured values ​​for optimization, which can greatly improve its accuracy and efficiency in predicting the motion response of large buoys and auxiliary buoys and the tension of the mooring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of the prediction method of the present invention;

[0031] Figure 2 Schematic diagram of a given large buoy mooring system;

[0032] Figure 3 It is the LSTM neural network prediction principle diagram of the present invention;

[0033] Figure 4 This is a diagram of the LSTM model training process of the present invention;

[0034] Figure 5 It is a schematic diagram of the convergence of the mean square error MSE of the present invention;

[0035] Figure 6 It is a schematic diagram comparing the prediction results of the tension of the mooring system of the present invention changing with time. DETAILED DESCRIPTION

[0036] 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.

[0037] Figure 1 The figure is a flow chart of the prediction method of the present invention. First, based on the computational fluid dynamics open source code OpenFOAM, a hydrodynamic mathematical model of a large buoy system is established to study the motion response characteristics of a large buoy, the motion response of an auxiliary buoy and the tension characteristics of an anchoring system under different wind, wave and current conditions. The motion response of the buoy and the auxiliary buoy and the tension information of the anchoring system are obtained by using a six-degree-of-freedom motion instrument and an underwater dynamometer, and compared and verified with the calculation results of the hydrodynamic mathematical model of a large buoy system. At the same time, a deep learning framework LSTM model is established, and the motion response of a large buoy and an auxiliary buoy and the tension data of the anchoring system calculated by the optimized hydrodynamic mathematical model of a large buoy system are used to train the LSTM model, and the LSTM model is optimized based on the measured values. Based on the optimized LSTM model, accurate prediction of the motion response of a large buoy and an auxiliary buoy and the tension of the anchoring system can be achieved.

[0038] Figure 2 The diagram is a schematic diagram of a given large buoy mooring system. The large buoy system comprises a 10-meter buoy body, an auxiliary buoy and a mooring system, wherein the auxiliary buoy is connected to the 10-meter buoy body through an anchor chain; the buoy system is fixed by a three-point mooring method, and the mooring foundation of the large buoy adopts a Hall anchor.

[0039] Figure 3 This is the LSTM neural network prediction principle diagram of the present invention. Compared with the hidden layer of the original RNN, LSTM adds a cell state C t , as well as the original three gate control units, namely the forget gate, input gate, and output gate.

[0040] The forget gate f t and input gate it Using sigmoid as the activation function, the formula is as follows:

[0041] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0042] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0043] The activation function for the unit state update value usually uses tanh, and the formula is as follows:

[0044]

[0045] Output gate o t And the hidden state h t The formula is as follows:

[0046] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0047] h t =o t tanh(C t )

[0048] Among them, f t represents the forget gate, σ represents the nonlinear sigmoid function, W f ,b f Respectively represent the weight matrix and bias function of the forget gate, h t-1 ,x t Respectively represent the output result of the previous moment and the input of the current moment; i t represents the input gate, W i ,b i Respectively represent the weight matrix and bias function of the input gate; Represents the current state of the input gate, W c ,b c Respectively represent the weight matrix and bias function of the current state; C t Represents the state parameter at the current moment, C t-1 Indicates the state parameter of the previous moment; o t Represents the output gate, W o ,b oThey represent the weight matrix and bias function of the output gate respectively. t Represents the prediction result at the current moment calculated based on the current state parameters and the output gate result.

[0049] Figure 4 The LSTM model training process diagram of the present invention is shown in Figure 1. First, the initial parameters and target parameters are input, the weights and bias items of each hidden layer are obtained, the bias difference (loss) between the actual output value and the target value is calculated, and it is determined whether it exceeds the allowable error range. If it exceeds the allowable error range, a back propagation operation is performed, the weights are updated, and the parameters of each hidden layer are calculated again in a loop until the loss is less than the allowable error. The training is completed and the determined weights and biases are obtained.

[0050] Figure 5 Schematic diagram of the mean square error (MSE) convergence of the present invention. The predicted result is compared with the actual mooring system tension during the training process through the loss function. In order to minimize this loss, the parameters of the training process are continuously updated. Figure 5 As shown, when the loss value stabilizes and no longer decreases, the model is considered to be trained.

[0051] Figure 6 It is a schematic diagram comparing the predicted results of the tension of the mooring system of the present invention over time. After training, the predicted results are consistent with the actual results.

[0052] 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 large buoy motion response and mooring tension combined with deep learning, characterized in that: Based on the computational fluid dynamics open source code OpenFOAM, a hydrodynamic mathematical model of a large buoy system under the action of wind, waves and current is established; the input of the hydrodynamic mathematical model of the large buoy system is the wind field, wave height, wave period and current velocity, and the output is the motion response of the large buoy and auxiliary buoy and the tension of the mooring system; the motion response of the large buoy and auxiliary buoy and the tension of the mooring system are obtained by using a six-degree-of-freedom motion instrument and an underwater dynamometer, and the results are compared and verified with the calculation results of the hydrodynamic mathematical model of the large buoy system, so as to obtain an optimized hydrodynamic mathematical model of the large buoy system; Taking wind field, wave height, wave period, and current velocity as input, and the tension of the mooring system and the motion responses of large buoys and auxiliary buoys as output, a deep learning framework LSTM model is established. The LSTM model is trained using the motion responses of large buoys and auxiliary buoys and the tension of the mooring system calculated using the optimized hydrodynamic mathematical model of the large buoy system. The LSTM model is optimized based on the measured values. Based on the optimized LSTM model, accurate prediction of the motion responses of large buoys and auxiliary buoys and the tension of the mooring system can be achieved.

2. According to claim 1, a large buoy motion response and anchor tension prediction method combined with deep learning is characterized by: The large buoy system includes a ten-meter buoy body, an auxiliary buoy and an anchoring system. The auxiliary buoy is connected to the ten-meter buoy body through an anchor chain. The large buoy system is fixed by a three-point anchoring method, and the anchoring foundation of the large buoy adopts a Hall anchor.

3. The method for predicting large buoy motion response and mooring tension combined with deep learning according to claim 1, characterized in that: The hydrodynamic mathematical model of a large buoy system under the action of wind, waves and current is established based on the computational fluid dynamics open source code OpenFOAM, specifically: Set initial boundary conditions and input physical parameters in OpenFOAM, including wind field, wave height, wave period and current velocity; Based on the OpenFOAM standard solver olaFlow, the anchor chain solution module MOODY is implanted in the motion solution calculation of the large buoy system to solve the hydrodynamic mathematical model of the large buoy system, and obtain the motion response of the large buoy and auxiliary buoy and the tension of the mooring system; The method for obtaining the tension of the mooring system is as follows: based on the expansion of the dynamic equilibrium equation, the tension is calculated using the discontinuous finite element numerical method.

4. The method for predicting large buoy motion response and mooring tension combined with deep learning according to claim 1, characterized in that: The large buoy motion response equation obtained based on the hydrodynamic mathematical model of the large buoy system is as follows: Among them, K(t-τ) is the system delay function matrix; t is time; τ is the delay time; F i (t) is the environmental load acting on the buoy structure; F m (t) is the tension of the mooring system; X is the motion response matrix of the buoy in six degrees of freedom, including surge, sway, heave, roll, pitch and pitch; They represent the second-order and first-order derivatives of X respectively; M represents the mass matrix of the buoy; μ represents the additional mass matrix of the buoy; and C is the damping matrix.

5. According to claim 1, a large buoy motion response and anchor tension prediction method combined with deep learning is characterized by: Before using the LSTM model, the motion response of the large buoy, auxiliary buoy and the tension of the mooring system are normalized. The normalization equation is as follows: x norm =(x-x min ) / (x max -x min ) Among them, x is the actual value of the training data; x norm is the normalized value; x max and x min are the maximum and minimum values ​​of the training data respectively; Similarly, after the normalized data has been trained on the network, it must go through a denormalization process to generate the actual output data. The calculation equation for the actual output data is as follows: x=x norm (x max -x min )+x min 。 6. The method for predicting large buoy motion response and mooring tension combined with deep learning according to claim 1, characterized in that: According to the complexity of the data, different time window values, LSTM layers, number of neurons in each layer and optimizer are set for the LSTM model to find the optimal model; the judgment criterion of the optimal model is: the error between the model's prediction result and the measured value is the smallest.

7. The method for predicting large buoy motion response and mooring tension combined with deep learning according to claim 1, characterized in that: During the training process, the LSTM model is trained according to the error analysis function mean absolute error MAE, mean square error MSE, and maximum percentage of prediction error E max Train the neural network and update the parameters 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.

Images