CN-120892993-B - Dynamic submarine cable response extremum prediction method and device

CN120892993BCN 120892993 BCN120892993 BCN 120892993BCN-120892993-B

Abstract

The invention discloses a dynamic submarine cable response extremum prediction method and equipment integrating physical mechanism and dynamic time sequence characteristics, which are characterized in that a dynamic submarine cable control equation and data driving training are integrated deeply through a physical information neural network, the mechanical mechanism of a dynamic submarine cable is converted into a residual constraint item, the defect of poor generalization of a pure data driving method is overcome, the amplitude characteristic of six-degree-of-freedom motion of a floating body is extracted through an LSTM time sequence prediction model, a long time sequence dynamic coupling effect is compressed into a physical interpretable characteristic, a structural type judgment module is designed, two types of main submarine cable structures can be covered without reconstructing the model, engineering applicability is greatly enhanced, data fitting loss, physical residual error loss and motion prediction loss are dynamically balanced by adopting a time-varying weight strategy, convergence is accelerated, prediction results are ensured to meet a physical rule at the same time, and the extreme wind power strength failure of the dynamic submarine cable is provided by outputting key node response and motion amplitude characteristic in real time, so that the offshore wind power warning wind power is greatly reduced.

Inventors

GAO YANGYANG
WANG KAIMING
WANG LIZHONG

Assignees

浙江大学

Dates

Publication Date: 20260505
Application Date: 20250808

Claims (10)

1. A dynamic submarine cable response extremum prediction method integrating a physical mechanism and dynamic time sequence characteristics is characterized by comprising the following steps: Constructing a multi-working-condition training set, wherein the multi-working-condition training set comprises sea-condition parameter vectors and floating body six-degree-of-freedom displacement time sequence data which are input by a model, and response extremum and motion amplitude characteristics of dynamic sea cable key nodes which are output by the model; Constructing LSTM time sequence prediction model, inputting into six-freedom displacement time sequence data of floating body, the output is the motion amplitude characteristic; constructing a physical information neural network, inputting a sea state parameter vector, outputting a response extremum, and dynamically selecting whether to output geometric parameters according to the number parameters of the pontoons; a structure type judging module is constructed, a corresponding physical residual calculation path is selected according to the number parameters of the pontoons received in real time, and a physical residual is calculated based on the output item of the physical information neural network; Constructing a composite loss function integrating data fitting loss, physical residual loss and motion prediction loss, and dynamically balancing three losses by adopting a time-varying weight coefficient, wherein the data fitting loss represents the mean square error of a response extremum predicted value and a simulation value output by a physical information neural network; calculating the gradient of the composite loss function to the network weight through back propagation, and updating the network weight by adopting a gradient descent method until convergence; inputting real-time sea state parameters and floating body motion time sequence data into a fusion network after training is completed, and outputting a dynamic sea cable key node response extremum and a motion amplitude characteristic prediction result.
2. The method for predicting the dynamic submarine cable response extremum by fusing physical mechanism and dynamic time sequence characteristics according to claim 1, wherein the constructing the multi-working-condition training set specifically comprises: Sampling in a set sea state parameter range by adopting a space sampling method to obtain sea state parameter vectors and six-degree-of-freedom displacement time sequence data of the floating body, which are input as a model; and establishing a parameterized dynamic sea cable model through hydrodynamic analysis software, taking sea state parameter vectors as the model input of the dynamic sea cable model, and outputting the response extremum and the motion amplitude characteristic of the dynamic sea cable key nodes.
3. The method for predicting the response extremum of the dynamic submarine cable by combining the physical mechanism and the dynamic time sequence characteristic according to claim 2, wherein the sea state parameter vector comprises sense wave height, wave period, wave direction, surface flow velocity, surface flow direction, wind speed and wind direction, the response extremum comprises effective tension and bending radius, and the motion amplitude characteristic comprises displacement standard deviation and maximum displacement amplitude.
4. The method for predicting the dynamic submarine cable response extremum by fusing physical mechanism and dynamic time sequence characteristics according to claim 1, wherein in the LSTM time sequence predicting model, an input layer receives six-degree-of-freedom displacement time sequence data of a floating body, a hidden layer adopts a long-short-time memory network, and an output layer generates a motion amplitude characteristic vector.
5. The method for predicting the response extremum of the dynamic submarine cable by combining the physical mechanism and the dynamic time sequence characteristics according to claim 1 is characterized in that in the physical information neural network, an input layer receives sea state parameter vectors, a hidden layer adopts a fully-connected neural network, and an output layer dynamically selects output items according to the number of pontoons, wherein when the number of pontoons is 0, only the response extremum is output, and when the number of pontoons is >0, the response extremum and the geometric parameter are output.
6. The method for predicting a dynamic submarine cable response extremum by combining a physical mechanism and dynamic time sequence characteristics according to claim 1, wherein the selecting a corresponding physical residual calculation path according to the number parameters of the pontoons received in real time is as follows: when the number of the pontoons is 0, the dynamic submarine cable is in a catenary type, and a catenary control equation is selected to calculate a physical residual error; when the number of the pontoons is more than 0, the dynamic submarine cable is of a buffer type, and a buffer type control equation is selected to calculate a physical residual error.
7. The method for predicting the dynamic submarine cable response extremum by fusing physical mechanism and dynamic time sequence characteristics according to claim 6, wherein the calculation of the physical residual error by using a catenary control equation is as follows: , Wherein, the As a physical residual error, the difference between the two coefficients, Outputting a dynamic submarine cable tension predicted value at a suspension point by a physical information neural network; The horizontal external force is input by simulation for the suspension point; The mass of the dynamic submarine cable in unit length belongs to the material constant; Gravitational acceleration; For the length of the catenary portion, belongs to design parameters; the physical residual error is calculated by adopting a catenary control equation: , Wherein, the The angle of the connecting point of the buoy section and the lower catenary section is output by a physical information neural network; outputting the angle of the contact point of the seabed by a physical information neural network; the length of the lower suspension link section belongs to design parameters; is the wet weight ratio of the catenary segment, equal to , Is a horizontal tension, is input by simulation, Wet weight of sea cable unit length belongs to material constant; Is water depth, belongs to environmental input; For the segment height, k=1, 2,3, output by the physical information neural network.
8. The method for predicting dynamic sea cable response extremum by combining physical mechanism and dynamic time sequence characteristics according to claim 1, wherein the composite loss function The method comprises the following steps: , A loss term is fitted for data, and the mean square error of a response extremum predicted value and a simulation value output by the physical information neural network is represented; as a physical residual error loss term, the physical residual error obtained by the structure type judging module Calculated, i.e ; Representing the mean square error between the motion amplitude characteristic predicted value and the simulation value output by the LSTM time sequence prediction model as a motion prediction loss term; 、 And And respectively fitting a loss term, a physical residual error loss term and a time-varying weight coefficient of a motion prediction loss term for the data, wherein t is a training round.
9. The method for predicting dynamic sea cable response extremum by combining physical mechanism and dynamic time sequence characteristics according to claim 8, wherein the time-varying weight coefficient 、 And Dynamically adjusting according to training round t, and changing weight coefficient in time 、 And The specific calculation formula is as follows: , , , Wherein, the Losing initial weight for data fitting; the maximum weight is lost for the physical residual error; Loss of maximum weight for motion prediction; The rate of increase is lost for the physical residual, The loss decay rate is fitted to the data, Predicting a loss growth rate for motion, satisfying And leading the data fitting loss to be dominant in the initial training period, leading the physical residual error loss to be increased in the middle training period, and continuously increasing the motion prediction loss to a preset maximum value in the later training period.
10. A computer device, the computer device comprising: One or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the dynamic submarine cable response extremum prediction method according to any one of claims 1 to 9 that fuses physical mechanisms and dynamic timing characteristics.

Description

Dynamic submarine cable response extremum prediction method and device Technical Field The invention belongs to the technical field of offshore wind power generation, relates to a dynamic submarine cable response extremum prediction method and equipment, and particularly relates to a dynamic submarine cable response extremum prediction method and equipment integrating a physical mechanism and dynamic time sequence characteristics. Background With the progress of the offshore wind power industry, deep-open-sea floating wind power has become the main flow direction of offshore wind power development, and is also an international research hotspot of offshore wind power. However, the development of offshore floating fans is still facing serious technical challenges, one of which is to ensure the reliability and integrity of the transmission sea cable. The dynamic sea cable is used as a key power transmission component of the floating wind power system, is subjected to various complex load actions of the motion of a floating body and the load of the marine environment during service, and is easy to generate strength failure caused by overrun of tension or insufficient bending radius at key nodes (such as a top suspension point and a floating drum connecting area). Therefore, accurate prediction of the response extremum of the dynamic sea cable is important for real-time safety precaution of offshore wind power. However, there are a number of disadvantages to the prior art. The traditional numerical simulation method at present mainly comprises the steps of establishing a finite element model and calculating a response extremum under extreme load through time domain dynamic simulation. Although the method is high in precision, the method needs to be independently modeled and calculated for each sea condition, and has long time consumption and high calculation resource consumption, and cannot meet the real-time early warning requirement. In order to improve the calculation efficiency, part of researches adopt neural networks (such as BP, GA-BP and the like) to establish a mapping model of sea state parameters and response extremum. Although the method can obviously accelerate prediction, the method is essentially a 'black box model', and has serious defects such as strong dependence of training samples, lack of physical laws, poor structural adaptability and the like, and has the defect of insufficient generalization. In addition, the current research fails to effectively extract the dynamic influence of six-degree-of-freedom displacement time sequence characteristics of the floating body on submarine cable response, and the prediction reliability of motion amplitude characteristics (displacement standard deviation and maximum displacement amplitude) is difficult to ensure. Based on the above consideration, there is a need to develop a dynamic submarine cable response extremum prediction method integrating a physical mechanism and dynamic time sequence characteristics, which breaks through the bottleneck of numerical simulation and neural network simulation by a new method integrating the physical mechanism, the time sequence dynamic characteristics and the data driving advantages, and provides a high-reliability early warning means for the dynamic submarine cable strength failure of a large-scale floating fan. Disclosure of Invention Aiming at the problems that the traditional numerical method consumes a long time to cause the incapability of meeting the real-time early warning requirement and the traditional pure data driving neural network has insufficient generalization, the invention provides a dynamic submarine cable response extremum prediction method and equipment integrating a physical mechanism and dynamic time sequence characteristics, which are used for realizing the extreme load prediction and strength failure early warning of a dynamic submarine cable of a large-scale floating fan. The method is more accurate in response extremum prediction of the dynamic submarine cable of the large-scale floating fan, and is more efficient in iterative optimization. The technical scheme adopted by the invention is as follows: A dynamic submarine cable response extremum prediction method integrating physical mechanism and dynamic time sequence characteristics comprises the following steps: Constructing a multi-working-condition training set, wherein the multi-working-condition training set comprises sea-condition parameter vectors and floating body six-degree-of-freedom displacement time sequence data which are input by a model, and response extremum and motion amplitude characteristics of dynamic sea cable key nodes which are output by the model; Constructing LSTM time sequence prediction model, inputting into six-freedom displacement time sequence data of floating body, the output is the motion amplitude characteristic; constructing a physical information neural network, inputting a sea state parameter vector, out