Load quantification method for floating fan in typhoon scene

Abstract

The invention relates to a load quantification method for a floating fan in a typhoon scene, and the method comprises the steps: constructing an environment data simulation frame which comprises an external nesting and an internal nesting, the internal nesting is used for achieving wave flow coupling and generating wave flow scene data, and the external nesting is used for carrying out wave flow coupling and obtaining external environment data of the floating fan; the method comprises the following steps: constructing a multi-time-scale load prediction framework, inputting environmental data into the multi-time-scale load prediction framework, and outputting load distribution of typhoon to a floating fan under a long time scale, the multi-time-scale load prediction framework comprising a short-time-scale dynamic coupling analysis model and a long-time-scale load prediction model, and the long-time-scale dynamic coupling analysis model comprises a short-time-scale dynamic coupling analysis model and a long-time-scale dynamic coupling analysis model. The short-time-scale dynamic coupling analysis model is used for obtaining a load prediction sample, and the long-time-scale load prediction model is used for predicting the load prediction sample and outputting load distribution under a long time scale. Compared with the prior art, the method has the advantages of being more practical in environment simulation, accurate and efficient in load prediction and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of floating wind turbine load analysis, and in particular to a method for quantifying the load of a floating wind turbine in a typhoon scenario. Background Art

[0002] With the increasing depletion of offshore resources, the offshore wind power industry is gradually expanding from offshore areas to deep sea areas. In this context, the operation of floating wind turbines in deep sea areas faces severe challenges from extreme weather events such as typhoons. Once these wind turbines are hit by a typhoon, maintenance and overhaul work will become extremely difficult due to their remote geographical location and bad sea conditions. Therefore, developing an effective load prediction method and conducting in-depth research on wind, wave, and flow loads under typhoon conditions, especially under the interaction of strong sea air waves, is of great significance for improving the operation and maintenance efficiency of wind turbines, reducing costs, and ensuring the safe operation of wind turbines.

[0003] At present, the method of constructing typhoon environment scenarios mainly relies on historical meteorological observation data, which is used to establish a joint probability distribution model. However, since meteorological observation data under typhoon conditions are usually very limited, especially the lack of long-term and continuous wind and wave data, this limits the accuracy of the model. At the same time, the spatial resolution of existing meteorological observation data is usually around 30 kilometers, which is far from meeting the accuracy required for floating wind turbine load assessment.

[0004] In the load analysis of floating wind turbines, commonly used tools include aerodynamic-hydrodynamic-servo-elastic analysis software (FAST) and computational fluid dynamics (CFD) methods. Although these tools can simulate the load characteristics of wind turbines on a short time scale, they are unable to cope with long-time scale events such as typhoons. Therefore, it is necessary to develop new technologies or improve existing methods to achieve load analysis on a long time scale of typhoons, which has important guiding significance for the design, construction, and operation and maintenance of wind turbines. By improving the accuracy and reliability of load prediction, the maintenance plan of wind turbines can be planned more effectively, reducing downtime caused by extreme weather events, and ensuring the safe and stable operation of wind turbines in severe sea conditions. Summary of the invention

[0005] The purpose of the present invention is to provide a load quantification method for floating wind turbines in typhoon scenarios that can realize long-term load analysis.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for quantifying the load of a floating wind turbine in a typhoon scenario comprises the following steps:

[0008] Construct an environmental data simulation framework, which includes two layers, namely, external nesting and internal nesting. The internal nesting realizes wave-current coupling by improving the FVCOM-SWAVE model to generate wave-current scenario data. The external nesting performs wind-wave-current coupling under the typhoon scenario through the operation mode of the FVCOM-SWAVE-WRF large model to obtain the external environmental data of the floating wind turbine.

[0009] A multi-time-scale load prediction framework is constructed, the environmental data is input into the multi-time-scale load prediction framework, and the load distribution of the typhoon on the floating wind turbine on a long time scale is output, wherein the multi-time-scale load prediction framework includes a short-time-scale dynamic coupling analysis model and a long-time-scale load prediction model, the short-time-scale dynamic coupling analysis model is used to calculate the short-time-scale load distribution of the floating wind turbine according to the environmental data as a load prediction sample, and the long-time-scale load prediction model is used to predict the load prediction sample and output the load distribution of the typhoon on the floating wind turbine on a long time scale.

[0010] Furthermore, the improved FVCOM-SWAVE model realizes wave-current coupling and generates wave-current scenario data by simulating the interaction between ocean currents and wave currents, wherein in the simulation of the interaction between ocean currents and wave currents, at each time step, the FVCOM sub-model transmits variables including water level and vertical average flow velocity to the SWAVE sub-model, and the SWAVE sub-model transmits variables including significant wave height, spectral peak period, average period, wavelength and near-bottom maximum circumferential velocity to the FVCOM sub-model.

[0011] Furthermore, the process of generating wave and current scene data includes:

[0012] According to the wave-current coupling process, the FVCOM sub-model obtains the wave-current scenario data by solving the dynamic equation and the SWAVE sub-model obtains the wave-current scenario data by solving the wave spectrum energy equation, wherein the dynamic equation is:

[0013]

[0014] in:

[0015]

[0016] In the formula, is the current velocity; p is the water pressure; v is the dynamic viscosity; is an external force; is the radiation stress term, reflecting the momentum transfer effect of waves on the fluid; is a gradient operator, which is used to calculate the distribution change of wave energy in space; is the wave energy density spectrum, N(t) is the wave energy spectrum representing the wave state; is the group velocity vector; t is the time;

[0017] The wave spectrum energy equation is:

[0018]

[0019] Where, σ is the wave frequency; θ is the wave propagation direction; is the spatial coordinate; c σ is the propagation speed in frequency space; c θ is the propagation speed in the directional space; S tot It is the source and sink term.

[0020] Furthermore, the steps of solving the wave spectrum energy equation of the SWAVE sub-model include:

[0021] The SWAVE sub-model solves the wave spectrum energy equation to obtain the wave energy spectrum, which is simplified as:

[0022]

[0023] Where: f(N,t)=f propagation +f breaking +f wind +f bottom

[0024] Where N is N(t), which represents the wave energy spectrum of the wave state; f(N,t) represents the sum of the physical processes; f propagation is the wave propagation term, which represents the phase velocity and group velocity of the wave in the ocean; f breaking is the energy loss caused by wave breaking; f wind is the wind field term, which represents the input of wind field to wave energy; f bottom is the low friction term, which represents the energy dissipation caused by the friction between the wave and the seabed;

[0025] According to the wave energy spectrum, the time integration method is adopted and the fourth-order Runge-Kutta method is introduced to calculate and update the physical process items in 1) in each step in time series, wherein the expressions of the calculation and update process include:

[0026] k1=f propagation (N n )+f breaking (N n )+f wind (N n )+f bottom (N n )

[0027]

[0028]

[0029]

[0030] Where k1, k2, k3, k4 are calculated values ​​of different intermediate time steps, which are used to update the wave state of the next time step; the first step k1 is at the nth time step t n At, based on the current wave state N n The combined effect of all physical processes in the calculation; the second step size k2 uses half of the time step, that is, Update based on Calculate the physical process at the next moment; the third time step k3 is also based on the half-step time The intermediate step size k2 is updated; the last step size k4 is updated based on the complete time step Δt;

[0031] Take the weighted average of the four time steps to get the wave state of the next time step:

[0032]

[0033] Where N n+1 is the wave state for the next time step.

[0034] Furthermore, the FVCOM-SWAVE model is coupled with the WRF sub-model by using an MCT coupler to realize the wind-wave-current coupling process in a typhoon scenario, wherein the wind-wave-current coupling process specifically includes:

[0035] The wave and current scenario data are used as the initial and boundary conditions of the WRF sub-model;

[0036] The WRF sub-model calculates the meteorological data variables under the typhoon scenario and inputs them into the FVCOM-SWAVE model as the boundary condition input of the FVCOM-SWAVE model to affect the wave-current coupling process of the FVCOM-SWAVE model, wherein the meteorological data variables include heat flux, wind speed, wave radiation, and sea surface pressure;

[0037] The FVCOM-SWAVE model outputs variables including sea surface temperature, spectrum peak period, significant wave height, and wavelength, which are input into the WRF sub-model to affect the next generation of iterative calculations of the WRF sub-model;

[0038] In each iteration, the WRF submodel and the FVCOM-SWAVE model continuously exchange data until the coupling is completed.

[0039] Furthermore, the short-time scale dynamic coupling analysis model uses the FAST-AQWA model to calculate the short-time scale load distribution of the floating wind turbine. The specific calculation process includes:

[0040] The data of each time point in the environmental data is taken as the mean value, and combined with the wind wave current spectrum, a sliding window technology is used to select a short time scale time window environmental sample, wherein the wind wave current spectrum includes Kaimal wind spectrum, JONSWAP wave spectrum, and Garrett-Munk current spectrum;

[0041] The time window environmental sample is input into the FAST-AQWA model, and the short-time scale load distribution of the floating wind turbine is obtained by cyclic calculation. In the cyclic calculation process, the FAST sub-model obtains the tower base load by solving the dynamic equation, and the tower base load is transmitted to the AQWA sub-model as an external force. The AQWA sub-model calculates the floating platform position, floating platform speed and floating platform acceleration, and then transmits the calculation results of the AQWA sub-model to the FAST sub-model. The calculation is cyclically performed until the end of the cycle. The AQWA sub-model outputs the dynamic response of the wind turbine superstructure, and obtains the short-time scale load distribution of the floating wind turbine according to the dynamic response result. The dynamic equation is:

[0042]

[0043] Where M is the mass matrix composed of inertial mass and additional mass; C and K are the damping matrix and stiffness matrix after assembly, respectively; is the unknown velocity vector at time t; X(t) is the unknown displacement at time t; τ is the integral element; F(t) is the external load action term; is the unknown acceleration vector at time t; h(t) is the acceleration pulse function matrix, which is used to verify the radiation memory effect defined by the following formula:

[0044]

[0045] Where B(ω) is the radiation damping corresponding to the wave frequency ω.

[0046] Furthermore, each cycle calculation process of the FAST-AQWA model further includes the following steps:

[0047] When the tower foundation load obtained by the FAST sub-model is transferred to the AQWA sub-model, the translation load and the rotation load are converted, wherein the tower foundation load includes the translation load and the rotation load, and the conversion expression is:

[0048]

[0049] In the formula, F AQWA and FFAST are the translation force vectors in the AQWA submodel and the FAST submodel, respectively; It is T mat The inverse matrix of AQWA is the moment vector of the center of mass of the buoy relative to the inertial coordinate system of the AQWA submodel; M FAST It is the moment vector acting on the tower base obtained in the FAST sub-model relative to the local coordinate system of the tower base; COG is the position vector from the reference point to the center of mass of the floating platform.

[0050] Furthermore, each cycle calculation process of the AQWA sub-model further includes the following steps:

[0051] When the platform velocity obtained by the AQWA sub-model is passed to the FAST sub-model, the platform velocity is corrected:

[0052] V=UT mat ·COG×ω

[0053] Where V is the translation velocity vector used in FAST; U and ω are the translation velocity vector and rotation velocity vector of the platform obtained in AQWA, respectively; COG is the position vector from the reference point to the center of mass of the platform.

[0054] Furthermore, the long-time scale load prediction model includes a multi-scale convolutional neural network and a bidirectional long short-term memory network based on an attention mechanism, the bidirectional long short-term memory network based on the attention mechanism includes a Bi-LSTM layer and an attention layer, and the execution steps of the long-time scale load prediction model include:

[0055] The load prediction sample is input into a multi-scale convolutional neural network to extract multi-scale features, wherein the expression for extracting multi-scale features by the multi-scale convolutional neural network is:

[0056] y out =(X i *W)+b

[0057] In the formula, y out is the output feature after convolution; X i is the i-th input data, including wind speed, wave height, flow velocity and other features in the i-th time window; W represents the filter used for local feature extraction; b is the bias term, which is used to adjust the output features;

[0058] The multi-scale features are input into the bidirectional long short-term memory network based on the attention mechanism, and the load distribution under the long time scale is output. During the execution of the bidirectional long short-term memory network based on the attention mechanism, the Bi-LSTM layer performs time series modeling on the multi-scale features in time steps, and the attention layer calculates the weight of each time step. The operation expression of the Bi-LSTM layer is:

[0059] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0060] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0061] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0062] c t =f t *c t-1 +i t *tanh(W c ·[h t-1 ,x t ]+b c )

[0063] h t =o t *tanh(c t )

[0064] In the formula, x t is the input feature of time step t, i.e., multi-scale feature; h t is the hidden state at time step t, indicating the model’s memory of the current time step; c t is the state of the memory unit, which is used to save the long-term dependency information of the network; i t is the input gate; σ is the sigmoid activation function; b i 、b f 、b o 、b c is a bias term used to adjust the output result; W i , W f , W o , W c is the weight matrix; f tFor the forget gate; t is the output gate; c t is the updated memory cell state.

[0065] Furthermore, the operational expression of the attention layer is:

[0066]

[0067] In the formula, h t is the hidden state output by the Bi-LSTM layer, representing the features of time step t; W att is the attention weight matrix; e t is the attention score at time step t; c is the context vector, which represents the most important information in the sequence; T is the total length of the time series; e k is the kth attention score; α t is the attention weight at time step t; α k is the kth attention weight; h k is the kth output hidden state.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] (1) The present invention constructs an environmental data simulation framework, performs wave-current coupling by improving the FVCOM-SWAVE model, and performs wind-wave-current coupling by using the FVCOM-SWAVE-WRF large model, providing accurate simulation of external scene data, and performs predictions through a multi-time scale load prediction framework, wherein short-time scale load distribution is first analyzed through a short-time scale dynamic coupling analysis model, and long-time scale load prediction is then performed through a long-time scale load prediction model, thereby realizing long-time scale load analysis during typhoons.

[0070] (2) In the SWAVE sub-model of the present invention, the physical process items are calculated and updated in time series in the time step by combining the time integration method and the fourth-order Runge-Kutta method, which can solve the high gradient problem caused by typhoons. The FVCOM-SWAVE-WRF large model takes into account the multi-physical field coupling of typhoon processes, wind, waves and currents, and can extract the environmental data of a certain coordinate point or area in the simulation area, solving the problem of insufficient accuracy of historical observation data and providing more practical environmental conditions for the dynamic analysis of floating wind turbines.

[0071] (3) The present invention proposes a multi-time scale load prediction framework, in which the FAST-AQWA model transfers variables between the FAST sub-model and the AQWA sub-model, and through a cyclic calculation method, it can achieve complementary advantages and solve more accurate aerodynamic and hydrodynamic loads; the long-time scale load prediction model combines a multi-scale convolutional neural network with a bidirectional long short-term memory network based on an attention mechanism, which can quickly extract features and assign weights, thereby solving the problem of being unable to perform long-term load simulation on floating wind turbines and providing a more reliable basis for reliability analysis and operation and maintenance of floating wind turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0073] Figure 2 The unstructured triangular mesh area calculated when solving the FVCOM sub-model and the SWAVE model of the present invention;

[0074] Figure 3 It is the FVCOM-SWAVE-WRF large model simulation framework of the present invention;

[0075] Figure 4 is a multi-time-scale load prediction framework of the present invention;

[0076] Figure 5 The wave height data comparison before and after the improvement of the present invention and the wind speed field and flow velocity field at the same time;

[0077] Figure 6 Comparison of wave height and wave period between the simulation results of the present invention and the observation data of a certain station;

[0078] Figure 7 It is the time series environment data of a certain sample of the present invention;

[0079] Figure 8 The wind wave current load training set prediction effect of the present invention;

[0080] Fig. 9 This is the prediction effect of the wind wave and current load test set of the present invention. DETAILED DESCRIPTION

[0081] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0082] This embodiment provides a method for quantifying the load of a floating wind turbine in a typhoon scenario. Figure 1 As shown, the method comprises the following steps:

[0083] S1. Construct an improved FVCOM-SWAVE model (Finite Volume Coastal Ocean Model / Simulating WAves Nearshore) to simulate the interaction between ocean currents and wave currents, realize wave-current coupling, and generate wave-current scenario data.

[0084] S2. Construct the FVCOM-SWAVE-WRF (Finite Volume Coastal Ocean Model / Simulating WAves Nearshore-Weather Research and Forecasting) large model, simulate the combined effects of wind, waves and currents under typhoons according to its operation mode, and generate external environment data of floating wind turbines under typhoons.

[0085] S3. Select environmental samples from the environmental data and input them into the FAST-AQWA (Fatigue, Aerodynamics, Structures, and Turbulence-Airborne and Quantitative Wave Analysis) model to simulate the short-time scale load distribution of the floating wind turbine as a load prediction sample.

[0086] S4. Construct a long-time scale load prediction model, use the attention mechanism combined with the long short-term memory network to learn the samples, predict the wind turbine load conditions in different scenarios, and obtain the load distribution of typhoons on wind turbines on a long time scale.

[0087] S1~S2 is the environmental data simulation framework, which is divided into two layers of architecture, mainly including internal nesting and external nesting. The internal nesting is intended to realize the wave-current coupling between the FVCOM sub-model and the SWAVE sub-model. The external framework is intended to realize the wind-wave-current coupling through the operation mode of the FVCOM-SWAVE-WRF large model. The FVCOM-SWAVE model in step S1 includes the FVCOM sub-model and the SWAVE sub-model, and the FVCOM-SWAVE-WRF large model in step S2 includes the FVCOM-SWAVE model and the WRF sub-model.

[0088] S3-S4 is a multi-time scale load prediction framework, which generates short-time scale load samples based on environmental data and the FAST-AQWA model, and performs long-time scale prediction through multi-scale convolutional neural networks, self-attention mechanisms, and bidirectional long short-term memory networks. The FAST-AQWA model in step S3 includes a FAST sub-model and an AQWA sub-model.

[0089] The specific steps of step S1 include:

[0090] The FVCOM sub-model was jointly developed by the Institute of Marine Science and Technology at the University of Massachusetts and the Woods Hole Oceanographic Institution. It is mainly used for three-dimensional hydrodynamic numerical simulation. The SWAVE sub-model is a sub-module of the FVCOM sub-model. Variables can be assigned directly between memories without the need for a coupler. When the FVCOM sub-model and SWAVE are run separately, the mutual influence of ocean currents and waves cannot be taken into account in the calculation process, and the error between the solution and the true value is large. Therefore, the two need to be coupled and run. When the FVCOM sub-model and the SWAVE sub-model are running, the same unstructured triangular mesh area is used for calculation. The simulation area is as follows: Figure 2 As shown. Among them, FVOCM, as a current field simulation model, transmits the following variables to the SWAVE sub-model: water level and vertical average velocity; the SWAVE sub-model, as a wave field simulation model, transmits the following variables to the FVCOM sub-model: effective wave height, spectral peak period, average period, wavelength and maximum circumferential velocity near the bottom; the time step of the SWAVE sub-model is set to 30s, and the time step of the FVCOM sub-model is set to 5s, and the two transmit data every 30s. Through the set time step and transmission interval, the two sub-models continuously interact with each other to achieve wave and current coupling. According to the wave and current coupling process, the FVCOM sub-model obtains the wave and current scene data by solving the dynamic equations and the SWAVE sub-model obtains the wave spectrum energy equations. Specifically:

[0091] The spectral energy equation is solved according to the wave propagation, interaction, nonlinearity and dissipation processes. The SWAVE sub-model discretizes the spectral energy equation and calculates the distribution at different frequencies and energies.

[0092] The wave spectrum energy equation solved by the SWAVE sub-model is:

[0093]

[0094] In the formula, is the wave energy density spectrum; σ is the wave frequency; θ is the wave propagation direction; is the spatial coordinate; t is the time; is the group velocity vector, which is related to the wave number and frequency; The ocean current velocity is the input provided by the FVCOM sub-model, which affects the propagation speed and direction of wave energy; c σ is the propagation speed in frequency space; c θ is the propagation speed in the directional space; S tot It is the source and sink term.

[0095] The wave energy spectrum of the SWAVE sub-model can be simplified as:

[0096]

[0097] Where N(t) is the wave energy spectrum representing the wave state; f(N,t) represents the sum of the physical processes, which can be expressed as:

[0098] f(N,t)=f propagation +f breaking +f wind +f bottom (3)

[0099] In the formula, f propagation is the wave propagation term, which represents the phase velocity and group velocity of the wave in the ocean; f breaking is the energy loss caused by wave breaking; f wind is the wind field term, which represents the input of wind field to wave energy; f bottom is a low friction term, which represents the energy dissipation caused by the friction between the wave and the seabed.

[0100] Since the interaction between strong waves caused by wind stress and nonlinear waves is a typical high-gradient problem, the traditional time integration method using the SWAVE submodel is prone to dissipation of wave energy and instability of the solution. The fourth-order Runge-Kutta method is introduced. The physical process terms of formula (3) will be calculated and updated in time series in each step. The specific steps are as follows:

[0101] k1=f propagation (N n )+f breaking (N n )+f wind (N n )+f bottom (N n ) (4)

[0102]

[0103]

[0104]

[0105] Where k1, k2, k3, k4 are calculated values ​​of different intermediate time steps, which are used to update the wave state of the next time step. The first step k1 is at the nth time step t n At, based on the current wave state N n The combined effect of all physical processes in the calculation; the second step size k2 uses half of the time step, that is, Update based on Calculate the physical process at the next moment; the third time step k3 is also based on the half-step time The intermediate step size k2 is updated; the last step size k4 is based on the update of the complete time step Δt.

[0106] The four time steps are weighted averaged to get the next time step N. n+1 Wave state:

[0107]

[0108] The dynamic equations solved by the FVCOM sub-model are:

[0109]

[0110] In the formula, is the current velocity; p is the water pressure; v is the dynamic viscosity; is external force, such as gravity, Coriolis force, etc.; is the radiation stress term, reflecting the momentum transfer effect of waves on the fluid:

[0111]

[0112] The specific steps of step S2 include:

[0113] The WRF submodel is a numerical weather prediction system used for weather research and forecasting. It is suitable for weather forecasts of various scales and local weather forecasts, and can be used to simulate the wind environment under typhoons. The MCT coupler is a toolkit for coupling between different numerical models, usually used in climate and weather models. The solution method of this coupling model is as follows Figure 3 As shown in the figure, the variables transmitted from the WRF sub-model to the FVCOM-SWAVE model are: heat flux, 10m wind speed, wave radiation, and sea surface pressure; the variables transmitted from the FVCOM-SWAVE model to the WRF sub-model are: sea surface temperature, spectral peak period, significant wave height, and wavelength.

[0114] In the FVCOM-SWAVE-WRF large model coupling simulation, WRF itself solves the atmospheric dynamics equations (including momentum equations, continuity equations, thermodynamic equations, etc.), and provides atmospheric boundary conditions such as wind speed, temperature, and humidity. The FVCOM sub-model and the SWAVE sub-model provide ocean flow field and wave data respectively as boundary condition inputs for the WRF sub-model of the atmospheric model. The meteorological data (such as wind speed, temperature, etc.) output by the WRF sub-model are fed back to the FVCOM-SWAVE model, affecting the ocean flow field and wave field, thus forming a two-way coupling system for more accurate ocean-air coupling simulation.

[0115] The FVCOM submodel and the SWAVE submodel do not appear as direct source terms in the WRF equation. They affect the surface boundary conditions in the WRF submodel by providing information on sea surface temperature, ocean currents, and waves, and indirectly affect atmospheric parameters such as wind speed, air temperature, and humidity. Specifically, they affect the meteorological forecast results of the WRF submodel by affecting the friction, heat exchange, and water vapor exchange in the atmospheric boundary layer. Therefore, the influence of the FVCOM submodel and the SWAVE submodel on the WRF submodel is mainly reflected in the boundary conditions and sea-air interactions.

[0116] The specific steps of step S3 include:

[0117] After extracting the environmental data of the simulation area of ​​the FVCOM-SWAVE-WRF large model, considering the uncertainty of wind, wave and current under the typhoon, the data at each time point was taken as the mean, and the Kaimal wind spectrum, JONSWAP wave spectrum and Garrett-Munk flow spectrum were used to generate 1000s of input data. Then, using the sliding window technology, a time window of 660s was selected, that is, short-time scale environmental samples were used as the data input of the FAST-AQWA model. The FAST-AQWA model solves the blade, tower, floating platform, and mooring responses of the wind turbine through a cyclic calculation method, and further calculates the short-time scale wind, wave, and current loads of the floating wind turbine. During the cyclic calculation process, the FAST sub-model obtains the tower base load by solving the dynamic equation. The tower base load is used as an external force to transfer the sub-model to calculate the floating platform position, floating platform velocity, and floating platform acceleration, and then the calculation results of the AQWA sub-model are transferred to the FAST sub-model. The calculation is repeated in this way until the end of the cycle, and the floating platform position, floating platform velocity, and floating platform acceleration results finally calculated by the AQWA sub-model are obtained to determine the dynamic response of the wind turbine superstructure, and the short-time scale wind, wave, and current load distribution of the floating wind turbine is obtained based on the dynamic response results.

[0118] The equation of motion to be solved is:

[0119]

[0120] Where M is the mass matrix composed of inertial mass and additional mass; C and K are the damping matrix and stiffness matrix after assembly, respectively; is the unknown acceleration vector at time t; h(t) is the acceleration pulse function matrix, which is used to verify the radiation memory effect defined by the following formula:

[0121]

[0122] Where B(ω) is the radiation damping corresponding to the wave frequency ω.

[0123] When the platform velocity obtained by the AQWA sub-model is passed to the FAST sub-model, the platform translation velocity needs to be corrected. When the platform reference point defined in the FAST sub-model is the origin of the inertial coordinate system, that is, (0, 0, 0), the platform translation velocity can be corrected by the following formula:

[0124] V=UT mat ·COG×ω (13)

[0125] Where V is the translation velocity vector used in FAST; U and ω are the translation velocity vector and rotation velocity vector of the platform obtained in the AQWA submodel, respectively; COG is the position vector from the reference point to the center of mass of the platform.

[0126] The load vector in the FAST submodel also needs to be corrected when it is transferred to the AQWA submodel. When the tower base is the origin of the inertial coordinate system, the translation load and rotation load conversion need to be performed using the following formula:

[0127]

[0128] In the formula, F AQWA and F FAST are the translation force vectors in the AQWA submodel and the FAST submodel, respectively; It is T mat The inverse matrix of AQWA is the moment vector of the center of mass of the buoy relative to the inertial coordinate system of the AQWA submodel; M FAST It is the moment vector acting on the tower foundation obtained in the FAST sub-model relative to the tower foundation local coordinate system.

[0129] The specific steps of step S4 include:

[0130] The specific process is as follows Figure 4 As shown, the long-time scale load prediction model includes a multi-scale convolutional neural network (MSCNN) and a bidirectional long short-term memory network based on an attention mechanism. The bidirectional long short-term memory network based on an attention mechanism includes a Bi-LSTM layer and an attention layer.

[0131] The convolution kernels of multiple different scales of the multi-scale convolutional neural network can capture the changing characteristics of environmental data at different time scales; the self-attention mechanism is used to enhance the model's ability to focus on key moments in the input features. By assigning different weights to each time step, the model can selectively focus on time points that have a greater impact on the load; the bidirectional long short-term memory network is used to model the forward and reverse time dependencies in environmental features, thereby capturing the temporal relationship between environmental data and load.

[0132] First, a multi-scale convolutional neural network is needed to extract multi-scale features, so as to effectively capture the local dependencies of environmental data such as wind speed, wave height, and flow velocity. The calculation formula of MSCNN is as follows:

[0133] y out =(X i *W)+b (15)

[0134] In the formula, y out is the output feature after convolution; X i is the i-th input data, containing the wind speed, wave height, flow velocity and other features in the i-th time window; W represents the filter used for local feature extraction; b is the bias term used to adjust the output features.

[0135] Then, the feature quantity of the convolution output is used as the input of the Bi-LSTM layer for time series modeling. The calculation formula of the long short-term memory network (LSTM) can be expressed as:

[0136]

[0137] In the formula, x t is the input feature of time step t, i.e., the convolution feature extracted by MSCNN; h t is the hidden state at time step t, indicating the model’s memory of the current time step; c t It is the state of the memory unit, which is used to save the long-term dependency information of the network.

[0138] The Bi-LSTM layer contains a forward LSTM and a backward LSTM, which process different vectors of the input sequence respectively, and the output hidden state is the concatenation of the two. In this way, the Bi-LSTM layer can capture both past and future information of the input sequence.

[0139] By using the attention layer's ability to focus on key moments in the input features and assigning different weights to each time step, the model can selectively focus on time points that have a greater impact on the load. The calculation is divided into attention score calculation, attention weight calculation, and weighted context vector, which are expressed as follows:

[0140]

[0141] In the formula, h t is the hidden state of BiLSTM output; W att is the attention weight matrix, which is a learnable parameter; e t is the attention score; α is the attention weight; c is the context vector, which represents the most important information in the sequence.

[0142] The following simulation analysis is carried out:

[0143] 1. Comparison of typhoon environment simulation effects of FVCOM-SWAVE-WRF large model

[0144] The present invention takes a super typhoon as an example to simulate the wind-wave-current coupling environment field. Figure 5 Shown is the distribution of wind, wave and flow fields of the super typhoon on August 9, 2019.

[0145] The wind speed field map shows that the wind field is a typical symmetrical circulation structure, with the highest wind speed in the center, which is shown in the red area, with wind speed exceeding 45 meters per second, showing strong storm characteristics. This high-wind-speed core area usually corresponds to the "eye wall" of the typhoon, which is the part with the most concentrated typhoon energy and the most violent weather phenomena, including heavy rainfall, strong winds and turbulent activities. This area is prone to cause great load impact on the blades, towers, moorings and other components of floating wind turbines. The figure shows that the high-wind-speed area of ​​the typhoon radially weakens outward. This spatial variation of wind speed is crucial to load prediction because it can directly affect the stress distribution and dynamic response characteristics of the wind turbine in the entire wind field.

[0146] The velocity field shows that the velocity is highest on the right side of the typhoon path, reaching about 1.5 m / s, which is represented by the yellow area. This high velocity phenomenon is caused by the Ekman flow and wind-driven circulation caused by the strong wind field of the typhoon. The strong wind of the typhoon accelerates the surface seawater through the sea-air coupling effect, forming a significant ocean vortex structure and velocity gradient. The distribution of higher velocity on the right side is consistent with the asymmetric characteristics of the typhoon wind field and ocean current, that is, the wind field on the right side of the typhoon path in the northern hemisphere is stronger, resulting in higher surface velocity in the ocean. This high velocity poses a severe challenge to the mooring system and overall dynamic stability of the wind turbine. In areas where the velocity exceeds 1.5 m / s, the action of the ocean current may cause significant tension fluctuations on the anchor chain and mooring system, and may even cause fatigue damage to the mooring system. At the same time, high velocity may cause the wind turbine to drift and tilt significantly, which will have an adverse effect on the windward angle control and power generation efficiency of the blades. In addition, if the wind turbines are densely arranged, the strong velocity may aggravate the wake effect between the wind turbines, further affecting the dynamic operation of the entire wind farm.

[0147] The wave field map shows that in the red area near the center of the typhoon, the effective wave height reaches more than 8 meters, indicating that the sea area is affected by violent waves driven by the strong wind field of the typhoon. This high wave area is highly correlated with the strong wind speed distribution in the center of the typhoon, and is one of the most concentrated manifestations of typhoon energy. The strong wave loads caused by high waves will cause significant heave, roll and pitch movements of floating wind turbines, thereby increasing the stress on the mooring system, and may even produce stronger asymmetric loads on wind turbines within the range. This unbalanced wave impact may aggravate fatigue damage to wind turbines. As the distance from the center of the typhoon increases, the wave height gradually decreases, showing a concentric circle distribution, transitioning outward from the red area to the yellow, green and blue areas, indicating that the wave energy gradually decays with spatial diffusion.

[0148] At the same time, based on the observation data of a certain station from 0:00 on August 7, 2019 to 0:00 on August 13, 2019, such as Figure 6 As shown, the wave simulation method of the present invention is verified, and the results show that the improved method proposed in the present invention can effectively improve the simulation accuracy of the SWAVE sub-model. Model verification is quantified based on the root mean square error (RMSE) and the skill number. RMSE is a statistical method for judging the difference between measured data and simulated data, and the skill number is an indicator of whether the model data is consistent with the real data. The equations for the two evaluations are defined as follows:

[0149]

[0150] In the formula, is the i-th observation data, is the ith simulated data; n is the total number of observations. The smaller the RMSE value, the higher the accuracy of the model.

[0151]

[0152] Where M i and D i are the ith simulation results and measured data respectively; is the average value of the field measured data; N is the number of observation data.

[0153] 2. Wind, wave and current load prediction and verification

[0154] Based on the above wind-wave-current coupling simulation data, the environmental data at each moment of the coordinate point 122.2°E 27.3°N were selected, and the instantaneous fluctuation data were simulated through the Kaimal wind spectrum, JONSWAP wave spectrum, and Garrett-Munk flow spectrum. A set of sample data was generated as follows Figure 7As shown, the wind speed needs to further push the WRF output data according to the height of the wind turbine to get the wind speed at the hub, and use a set of input data as the FAST-AQWA model. Taking the 10MW OO-Star semi-submersible wind turbine designed by the EU project Lifes50+ as an example, the wind, wave and current loads on the floating wind turbine are calculated to obtain a set of load data samples.

[0155] The prediction effect verification proposed by the present invention is as follows Figure 8-9 As shown, model validation is quantified based on mean absolute percentage error (MAPE) and root mean square percentage error (RMSPE). The mean absolute percentage error indicates the magnitude of the forecast error relative to the actual value, while the root mean square percentage error measures the overall level of error by calculating the square root of the mean of the squares of the percentage errors between the forecast and the actual values. The two indicators are calculated as follows:

[0156]

[0157] In the formula, y i represents the actual value of sample i, represents the predicted value of sample i, and n represents the total number of samples.

[0158] The prediction results show that the model shows high accuracy and stability in the prediction of wind load, wave load and flow load. In the training set, the predicted value of wind load is highly consistent with the true value. Even in the area with large fluctuations, the model can still accurately track the actual value, showing strong learning ability. The overall trend of wave load prediction is relatively accurate, but there is a slight deviation in some peak areas, indicating that the model is slightly insufficient in fitting complex nonlinear dynamics, while the prediction of flow load is almost completely in line with the true value, especially in the area with small changes, showing a very high fitting effect. In the test set, the prediction effect of wind load is the best, the model has a strong ability to capture its dynamic characteristics, and the true value and the predicted value almost overlap. The deviation of wave load prediction in the fast fluctuation area increases slightly, but the overall trend can still follow the true value well; the predicted value of flow load is generally stable, but the error increases slightly in some descending areas, and further optimization is needed to cope with local fluctuations. Overall, the performance of the model on the training set and the test set shows that its prediction ability for wind load and flow load is particularly outstanding, and the prediction of wave load has the potential for further optimization, which provides a solid foundation for the accurate prediction of loads in practical applications.

[0159] Table 1 shows the quantitative results of the prediction error. The results show that the model shows good accuracy and stability in the prediction of wind load, wave load and flow load. Especially in the prediction of wind load, the MAPE and RMSPE of the test set are 1.5668% and 3.4591% respectively, indicating that the model has a strong ability to capture the dynamic characteristics of wind load and has high generalization performance. In terms of wave load, although its error is slightly higher, the MAPE and RMSPE of the test set are 5.9646% and 3.6517% respectively, showing that the model can still maintain strong stability and consistency when dealing with unseen data, which provides a good foundation for further optimization. The prediction error of flow load is low, and the MAPE and RMSPE of the test set are 3.3603% and 6.9259% respectively, indicating that the model can accurately capture the overall trend and has a good response to local changes. In general, the performance of the model on different load types shows high prediction performance, especially the performance on the test set shows that the model has reliability in practical applications, laying a solid foundation for wind turbine load prediction and optimization.

[0160] Table 1 Quantification of wind wave and current load prediction error

[0161] Load Type Dataset MAPE(%) RMSPE(%) Wind load Training set 1.9951 4.7564 Wind load Test Set 1.5668 3.4591 Wave load Training set 6.1702 5.2768 Wave load Test Set 5.9646 3.6517 Flow Load Training set 3.3427 7.5133 Flow Load Test Set 3.3603 6.9259

[0162] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0163] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0164] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0165] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0166] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0167] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0168] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for quantifying the load of a floating wind turbine in a typhoon scenario, characterized in that: The following steps are involved: Construct an environmental data simulation framework, which includes two layers, namely, external nesting and internal nesting. The internal nesting realizes wave-current coupling by improving the FVCOM-SWAVE model to generate wave-current scenario data. The external nesting performs wind-wave-current coupling under the typhoon scenario through the operation mode of the FVCOM-SWAVE-WRF large model to obtain the external environmental data of the floating wind turbine. A multi-time-scale load prediction framework is constructed, the environmental data is input into the multi-time-scale load prediction framework, and the load distribution of the typhoon on the floating wind turbine on a long time scale is output, wherein the multi-time-scale load prediction framework includes a short-time-scale dynamic coupling analysis model and a long-time-scale load prediction model, the short-time-scale dynamic coupling analysis model is used to calculate the short-time-scale load distribution of the floating wind turbine according to the environmental data as a load prediction sample, and the long-time-scale load prediction model is used to predict the load prediction sample, and output the load distribution of the typhoon on the floating wind turbine on a long time scale.

2. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 1 is characterized in that: The improved FVCOM-SWAVE model realizes wave-current coupling and generates wave-current scenario data by simulating the interaction between ocean currents and wave currents. In the simulation of the interaction between ocean currents and wave currents, at each time step, the FVCOM sub-model transmits variables including water level and vertical average flow velocity to the SWAVE sub-model, and the SWAVE sub-model transmits variables including significant wave height, spectral peak period, average period, wavelength and near-bottom maximum circumferential velocity to the FVCOM sub-model.

3. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 1 is characterized in that: The process of generating wave and current scene data includes: According to the wave-current coupling process, the FVCOM sub-model obtains the wave-current scenario data by solving the dynamic equation and the SWAVE sub-model obtains the wave-current scenario data by solving the wave spectrum energy equation, wherein the dynamic equation is: in: In the formula, is the velocity of the ocean current; p is the water pressure; v is the dynamic viscosity; is an external force; is the radiation stress term, reflecting the momentum transfer effect of waves on the fluid; is a gradient operator, which is used to calculate the distribution change of wave energy in space; is the wave energy density spectrum, N(t) is the wave energy spectrum representing the wave state; is the group velocity vector; t is the time; The wave spectrum energy equation is: Where, σ is the wave frequency; θ is the wave propagation direction; is the spatial coordinate; c σ is the propagation speed in frequency space; c θ is the propagation speed in the directional space; S tot It is the source and sink term.

4. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 3 is characterized in that: The steps of solving the wave spectrum energy equation by the SWAVE sub-model include: The SWAVE sub-model solves the wave spectrum energy equation to obtain the wave energy spectrum, which is simplified as: Where: f(N,t)=f propagation +f breaking +f wind +f bottom Where N is N(t), which represents the wave energy spectrum of the wave state; f(N,t) represents the sum of the physical processes; f propagation is the wave propagation term, which represents the phase velocity and group velocity of the wave in the ocean; f breaking is the energy loss caused by wave breaking; f wind is the wind field term, which represents the input of wind field to wave energy; f bottom is the low friction term, which represents the energy dissipation caused by the friction between the wave and the seabed; According to the wave energy spectrum, the time integration method is adopted and the fourth-order Runge-Kutta method is introduced to calculate and update the physical process terms in equation 1) in each step in time series, wherein the expressions of the calculation and update process include: k1=f propagation (N n )+f breaking (N n )+f wind (N n )+f bottom (N n ) k4=f propagation (N n +Δtk3)+f breaking (N n +Δtk3)+f wind (N n +Δtk3)+f bottom (N n +Δtk3) 4) Where k1, k2, k3, k4 are calculated values ​​of different intermediate time steps, which are used to update the wave state of the next time step; the first step k1 is at the nth time step t n At, based on the current wave state N n The combined effect of all physical processes in the calculation; the second step size k2 uses half of the time step, that is, Update based on Calculate the physical process at the next moment; the third time step k3 is also based on the half-step time The intermediate step size k2 is updated; the last step size k4 is updated based on the complete time step Δt; Take the weighted average of the four time steps to get the wave state of the next time step: Where N n+1 is the wave state for the next time step.

5. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 1, characterized in that: The FVCOM-SWAVE-WRF large model includes an FVCOM-SWAVE model and a WRF sub-model. The FVCOM-SWAVE model is coupled with the WRF sub-model using an MCT coupler to realize the wind-wave-current coupling process in a typhoon scenario. The wind-wave-current coupling process specifically includes: The wave and current scenario data are used as the initial and boundary conditions of the WRF sub-model; The WRF sub-model calculates the meteorological data variables under the typhoon scenario and inputs them into the FVCOM-SWAVE model as the boundary condition input of the FVCOM-SWAVE model to affect the wave-current coupling process of the FVCOM-SWAVE model, wherein the meteorological data variables include heat flux, wind speed, wave radiation, and sea surface pressure; The FVCOM-SWAVE model outputs variables including sea surface temperature, spectrum peak period, significant wave height, and wavelength, which are input into the WRF sub-model to affect the next generation of iterative calculations of the WRF sub-model; In each iteration, the WRF submodel and the FVCOM-SWAVE model continuously exchange data until the coupling is completed.

6. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 1, characterized in that: The short-time scale dynamic coupling analysis model uses the FAST-AQWA model to calculate the short-time scale load distribution of the floating wind turbine. The specific calculation process includes: The data of each time point in the environmental data is taken as the mean value, and combined with the wind wave current spectrum, a sliding window technology is used to select a short time scale time window environmental sample, wherein the wind wave current spectrum includes Kaimal wind spectrum, JONSWAP wave spectrum, and Garrett-Munk current spectrum; The time window environmental sample is input into the FAST-AQWA model, and the short-time scale load distribution of the floating wind turbine is obtained by cyclic calculation. In the cyclic calculation process, the FAST sub-model obtains the tower base load by solving the dynamic equation, and the tower base load is transmitted to the AQWA sub-model as an external force. The AQWA sub-model calculates the floating platform position, floating platform speed and floating platform acceleration, and then transmits the calculation results of the AQWA sub-model to the FAST sub-model. The calculation is cyclically performed until the end of the cycle. The AQWA sub-model outputs the dynamic response of the wind turbine superstructure, and obtains the short-time scale load distribution of the floating wind turbine according to the dynamic response result. The dynamic equation is: Where M is the mass matrix composed of inertial mass and additional mass; C and K are the damping matrix and stiffness matrix after assembly, respectively; is the unknown velocity vector at time t; X(t) is the unknown displacement at time t; τ is the integral element; F(t) is the external load action term; is the unknown acceleration vector at time t; h(t) is the acceleration pulse function matrix, which is used to verify the radiation memory effect defined by the following formula: Where B(ω) is the radiation damping corresponding to the wave frequency ω.

7. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 6, characterized in that: Each cycle calculation process of the FAST-AQWA model also includes the following steps: When the tower foundation load obtained by the FAST sub-model is transferred to the AQWA sub-model, the translation load and the rotation load are converted, wherein the tower foundation load includes the translation load and the rotation load, and the conversion expression is: In the formula, F AQWA and F FAST are the translation force vectors in the AQWA submodel and the FAST submodel, respectively; It is T mat The inverse matrix of AQWA is the moment vector of the center of mass of the buoy relative to the inertial coordinate system of the AQWA submodel; M FAST It is the moment vector acting on the tower base obtained in the FAST sub-model relative to the local coordinate system of the tower base; COG is the position vector from the reference point to the center of mass of the floating platform.

8. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 6, characterized in that: Each cycle calculation process of the AQWA sub-model also includes the following steps: When the platform velocity obtained by the AQWA sub-model is passed to the FAST sub-model, the platform velocity is corrected: V=U-T mat ·COG×ω Where V is the translation velocity vector used in FAST; U and ω are the translation velocity vector and rotation velocity vector of the platform obtained in AQWA, respectively; COG is the position vector from the reference point to the center of mass of the platform.

9. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 1, characterized in that: The long-time scale load prediction model includes a multi-scale convolutional neural network and a bidirectional long short-term memory network based on an attention mechanism, wherein the bidirectional long short-term memory network based on an attention mechanism includes a Bi-LSTM layer and an attention layer, and the execution steps of the long-time scale load prediction model include: The load prediction sample is input into a multi-scale convolutional neural network to extract multi-scale features, wherein the expression for extracting multi-scale features by the multi-scale convolutional neural network is: y out =(X i *W)+b In the formula, y out is the output feature after convolution; X i is the i-th input data, including wind speed, wave height, flow velocity and other features in the i-th time window; W represents the filter used for local feature extraction; b is the bias term, which is used to adjust the output features; The multi-scale features are input into the bidirectional long short-term memory network based on the attention mechanism, and the load distribution under the long time scale is output. During the execution of the bidirectional long short-term memory network based on the attention mechanism, the Bi-LSTM layer performs time series modeling on the multi-scale features in time steps, and the attention layer calculates the weight of each time step. The operation expression of the Bi-LSTM layer is: i t =σ(W i ·[h t-1 ,x t ]+b i ) f t =σ(W f ·[h t-1 ,x t ]+b f ) the t =σ(W o ·[h t-1 ,x t ]+b o ) c t =f t *c t-1 +i t *tanh(W c ·[h t-1 ,x t ]+b c ) h t =o t *tanh(c t ) In the formula, x t is the input feature of time step t, i.e., multi-scale feature; h t is the hidden state at time step t, indicating the model’s memory of the current time step; c t is the state of the memory unit, which is used to save the long-term dependency information of the network; i t is the input gate; σ is the sigmoid activation function; b i 、b f 、b o 、b c is a bias term used to adjust the output result; W i , W f , W o , W c is the weight matrix; f t For the forget gate; t is the output gate; c t is the updated memory cell state.

10. The method for quantifying the load of a floating wind turbine in a typhoon scenario according to claim 9, characterized in that: The operational expression of the attention layer is: In the formula, h t is the hidden state output by the Bi-LSTM layer, representing the features of time step t; W att is the attention weight matrix; e t is the attention score at time step t; c is the context vector, which represents the most important information in the sequence; T is the total length of the time series; e k is the kth attention score; α t is the attention weight at time step t; α k is the kth attention weight; h k is the kth output hidden state.

Images