AI-enhanced floating turbine six-degree-of-freedom motion actuation line calculation method

Abstract

The invention discloses an AI-enhanced floating turbine six-degree-of-freedom motion actuation line calculation method, and belongs to the crossing field of computational fluid mechanics and ocean renewable energy technologies. Comprising the following steps: S1, constructing a basic actuating line model, taking a discrete blade as a blade element, and calculating an initial flow field and hydrodynamic parameters; s2, a three-level coordinate transformation system is established, and speed and attack angle correction of six-degree-of-freedom motion is completed; s3, predicting a dynamic lift-drag coefficient based on the time sequence features through a pre-trained deep learning agent model; s4, smoothly mapping the blade element hydrodynamic force into a flow field momentum source item, and solving a transient flow field in combination with CFD; and S5, constructing a'motion-flow field-hydrodynamic force 'iterative closed loop to realize dynamic coupling simulation. According to the method, the advantages of physical modeling and data driving are fused, the dynamic load prediction deviation is reduced, the calculation efficiency is improved compared with that of a full-geometric LES, and the method can be widely applied to optimization design and performance evaluation of the floating water turbine.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of computational fluid dynamics (CFD) and marine renewable energy technology, specifically involving an AI-enhanced method for calculating the six-degree-of-freedom motion actuation lines of a floating turbine. Background Technology

[0002] As the global energy structure shifts towards cleaner and lower-carbon energy sources, tidal energy, as a abundant and highly predictable marine renewable energy source, has become an important direction for addressing the energy crisis and environmental issues through its large-scale development and utilization. Floating tidal turbines, with their core advantages such as flexible installation depth, low dependence on seabed topography, and suitability for deep-sea resource development, have become a research hotspot and industrialization focus in the field of marine energy equipment.

[0003] However, the service environment of floating turbines is complex. The platform experiences significant six-degree-of-freedom motion under the combined effects of waves and ocean currents. These include rotational degrees of freedom: yaw (rotation about the vertical axis), roll (rotation about the longitudinal axis), and pitch (rotation about the lateral axis); and translational degrees of freedom: sway (translation along the longitudinal direction), heave (translation along the vertical direction), and sway (translation along the lateral direction). These motions directly alter the relative flow conditions of the turbine blades: on the one hand, rotational degrees of freedom induce additional tangential velocity and Coriolis acceleration in the blade element, leading to high-frequency fluctuations in the local angle of attack; on the other hand, translational degrees of freedom cause a superimposed shift in the incoming flow velocity of the blade element, further exacerbating the unsteadiness of the flow field. These effects collectively cause dynamic stall hysteresis in the blade airfoil (hydrodynamics has a strong history dependence), periodic load oscillations, and energy conversion efficiency degradation, posing significant challenges to the structural strength design, fatigue life assessment, and control strategy optimization of the turbine.

[0004] The Actuator Line Method (ALM), a fluid-structure interaction simulation technique that balances computational efficiency and accuracy, discretizes blades into "blade elements" distributed along the spanwise direction. It calculates local hydrodynamics based on blade element momentum theory and maps these dynamics to the flow field mesh as momentum source terms. It has been widely applied in numerical simulations of stationary wind turbines and tidal turbines. However, the traditional Actuator Line Method suffers from three major technical drawbacks when adapting to the simulation of six-degree-of-freedom motion of floating hydro turbines: 1. Lack of motion-flow coupling mechanism: Traditional ALM assumes the impeller is in a fixed coordinate system or simple translation state, and does not establish a system coupling model between the six degrees of freedom motion and the local flow field of the blade element. Especially for the rotational degrees of freedom (such as yaw and roll), the influence of the Coriolis acceleration generated by these is completely ignored on the relative velocity of the blade element, resulting in an angle of attack calculation error of more than 10°, which directly causes hydrodynamic prediction deviation; 2. Dynamic stall effect cannot be characterized: Traditional ALM relies on static airfoil data tables (lookup table method) to obtain lift-drag coefficients. However, during dynamic stall, hydrodynamics depends not only on the current angle of attack but also on the rate of change of the angle of attack and its historical trajectory (i.e., "hysteresis effect"). Static data tables cannot reflect this nonlinear characteristic, resulting in insufficient accuracy in dynamic load prediction, with deviations from experimental results often exceeding 30%. 3. Difficulty in balancing accuracy and efficiency: Although high-fidelity CFD methods (such as large eddy simulation, LES) can capture dynamic stall and complex flow field structures through full geometric resolution, their computational load is huge—single-condition simulation requires more than one million grids and thousands of hours of computation time, which cannot meet the efficiency requirements of "multi-condition optimization" and "long-term performance evaluation" in engineering design; while traditional ALM is computationally efficient, its accuracy is difficult to guarantee due to the above-mentioned defects.

[0005] To address these issues, existing research attempts to improve the model in two ways: Firstly, some scholars have introduced motion corrections for single degrees of freedom (such as heave) into ALM, but this has not achieved coupling processing of all six degrees of freedom and has neglected the Coriolis effect of rotational degrees of freedom. Secondly, machine learning methods (such as Long Short-Term Memory networks LSTM) have begun to emerge in unsteady hydrodynamic prediction, but a technical solution specifically designed for six-degree-of-freedom motion simulation of floating turbines, deeply integrated with actuation line methods, has not yet been formed. Existing AI models mostly focus on hydrodynamic fitting under static conditions, without considering the influence of motion parameters on feature inputs, and have not achieved real-time collaboration with CFD solvers. Summary of the Invention

[0006] The purpose of this invention is to provide an AI-enhanced method for calculating the six-degree-of-freedom motion actuation lines of a floating turbine, thereby solving the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following solution: An AI-enhanced method for calculating the six-DOF motion actuation lines of a floating hydro turbine includes the following steps: Step S1. Construct a basic actuation line model: Discretize the blades of the floating tidal power turbine along the spanwise direction into several blade elements. Based on the blade element momentum theory, calculate the basic incoming flow velocity, local initial angle of attack, and initial hydrodynamic force of each blade element in a fixed coordinate system. The initial hydrodynamic force is initially calculated using the basic incoming flow velocity, initial angle of attack, and static airfoil data of the blade elements.

[0008] Step S2. Six-DOF motion coupling correction: Establish a three-level coordinate transformation system of "global coordinate system - body coordinate system - relative rotation coordinate system" to perform kinematic corrections on the yaw, roll, and pitch rotational degrees of freedom and the sway, heave, and roll translational degrees of freedom of the floating platform; calculate the additional velocities generated by each degree of freedom motion, including the tangential additional velocity and Coriolis acceleration of the rotational degree of freedom and the translational additional velocity of the translational degree of freedom, and synthesize the additional velocities with the basic incoming flow velocity vector of step S1 to update the true relative incoming flow velocity and true local angle of attack of the blade elements.

[0009] Step S3. AI-enhanced hydrodynamic coefficient prediction: Based on the real local angle of attack, real relative inflow velocity, and real-time six-degree-of-freedom motion parameters of the floating platform updated in Step S2, a time-series feature vector is constructed; the time-series feature vector is input into a pre-trained deep learning proxy model, which outputs the dynamic lift coefficient and dynamic drag coefficient of the blade element, and calculates the corrected blade element hydrodynamics in combination with the real relative inflow velocity; the deep learning proxy model can capture the dynamic stall hysteresis effect and the hydrodynamic history dependence.

[0010] Step S4. Volume force mapping and CFD transient solution: The blade element hydrodynamics corrected in step S3 are smoothly mapped to the flow field grid using a Gaussian distribution function to form momentum source terms; the momentum source terms are embedded in the CFD solver to solve the unsteady Navier-Stokes equations and obtain the dynamic hydrodynamic performance parameters and flow field structure parameters of the turbine.

[0011] Step S5. Iterative Update: Feed back the flow field structure parameters output in step S4 to step S2, update the motion parameters of the floating platform, and repeat steps S2-S4 until the transient simulation of the preset duration is completed.

[0012] Further optimization, in step S1, the specific process of calculation based on leaf element momentum theory includes: Step S11. Distribute the turbine blades evenly along the spanwise direction (from blade root to blade tip) into Ns blade elements (typically Ns = 20~30, to ensure a balance between accuracy and efficiency). The key number of each blade element includes the number of blades. n b Leaf element span dr (Intercenter distance between adjacent leaf elements), local chord length c Local pitch angle γ, blade rotational angular velocity ω, and blade element spanwise distance from the blade root. r , 0≤r≤R, where R is the blade radius.

[0013] Step S12. Calculate the basic velocity components: Based on the blade element momentum theory, neglecting platform motion, the basic incoming flow velocity of the blade element is the superposition of the free incoming flow velocity and the induced velocity. Among them, the axial velocity... Ux =U 0+ u tangential velocity U θ = ωr - v Basic relative inflow velocity ;in, U 0 represents the incoming flow velocity; u The axial velocity, generated by the wake effect, can be initially estimated using a simplified wake model. v To induce tangential velocity; Step S13. Calculate the initial angle of attack: Inflow angle φ0 = arctan( Ux / U θ The initial local angle of attack α0 = φ0 - γ is the angle between the airfoil chord and the direction of the relative incoming flow.

[0014] Step S14. Calculate the initial hydrodynamic force: initial lift L 0 = (1 / 2) n b ρU rel,0 2 cC L,0 dr Initial resistance D 0 = (1 / 2) n b ρU rel,0 2 cC D,0 dr ;in, ρ For fluid density, C L,0 , C D,0 The lift and drag coefficients are based on static airfoil data tables. The initial lift and drag are decomposed into axial (incoming flow direction) and tangential (rotational tangential direction) components: Initial axial force F x , θ =L0cosα0+D0sinα0、Tangential force F θ,0 =L0sinα0-D0cosα0, which prepares for the subsequent momentum source term mapping.

[0015] Further optimization involves the following step S2: six-degree-of-freedom motion coupling correction. A three-level coordinate transformation system is established: "global coordinate system - body coordinate system - relative rotation coordinate system". Kinematic corrections are made to the rotational degrees of freedom (bow, roll, pitch) and translational degrees of freedom (sway, heave, sway) of the floating platform. The additional velocities generated by each degree of freedom motion are calculated and combined with the basic incoming flow velocity vector from step S1 to update the true relative incoming flow velocity and true local angle of attack of the blade elements.

[0016] Step S21. Define three types of coordinate systems: ① Global coordinate system (G system): fixed to the seabed, with the X-axis along the direction of the incoming flow, the Y-axis along the transverse direction (perpendicular to the incoming flow), and the Z-axis along the vertical direction (positive upwards), used to describe the absolute motion of the platform; ② Body coordinate system (B system): fixed to the center of mass of the floating platform, with coordinate axes parallel to the G system, and translating synchronously with the platform (without rotation), used to describe the position of blade elements relative to the platform; ③ Relative rotation coordinate system (R system): for rotational degrees of freedom (such as yaw), with the center of rotation as the origin, the coordinate axes rotate synchronously with the platform, used to calculate the additional velocity caused by rotation.

[0017] The specific process of kinematic correction for rotational degrees of freedom includes: Step S22. Define the equation of rotational motion: For any rotational degree of freedom, let the rotation angle be ξ(t) and the angular velocity be... Angular acceleration is , where t is time; ψ0 is the maximum bow roll angle; and T is the motion period.

[0018] Bow rocking: ξ yaw (t)=ψ0sin(2πt / T), R yaw The radius of the bow rotation; Horizontal sway: ξ roll (t)=φ0sin(2πt / T), R roll The radius of the roll rotation; Swaying: ξ pitch (t)=θ0sin(2πt / T), R pitch The radius of the pitch rotation; Step S23. Determine the coordinates of the rotation center: Calculate the center point of the blade's rotation plane. Oc Real-time coordinates in the global coordinate system x c , y c , z c Let R be the radius of rotation for this rotational degree of freedom. ξ ; Step S24. Coordinate Transformation: Using rotation matrix M ξ The original coordinates of the leaf element in the global coordinate system ( x, y Transform the coordinates (z) to a relative rotating coordinate system. x r , y r , z r ),Right now( x r , y r , z r )ᵀ=M ξ ·( x - x c , y - y c ,z- z c )ᵀ; Step S25. Calculate the additional velocity: tangential additional velocity Coriolis acceleration Where Rxy is the distance from the leaf element to the center of rotation in the XY plane of the relative rotating coordinate system; Step S26. Update velocity and angle of attack: Combine the tangential additional velocity with the basic tangential velocity, correct the axial velocity with Coriolis acceleration, and obtain the relative incoming flow velocity after rotational correction. U rel,ξ ;based on U rel,ξ Recalculate the inflow angle φ ξ Update the actual local angle of attack α ξ =φ ξ -γ.

[0019] The specific process of kinematic correction for translational degrees of freedom includes: Step S27. Define the equation of translational motion: For any translational degree of freedom, let the translational displacement be s(t) and the velocity be... acceleration is ; Sway: Movement along the X-axis of the global coordinate system, displacement s surge (t), velocity ; Heave: Movement along the Z-axis of the global coordinate system, displacement s heave (t), velocity ; Sway: Movement along the Y-axis of the global coordinate system, displacement s sway (t), velocity .

[0020] Step S28. Calculate the additional velocity: Combine the translational velocity The relative incoming flow velocity after translational correction is obtained by vector superposition with the basic incoming flow velocity from step S1. U rel,s ; Step S29. Update angle of attack: based on U rel,s Recalculate the inflow angle φ based on the velocity direction. s Update the actual local angle of attack α s =φ s -γ.

[0021] Further optimization involves determining the true relative incoming flow velocity and the true local angle of attack in step S2 as follows: if both rotational and translational degrees of freedom exist simultaneously, the rotationally corrected... U rel,ξ With translational correction U rel,s Further vector synthesis yields the final true relative inflow velocity. U rel,true ;based on U rel,true Calculate the final inflow angle φ true Ultimately, the true local angle of attack α true =φ true -γ.

[0022] Further optimization is made in step S3, where the deep learning agent model is a recurrent neural network or a graph neural network; 1) If it is a recurrent neural network, specifically a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU); the mathematical model of the LSTM includes: Input Gate: ; Forgotten Gate: ; Candidate cell status: ; Cell status update: ; Output gate: ; Hidden state output: ; Where σ is the sigmoid activation function, and ⊙ represents element-wise multiplication. W i , W f , W C , W o This is the weight matrix. b i , b f , b C, b o For bias vectors, h t-1 The hidden state of the previous time step. C t-1 This represents the cell state at the previous time step. x t This is the feature vector at the current time step; 2) If it is a graph neural network, all blade elements of the turbine are constructed as a graph structure, with blade elements as nodes of the graph and wake interference relationships between blade elements as edges of the graph. The neighborhood information of the nodes is aggregated through a message passing mechanism to collaboratively predict the dynamic lift coefficient and dynamic drag coefficient of each blade element.

[0023] Further optimization is achieved by the following step S3: the time series feature vector is specifically constructed as follows: x t =[α true,t Re t V x,t V y,t V z,t ,ω roll,t ,ω pitch,t ,ω yaw,t ,φ t ,θ t ,ψ t M t ], where: α true,t For step S2, update the final true local angle of attack at time step t; Re t Let Re be the Reynolds number at time step t; the formula is Re t =ρU rel,true,t c / μ, where μ is the hydrodynamic viscosity, U rel,true,t V represents the final true relative inflow velocity at time step t. x,t V y,t V z,t Let ω represent the linear velocities of the floating platform's center of mass in the x, y, and z directions in the global coordinate system at time step t; roll,t ,ω pitch,t ,ω yaw,t These represent the roll, pitch, and bow angular velocities of the floating platform at time step t; φ t ,θ t ,ψ t These represent the Euler angles of the floating platform at time step t: roll, pitch, and bow. t The time step t is the Mach number based on the tip velocity, calculated using the formula M. t =(ωR+U0) / a, where R is the blade radius and a is the speed of sound in the fluid.

[0024] The time series is constructed as follows: Feature vectors from the past N time steps are selected to form the input sequence X = [ x t-N+1 , x t-N+2 ,..., x t-1 , x t The time series length N ranges from 20 to 100 time steps; the specific value of N is determined by the dynamic stall characteristic time scale T. stall Sure: Where Δt is the time step of the CFD transient simulation, This indicates rounding up to the nearest integer.

[0025] Further optimization, in step S3, the specific process of pre-training the deep learning agent model includes: Step S31. Dataset Construction: Training data is generated using Large Eddy Simulation (LES). The LES model includes the geometry of the entire turbine blades and the rigid body model of the floating platform, with a mesh count ≥ 5 million and a time step Δt. LES =0.001~0.005s, simulation duration ≥20 floating platform motion cycles; extract the "time series feature vector + true lift-drag coefficient label" of each leaf element, total sample size M≥5×10 4 ; Step S32. Data preprocessing: Standardize the feature vectors using Z-score; divide the training set, validation set, and test set into a 7:2:1 ratio; Step S33. Training parameter configuration: The backpropagation time series algorithm BPTT is used, the optimizer is Adam (β1=0.9, β2=0.999, ε=1e-8), the initial learning rate is 1e-4, and the learning rate is adjusted by cosine annealing; the batch size is 16~64, and the gradient clipping threshold is 1.0 (L2 norm). Step S34. Loss Function and Regularization: The loss function is the mean squared error. , Where B is the batch size. C L,AI , C D,AI To predict the lift-to-drag ratio, C L,true , C D,true Label the model as LES; employ an early stopping strategy (e.g., stop if the loss on the validation set does not decrease after 10-20 consecutive rounds) and dropout technique (dropout rate 0.2-0.5) to prevent overfitting; Step S35. Model Evaluation: Test set evaluation metrics include Mean Absolute Error (MAE) ≤ 0.05 and Coefficient of Determination (R²). 2 ≥0.9, to ensure the model's generalization ability.

[0026] Further optimization is achieved in step S3, where the revised formula for calculating the hydrodynamic force of the blade element is: Dynamic lift L AI =(1 / 2)ρU rel,true 2 cC L,AI dr, dynamic resistance D AI =(1 / 2)ρU rel,true 2 cC D,AI dr; will L AI With D AI Decomposed into axial force F x =L AI cosα true +D AI sinα true Tangential force F θ =L AI sinα true -D AI cosα true As the hydrodynamic input of the leaf element in step S4; The deployment method for the deep learning proxy model is as follows: the pre-trained model is converted to ONNX format, and a CFD solver is embedded through a user-defined function (UDF) interface; within each time step, the UDF reads the updated parameters from the CFD solver in step S2, constructs a time-series feature vector, and calls the ONNX model for inference output C. L,AI With C D,AI Inference time is <1ms / time step.

[0027] Further optimization is achieved by defining the Gaussian distribution function in step S4 as f(r) = (1 / (πσ)). 2 ))exp(-r g 2 / σ 2 ), where r g σ is the distance from the blade element center to the flow field grid node, and σ is the smoothing radius, which takes the value of 0.5 to 1.0 times the blade element spanwise length dr; dynamic hydrodynamic performance parameters include the turbine power coefficient C. P Thrust coefficient C T The flow field structure parameters include flow field velocity, pressure, and vorticity distribution; In step S5, the specific logic of the iterative update is as follows: after each CFD time step is completed, the force and torque of the flow field on the floating platform are extracted, the motion equation parameters (displacement, angular velocity, angular acceleration) of the platform are updated, and the updated motion parameters are input into step S2 to recalculate the additional velocity, so as to realize the dynamic coupling closed loop of "motion-flow field-hydrodynamics".

[0028] A computing system for performing the above method includes: Parameter input module: Receives turbine geometric parameters, environmental parameters, and initial motion parameters of the floating platform; Actuation line calculation module: performs the basic actuation line model construction in step S1 and the six-degree-of-freedom motion coupling correction in step S2, and outputs the true relative incoming flow velocity and the true local angle of attack; AI Inference Module: Loads a pre-trained ONNX format deep learning proxy model, performs feature construction and inference in step S3, and outputs the dynamic lift-drag coefficient and the corrected leaf element hydrodynamics. CFD Solver Module: Performs the volume force mapping and unsteady Navier-Stokes equation solution in step S4, and outputs dynamic hydrodynamic performance parameters and flow field structure parameters; Iterative control module: Executes parameter feedback and iterative scheduling in step S5 until the preset simulation duration is completed; Results visualization module: Outputs turbine performance curves and flow field cloud maps.

[0029] Compared with the prior art, the present invention has the following beneficial effects: 1. High precision: Through three-level coordinate transformation of "global-follower-relative rotation" and strict kinematic correction, it accurately reflects the coupling effect of the platform's complex motion on the local flow field of the blade element, reduces the angle of attack calculation error and the dynamic load prediction deviation.

[0030] 2. Complete physical mechanism: The use of AI proxy model (LSTM / graph neural network) to replace static data table can effectively capture dynamic stall hysteresis effect and hydrodynamic history dependence, and greatly improve the prediction accuracy of dynamic load.

[0031] 3. Balancing efficiency and versatility: While maintaining near-high-fidelity CFD accuracy, the transient simulation time is shortened, making it suitable for engineering applications, parameter optimization, and long-term performance evaluation; the method is highly versatile and applicable to various types of floating turbines and offshore wind turbines. Attached Figure Description

[0032] Figure 1 A schematic diagram of the leaf element momentum theory; Figure 2 This is a schematic diagram of the volume force distribution using the method of the moving lines; Figure 3A schematic diagram of the bow roll motion (including coordinate transformation); Figure 4 This is a schematic diagram of the roll motion (including coordinate transformation). Figure 5 This is a schematic diagram of the pitching motion (including coordinate transformation). Figure 6 This is a schematic diagram of oscillation motion (including coordinate transformation). Figure 7 A schematic diagram of heave motion (including coordinate transformation); Figure 8 A schematic diagram of sway motion (including coordinate transformation); Figure 9 Diagram of the tail vortex structure of a stationary water turbine; Figure 10 This is a structural diagram of the tail vortex at the bow of a water turbine. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Example: This example uses a 500kW floating turbine as a simulation object and, in conjunction with the accompanying drawings, details the deployment and implementation process of the present invention.

[0035] The parameters in this embodiment include: turbine parameters: number of blades. n b =3, blade radius R=3m, leaf element discrete number Ns=2, local chord length c=0.8~1.2m (decreasing from leaf root to leaf tip); environmental parameters: incoming current velocity U0=2.5m / s, seawater density ρ=1025kg / m³ 3 The dynamic viscosity μ = 1.07 × 10⁻⁶ -6 Pa·s; Platform motion parameters: Maximum yaw angle ψ0 = 15°, maximum roll angle φ0 = 10°. 。 Maximum pitch angle θ0 = 8 。 The maximum displacements of the longitudinal sway, heave, and transverse sway are respectively s s0 =1.5m, s h0 =1.0m, s w0 =0.8m, all motion cycles T=10s.

[0036] Calculation parameters: CFD mesh size = 5 million (including turbine flow field and platform region), transient time step Δt = 0.01s, total simulation duration = 50s, including 5 motion cycles.

[0037] In this embodiment, the implementation steps of the AI-enhanced six-degree-of-freedom motion actuation line calculation method for floating turbines are as follows: Step 1: Model Initialization 1. Flow Field and Mesh Construction: Establish the flow field computational domain in ANSYS Fluent: Length along the incoming flow direction (X-axis) = 50R (150m), width laterally (Y-axis) = 30R (90m), height vertically (Z-axis) = 20R (60m). (Refer to...) Figure 3-8 The coordinate system range.

[0038] Structured grids were used to divide the area into regions affected by leaf elements ( Figure 2 The grid is finer in the momentum source region (grid size = 0.1R = 0.3m) and sparser in the far field region (grid size = 2R = 6m) to ensure a balance between computational accuracy and efficiency.

[0039] 2. Turbine parameter input: based on Figure 1 A schematic diagram of the blade element momentum theory is provided. Key blade element parameters are input into Fluent's UDF: blade element spanwise length dr = R / (Ns-1) = 3 / 24 ≈ 0.125 m; local pitch angle γ: blade root γ = 15°. 。 Leaf tip γ=5 。 The blade rotational angular velocity ω = π × n / 30 = π × 15 / 30 = 1.57 rad / s.

[0040] 3. Reference for setting initial conditions for platform movement Figure 3-8 The motion diagram is shown below, and the six-degree-of-freedom motion equations are defined in the UDF: Bow rocking: ξ yaw (t)=15 。 ×sin(2πt / 10), such as Figure 3 As shown in (a); Horizontal sway: ξ roll (t)=10 。 ×sin(2πt / 10), such as Figure 4 As shown in (a); Swaying: ξ pitch (t)=8 。 ×sin(2πt / 10), such as Figure 5 As shown in (a); Swaying / Hanging / Horizontal: Press s respectively surge (t) = 1.5 × sin(2πt / 10), s heave(t) = 1.0 × sin(2πt / 10), s sway (t) = 0.8 × sin(2πt / 10) is defined as follows ( Figure 6 (a)-8(a)).

[0041] At the initial moment (t=0), all motion parameters are 0 (displacement, angular velocity, and angular acceleration are all 0).

[0042] Step 2: Solver Setup 1. UDF program integration: Compile and integrate an actuator line UDF that combines "motion correction + AI prediction": Core function 1: Read real-time motion parameters from the platform and execute... Figure 3-8 Coordinate transformations (such as those called during bow roll) Figure 3 (b) Rotation matrix M yaw Core Function 2: Implementation Figure 2 Volume force mapping – via Gaussian distribution function f (r)=(1 / (πσ 2 ))exp(-rg² / σ 2 The leaf element hydrodynamics is mapped onto the flow field mesh, where the smoothing radius σ = 0.8 × dr ≈ 0.1 m.

[0043] 2. CFD solver parameter configuration for turbulence model: The SSTk-ω model is adopted (balancing the accuracy of rotating flow field and dynamic stall simulation). Solution format: The pressure-velocity coupling uses the SIMPLEC algorithm, and the time term uses a second-order implicit scheme (to ensure transient response accuracy). Boundary conditions: The inlet (X=-75m) is set as a velocity inlet (U0=2.5m / s), the outlet (X=75m) is set as a pressure outlet (atmospheric pressure), the wall is set as a no-slip boundary, and the turbine blade region is defined as the momentum source term region through UDF (Uniform Derivative Function). Figure 2 (the "source term application region").

[0044] Step 3: Motion parameter update and AI force calculation Within each time step (Δt=0.01s), perform the following operations via UDF: 1. Real-time calculation of platform motion parameters: Calculate the displacement at the current moment based on the motion equation.

[0045] 2. Blade element relative speed and angle of attack correction: Taking yaw and roll as examples, yaw correction ( Figure 3 (b) ): Reading Figure 3 (b) Real-time coordinates of the rotation center Oct. x c , y c ,z c The equations of motion from step 1 are used to calculate the equations of motion.

[0046] Calculate the vector of leaf element point and Oc. x v , y v , z v )=( x - x c , y - y c ,z- z c ); Call the rotation matrix M yaw ,Will( x v , y v , z v Convert to bow-and-roll R-coordinate system ( x r , y r , z r ); Calculate additional velocity , ; Correction of relative velocity .

[0047] Roll correction, such as Figure 4 (b): Similarly, based on Figure 4 The center of rotation O in (b) b In relation to coordinate transformation, call the rotation matrix M. roll Calculate the additional velocity V θroll Correcting relative velocity U rel,roll .

[0048] Finally, by superimposing the additional velocities of all degrees of freedom, we obtain the final true relative velocity U of the leaf element. rel,true With true angle of attack α true .

[0049] 3. AI Model Inference (LSTM Prediction of Hydrodynamic Coefficients) Feature Vector Construction: according to x t =[α true,t Re t V x,t V y,t V z,t ,ω roll,t ,ωpitch,t ,ω yaw,t ,φ t ,θ t ,ψ t M t Constructed from the dynamic stall characteristic timescale T stall =0.5s, the time series length is calculated to be N=50; ONNX model call: UDF loads a pre-trained LSTM model, inputs time series features, and outputs dynamic lift-drag coefficients. C L,AI , C D,AI Inference time ≈ 0.5ms / time step; Hydrodynamic calculation: based on dynamic lift L AI =(1 / 2)ρU rel,true 2 cC L,AI dr, dynamic resistance D AI =(1 / 2)ρU rel, true 2 cC D,AI dr; Calculate the drag force of the leaf element (corresponding to Figure 1 The "lift / drag arrow" is decomposed into axial force F. x Tangential force F θ .

[0050] Step 4: Iterative solution 1. CFD transient solution: The blade element hydrodynamics calculated in step 3 are loaded into the flow field mesh in the form of momentum source terms via UDF ( Figure 2 The volume force distribution region is determined by solving the unsteady Navier-Stokes equations to obtain the flow field parameters (velocity, pressure, vorticity) and turbine performance parameters (power coefficient C) at the current time step. P Thrust coefficient C T ).

[0051] 2. Motion Parameter Feedback Update: Extract the forces and moments exerted by the flow field on the floating platform (such as bow moment and roll moment), substitute them into the platform's motion equations, and update the displacement, angular velocity, and angular acceleration for the next time step, forming a closed loop of "motion → flow field → hydrodynamics → motion". Figure 10 The shape of the bow roll tail vortex changes in real time with the motion, and... Figure 9 The fixed wake vortex is compared to verify the effectiveness of the closed loop.

[0052] 3. Convergence Judgment: Monitor the residuals at each time step (pressure residual < 1e-4, velocity residual < 1e-5). If the residuals do not converge, adjust the mesh refinement area (e.g., optimization). Figure 2To ensure computational stability, the source term region grid or solver parameters (such as increasing the relaxation factor) can be adjusted.

[0053] Step 5: Post-processing of results 1. Performance parameter analysis UDF automatically outputs C P C T The time-history data (5000 data points within 50 seconds) was plotted using Origin software: the baseline group (fixed, corresponding to...) Figure 9 ):C P Mean = 0.42, fluctuation range < 5%; Experimental group (six degrees of freedom motion, corresponding to) Figure 10 ):C P Mean = 0.39, fluctuation range = 15% (and Figure 10 The consistent tail vortex distortion reflects the impact of motion on performance. Results Validation: AI Predicted C P The deviation from high-fidelity LES is less than 8%, which meets the engineering accuracy requirements.

[0054] 2. Flow field visualization analysis: Load the flow field results using Tecplot360 and plot the vorticity contour plot, such as... Figure 9 , Figure 10 : Figure 9 (Fixed type): The wake vortex exhibits a symmetrical shedding pattern, with a peak vortex volume of 50s. -1 , the length of the wake = 20R (60m); Figure 10 (Bowing motion): The wake vortex is deflected due to bow rolling, with peak vortex volume at 65s. -1 The wake length is 15R (45m) (consistent with the motion correction result in step 3, verifying the accuracy of the additional velocity calculation).

[0055] Drawing the leaf element angle of attack time history diagram: comparison Figure 1 The initial angle of attack α0 and the corrected α true , showing α true The fluctuation range has increased, which matches the dynamic stall phenomenon.

[0056] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for calculating the six-degree-of-freedom motion actuation lines of a floating turbine using AI enhancement, characterized in that, Includes the following steps: Step S1. Construct the basic actuation line model: Discretize the blades of the floating turbine along the spanwise direction into several blade elements. Based on the blade element momentum theory, calculate the basic incoming flow velocity, local initial angle of attack, and initial hydrodynamic force of each blade element in a fixed coordinate system. The initial hydrodynamics are initially calculated using the basic incoming flow velocity, initial angle of attack, and static airfoil data of the blade element. Step S2. Six-DOF motion coupling correction: Establish a three-level coordinate transformation system of "global coordinate system - body coordinate system - relative rotation coordinate system" to perform kinematic corrections on the yaw, roll, and pitch rotational degrees of freedom and the sway, heave, and roll translational degrees of freedom of the floating platform; calculate the additional velocities generated by each degree of freedom motion, including the tangential additional velocity and Coriolis acceleration of the rotational degree of freedom and the translational additional velocity of the translational degree of freedom, and synthesize the additional velocities with the basic incoming flow velocity vector of step S1 to update the true relative incoming flow velocity and true local angle of attack of the blade elements; Step S3. AI-enhanced hydrodynamic coefficient prediction: Based on the updated real local angle of attack, real relative inflow velocity, and real-time six-degree-of-freedom motion parameters of the floating platform in Step S2, a time-series feature vector is constructed; the time-series feature vector is input into a pre-trained deep learning proxy model, which outputs the dynamic lift coefficient and dynamic drag coefficient of the blade element, and calculates the corrected blade element hydrodynamics by combining the real relative inflow velocity; the deep learning proxy model can capture the dynamic stall hysteresis effect and hydrodynamic history dependence; Step S4. Volume force mapping and CFD transient solution: The blade element hydrodynamics corrected in step S3 are smoothly mapped to the flow field grid using a Gaussian distribution function to form momentum source terms; the momentum source terms are embedded in the CFD solver to solve the unsteady Navier-Stokes equations and obtain the dynamic hydrodynamic performance parameters and flow field structure parameters of the turbine. Step S5. Iterative Update: Feed back the flow field structure parameters output in step S4 to step S2, update the motion parameters of the floating platform, and repeat steps S2-S4 until the transient simulation of the preset duration is completed.

2. The method according to claim 1, characterized in that, In step S1, the specific process of calculation based on leaf element momentum theory includes: Step S11. Define key parameters for leaf elements: number of leaves n b Leaf element span dr Local string length c Local pitch angle γ, blade rotational angular velocity ω, and blade element spanwise distance from the blade root. r ; Step S12. Calculate the basic velocity components: axial velocity Ux = U 0+ u tangential velocity U θ = ωr - v Basic relative inflow velocity ;in, U 0 represents the incoming flow velocity. u To induce axial velocity, v To induce tangential velocity; Step S13. Calculate the initial angle of attack: Inflow angle φ0 = arctan( Ux / U θ The initial local angle of attack is α0 = φ0 - γ. Step S14. Calculate the initial hydrodynamic force: initial lift L 0 = (1 / 2) n b ρU rel,0 2 cC L,0 dr Initial resistance D 0 = (1 / 2) n b ρ U rel,0 2 cC D,0 dr ;in, ρ For fluid density, C L,0 , C D,0 The lift-drag coefficient is based on the static airfoil data table.

3. The method according to claim 1, characterized in that, In step S2, the specific process of kinematic correction for rotational degrees of freedom includes: Step S21. Define the equation of rotational motion: For any rotational degree of freedom, let the rotation angle be ξ(t) and the angular velocity be... Angular acceleration is , where t is time; Step S22. Determine the coordinates of the rotation center: Calculate the center point of the blade's rotation plane. O c Real-time coordinates in the global coordinate system x c , y c , z c Let R be the radius of rotation for this rotational degree of freedom. ξ ; Step S23. Coordinate Transformation: Using rotation matrix M ξ The original coordinates of the leaf element in the global coordinate system ( x , y Transform the coordinates (z) to a relative rotating coordinate system. x r , y r , z r ,Right now( x r , y r , z r )ᵀ=M ξ ·( x - x c , y - y c ,z- z c )ᵀ; Step S24. Calculate the additional velocity: tangential additional velocity Coriolis acceleration Where Rxy is the distance from the leaf element to the center of rotation in the XY plane of the relative rotating coordinate system; Step S25. Update velocity and angle of attack: Combine the tangential additional velocity with the basic tangential velocity, correct the axial velocity with Coriolis acceleration, and obtain the relative incoming flow velocity after rotational correction. U rel,ξ ;based on U rel,ξ Recalculate the inflow angle φ ξ Update the actual local angle of attack α ξ =φ ξ -γ; The specific process of kinematic correction for translational degrees of freedom includes: Step S26. Define the equation of translational motion: For any translational degree of freedom, let the translational displacement be s(t) and the velocity be... acceleration is ; Step S27. Calculate the additional velocity: Combine the translational velocity The relative incoming flow velocity after translational correction is obtained by vector superposition with the basic incoming flow velocity from step S1. U rel,s ; Step S28. Update angle of attack: based on U rel,s Recalculate the inflow angle φ based on the velocity direction. s Update the actual local angle of attack α s =φ s -γ.

4. The method according to claim 3, characterized in that, In step S2, the final determination of the true relative incoming flow velocity and the true local angle of attack is as follows: if both rotational and translational degrees of freedom exist simultaneously, the rotationally corrected... U rel,ξ With translational correction U rel,s Further vector synthesis yields the final true relative inflow velocity. U rel,true ;based on U rel,true Calculate the final inflow angle φ true Ultimately, the true local angle of attack α true =φ true -γ.

5. The method according to claim 44, characterized in that, In step S3, the deep learning agent model is a recurrent neural network or a graph neural network. 1) If it is a recurrent neural network, specifically a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU); the mathematical model of the LSTM includes: Input Gate: ; Forgotten Gate: ; Candidate cell status: ; Cell status update: ; Output gate: ; Hidden state output: ; Where σ is the sigmoid activation function, and ⊙ represents element-wise multiplication. W i , W f , W C , W o This is the weight matrix. b i , b f , b C , b o For bias vectors, h t-1 The hidden state of the previous time step. C t-1 This represents the cell state at the previous time step. x t This is the feature vector at the current time step; 2) If it is a graph neural network, all blade elements of the turbine are constructed as a graph structure, with blade elements as nodes of the graph and wake interference relationships between blade elements as edges of the graph. The neighborhood information of the nodes is aggregated through a message passing mechanism to collaboratively predict the dynamic lift coefficient and dynamic drag coefficient of each blade element.

6. The method according to claim 5, characterized in that, In step S3, the specific structure of the time series feature vector is as follows: x t =[α true,t Re t V x,t V y,t V z,t ,ω roll,t ,ω pitch,t ,ω yaw,t ,φ t ,θ t ,ψ t M t ], where: α true,t For step S2, update the final true local angle of attack at time step t; Re t Let Re be the Reynolds number at time step t; the formula is Re t =ρU rel,true,t c / μ, where μ is the hydrodynamic viscosity, U rel,true,t V represents the final true relative inflow velocity at time step t. x,t V y,t V z,t Let ω represent the linear velocities of the floating platform's center of mass in the x, y, and z directions in the global coordinate system at time step t; roll,t ,ω pitch,t ,ω yaw,t These represent the roll, pitch, and bow angular velocities of the floating platform at time step t; φ t ,θ t ,ψ t These represent the Euler angles of the floating platform at time step t: roll, pitch, and bow. t The time step t is the Mach number based on the tip velocity, calculated using the formula M. t =(ωR+U0) / a, where R is the blade radius and a is the speed of sound in the fluid; The time series is constructed as follows: Feature vectors from the past N time steps are selected to form the input sequence X = [ x t-N+1 , x t-N+2 ,..., x t-1 , x t The time series length N ranges from 20 to 100 time steps; the specific value of N is determined by the dynamic stall characteristic time scale T. stall Sure: Where Δt is the time step of the CFD transient simulation, This indicates rounding up to the nearest integer.

7. The method according to claim 6, characterized in that, In step S3, the specific process of pre-training the deep learning agent model includes: Step S31. Dataset Construction: Training data is generated using Large Eddy Simulation (LES). The LES model includes the geometry of the entire turbine blades and the rigid body model of the floating platform, with a mesh count ≥ 5 million and a time step Δt. LES =0.001~0.005s, simulation duration ≥20 floating platform motion cycles; extract the "time series feature vector + true lift-drag coefficient label" of each leaf element, total sample size M≥5×10 4 ; Step S32. Data preprocessing: Standardize the feature vectors using Z-score; divide the training set, validation set, and test set into a 7:2:1 ratio; Step S33. Training parameter configuration: The backpropagation time series algorithm BPTT is used, the optimizer is Adam, the initial learning rate is 1e-4, and the learning rate is adjusted using the cosine annealing strategy; Step S34. Loss Function and Regularization: The loss function is the mean squared error. ， Where B is the batch size. C L,AI , C D,AI To predict the lift-to-drag ratio, C L,true , C D,true Label the LES; employ early stopping strategy and dropout technique to prevent overfitting; Step S35. Model Evaluation: Test set evaluation metrics include Mean Absolute Error (MAE) ≤ 0.05 and Coefficient of Determination (R²). 2 ≥0.9, to ensure the model's generalization ability.

8. The method according to claim 1, characterized in that, In step S3, the specific formula for the corrected hydrodynamic calculation of the blade element is: Dynamic lift L AI =(1 / 2)ρU rel,true 2 cC L,AI dr, dynamic resistance D AI =(1 / 2)ρU rel,true 2 cC D,AI dr; will L AI With D AI Decomposed into axial force F x =L AI cosα true +D AI sinα true Tangential force F θ =L AI sinα true -D AI cosα true As the hydrodynamic input of the leaf element in step S4; The deployment method for the deep learning proxy model is as follows: the pre-trained model is converted to ONNX format, and a CFD solver is embedded through a user-defined function (UDF) interface; within each time step, the UDF reads the updated parameters from the CFD solver in step S2, constructs a time-series feature vector, and calls the ONNX model for inference output C. L,AI With C D,AI Inference time is <1ms / time step.

9. The method according to claim 8, characterized in that, In step S4, the specific form of the Gaussian distribution function is f(r) = (1 / (πσ)). 2 ))exp(-r g 2 / σ 2 ), where r g σ is the distance from the blade element center to the flow field grid node, and σ is the smoothing radius, which takes the value of 0.5 to 1.0 times the blade element spanwise length dr; dynamic hydrodynamic performance parameters include the turbine power coefficient C. P Thrust coefficient C T The flow field structure parameters include flow field velocity, pressure, and vorticity distribution; In step S5, the specific logic of the iterative update is as follows: after each CFD time step is completed, the force and torque of the flow field on the floating platform are extracted, the motion equation parameters (displacement, angular velocity, angular acceleration) of the platform are updated, and the updated motion parameters are input into step S2 to recalculate the additional velocity, so as to realize the dynamic coupling closed loop of "motion-flow field-hydrodynamics".

10. A computing system for performing the method according to any one of claims 1-9, characterized in that, include: Parameter input module: Receives turbine geometric parameters, environmental parameters, and initial motion parameters of the floating platform; Actuation line calculation module: performs the basic actuation line model construction in step S1 and the six-degree-of-freedom motion coupling correction in step S2, and outputs the true relative incoming flow velocity and the true local angle of attack; AI Inference Module: Loads a pre-trained ONNX format deep learning proxy model, performs feature construction and inference in step S3, and outputs the dynamic lift-drag coefficient and the corrected leaf element hydrodynamics. CFD Solver Module: Performs the volume force mapping and unsteady Navier-Stokes equation solution in step S4, and outputs dynamic hydrodynamic performance parameters and flow field structure parameters; Iterative control module: Executes parameter feedback and iterative scheduling in step S5 until the preset simulation duration is completed; Results visualization module: Outputs turbine performance curves and flow field cloud maps.

Images