Offshore floating wind turbine wake sparsity monitoring reconstruction method and system

Abstract

The present application relates to the technical field of sparse monitoring, and discloses a sparse monitoring reconstruction method and system for wake of offshore floating wind turbine, comprising: arranging flexible pressure measuring belts at key positions of the wind turbine to obtain dynamic wind pressure distribution data of the surface of the wind turbine blade, real-time rotating speed of the wind turbine and vibration acceleration data of the wind turbine platform, constructing a sparse monitoring data set based on the dynamic wind pressure distribution data, the real-time rotating speed and the vibration acceleration data; uniformly fusing the sparse monitoring data set, inputting the processed data into a trained neural network model to reconstruct a complete wake velocity field distribution; dynamically correcting the reconstructed wake field in combination with platform motion state data, and outputting a wake evolution prediction result with spatiotemporal continuity.

Description

Technical Field

[0001] This invention relates to the field of sparse monitoring technology, and in particular to a method and system for sparse monitoring and reconstruction of the wake of a floating offshore wind turbine. Background Technology

[0002] In the field of offshore floating wind power generation, the accurate monitoring and reconstruction of wake effects has always been a key technical bottleneck restricting the overall performance optimization of wind farms. Traditional wake monitoring methods mainly rely on high-density sensor arrays or lidar systems. These methods are not only costly to deploy, but also face practical problems such as decreased reliability and maintenance difficulties in harsh marine environments. In particular, for floating platforms with six degrees of freedom of motion, their dynamic characteristics make the wake field exhibit more complex spatiotemporal evolution features, further increasing the difficulty of traditional monitoring methods.

[0003] In recent years, flow field reconstruction technology based on sparse monitoring has demonstrated significant advantages. By utilizing an unlimited number of sensors at key locations, combined with advanced signal processing algorithms, accurate capture of flow field characteristics can be achieved while significantly reducing system complexity and cost. Particularly in this technological approach, the application of diffusion models has brought breakthrough progress to flow field reconstruction. Compared with traditional neural networks, diffusion models, through a progressive data generation process, can better maintain the physical consistency of the flow field and accurately reproduce the multi-scale turbulent structure of the wake field. This generative method based on a probabilistic framework is more robust to the sparsity of input data, making it particularly suitable for applications where sensor deployment is limited, such as offshore wind turbines.

[0004] Currently, combining the convenience of sparse monitoring with the powerful generative capabilities of diffusion models to develop a real-time wake reconstruction method suitable for offshore floating wind turbines has become a key technical challenge for improving the operational efficiency of wind farms. This research direction not only has significant theoretical value but also provides practical engineering solutions for the intelligent operation and maintenance of offshore wind power. Summary of the Invention

[0005] This invention provides a method and system for sparse monitoring and reconstruction of the wake of a floating offshore wind turbine, in order to solve the problem of poor robustness of existing sparse monitoring and reconstruction methods.

[0006] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a method for monitoring and reconstructing the wake sparseness of a floating offshore wind turbine, comprising: S1. Flexible pressure measuring belts are deployed at key locations of the wind turbine to obtain dynamic wind pressure distribution data on the surface of the wind turbine blades, real-time rotational speed of the wind turbine, and vibration acceleration data of the wind turbine platform in real time. A sparse monitoring dataset is constructed based on the dynamic wind pressure distribution data, the real-time rotational speed, and the vibration acceleration data. S2. Perform unified fusion processing on the sparse monitoring dataset, and input the processed data into the trained neural network model to reconstruct the complete wake velocity field distribution; S3. Combine the platform motion state data to dynamically correct the reconstructed wake field and output wake evolution prediction results with spatiotemporal continuity.

[0007] Optionally, the key locations mentioned in step S1 include the leading edge, trailing edge, and different cross sections in the spanwise direction of the blade. In the leading edge region, multiple rows of pressure measuring holes are arranged along the chord, and pressure measuring holes are arranged in an alternating pattern in the trailing edge region. Pressure measuring bands are arranged in the spanwise direction at three characteristic cross sections: the root, middle, and tip of the blade.

[0008] Optionally, the unification and fusion process in step S2 includes: Interpolation is used to unify sensor data from different frequencies to the same sampling time point; The differences in physical dimensions are eliminated by using the maximum-minimum normalization or Z-score normalization methods. The normalized multidimensional data is integrated into a unified input matrix according to the time series and divided into sliding windows of fixed length.

[0009] Optionally, the neural network model described in step S2 is one or more combinations of a diffusion model, a convolutional neural network, a recurrent neural network, a long short-term memory network, or a Transformer.

[0010] Optionally, the training data for the neural network model in step S2 includes simulation data or experimentally measured wake field data, and real wake field data labeled as PIV or CFD. Mean square error, correlation coefficient or vorticity error are used as verification indicators during the training process.

[0011] Optionally, the dynamic correction described in step S3 involves spatially transforming or compensating the reconstructed wake field by fusing the platform's six-degree-of-freedom motion data, in order to reflect the influence of the platform's motion on the wake field.

[0012] In a second aspect, embodiments of this application provide a marine floating wind turbine wake sparse monitoring and reconstruction system for implementing the method described in any one of the first aspects, comprising: Flexible pressure measuring strips, pressure sensors, speed sensors, and six-axis accelerometers are arranged on the surface of the blades; Multi-physical quantity synchronous measurement and data acquisition system; Data processing and fusion module; Neural network inference module; Wake field visualization and output module.

[0013] Beneficial effects: The sparse monitoring and reconstruction method for wakes of offshore floating wind turbines provided by this invention sparsely deploys flexible pressure gauges and a limited number of motion sensors only at key locations on the blades (such as the leading edge, trailing edge, and specific spanwise sections), replacing traditional high-density sensor arrays or expensive lidar systems. This greatly simplifies the structure of the monitoring system and reduces hardware, installation, and maintenance costs, making it particularly suitable for applications on offshore floating platforms in harsh environments with poor accessibility. By integrating physical mechanisms (such as the platform's six-degree-of-freedom motion) with data-driven algorithms (such as diffusion models), the constructed intelligent reconstruction model can fully utilize the physical characteristics in the sparse monitoring data. This method can not only accurately reconstruct the complete wake velocity field at the current moment but also predict the dynamic evolution of the wake by combining time-series data. It has high reconstruction accuracy and good spatiotemporal continuity, providing a reliable data foundation for real-time optimization control of wind farms. By adopting a unified data fusion process (including time alignment, normalization, and sliding window processing) and a lightweight neural network model, the entire system operates efficiently from data acquisition and processing to wake field reconstruction and prediction, meeting the stringent real-time requirements of engineering practice. Attached Figure Description

[0014] Figure 1 A flowchart of a preferred embodiment of the offshore floating wind turbine wake sparse monitoring and reconstruction method of the present invention; Figure 2 This is a schematic diagram of the wind turbine measurement system according to a preferred embodiment of the present invention; Figure 3 This is a schematic diagram of the neural network structure provided in a preferred embodiment of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0016] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0017] Please see Figure 1-3 This application provides a method for monitoring and reconstructing the sparse wake of a floating offshore wind turbine, including: S1. Flexible pressure measuring belts are deployed at key locations of the wind turbine to obtain dynamic wind pressure distribution data on the surface of the wind turbine blades, real-time rotational speed of the wind turbine, and vibration acceleration data of the wind turbine platform in real time. A sparse monitoring dataset is constructed based on the dynamic wind pressure distribution data, the real-time rotational speed, and the vibration acceleration data. S2. Perform unified fusion processing on the sparse monitoring dataset, and input the processed data into the trained neural network model to reconstruct the complete wake velocity field distribution; S3. Combine the platform motion state data to dynamically correct the reconstructed wake field and output wake evolution prediction results with spatiotemporal continuity.

[0018] Optionally, the key locations mentioned in step S1 include the leading edge, trailing edge, and different cross sections in the spanwise direction of the blade. In the leading edge region, multiple rows of pressure measuring holes are arranged along the chord, and pressure measuring holes are arranged in an alternating pattern in the trailing edge region. Pressure measuring bands are arranged in the spanwise direction at three characteristic cross sections: the root, middle, and tip of the blade.

[0019] In the above embodiments, flexible pressure measurement strips are deployed at key locations on the wind turbine blade surface (including the leading edge, trailing edge, and different cross-sections along the span). Pressure signals are transmitted to pressure sensors through pressure measurement holes and air guide hoses to acquire real-time dynamic wind pressure distribution data on the blade surface. Simultaneously, a speed sensor measures the real-time wind turbine speed, and a six-axis accelerometer collects six-degree-of-freedom vibration acceleration data of the wind turbine platform under wave loads. All sensor data are synchronously acquired and preprocessed through a high-speed data acquisition system to form a multi-source heterogeneous sparse monitoring dataset. The measurement system is as follows: Figure 2 As shown.

[0020] (1) A layered arrangement strategy is adopted on the surface of the wind turbine blade. Three rows of pressure measuring holes are arranged along the chord in the leading edge region, with a spacing of 15% of the chord length; two rows of staggered pressure measuring holes are set in the trailing edge region, covering the trailing 20% ​​of the chord length of the blade; pressure measuring bands are arranged in the spanwise direction at three characteristic sections: the root, middle and tip of the blade. Each pressure measuring hole is connected to a miniature pressure sensor module through a 2mm diameter PTFE air guide hose, and a dedicated wire groove is arranged inside the blade to protect the signal transmission line.

[0021] (2) A multi-physical quantity synchronous measurement system is adopted, including a dynamic pressure measurement subsystem and a motion state monitoring subsystem, to measure the spindle speed and monitor the platform motion. An environmental parameter compensation module is used for real-time environmental compensation of the measurement data.

[0022] (3) Next, high-speed data acquisition and synchronization control are performed. A distributed acquisition network is constructed, with each blade configured with a local data acquisition node to achieve multi-channel synchronous triggering. The data is preprocessed simultaneously, and online moving average filtering (outlier detection and removal algorithm) and dimension normalization are performed.

[0023] (4) Finally, multi-source data fusion and feature extraction are performed. A data buffer with a unified timestamp is established, and Kalman filtering is used to achieve spatiotemporal alignment of multi-sensor data. The time-domain features (mean, variance) and frequency-domain features (FFT main frequency) of the pressure signal are extracted, the Euler angles and motion trajectory of the platform motion parameters are calculated, and a dimensionless feature parameter matrix is ​​generated to obtain a monitoring dataset with quality label.

[0024] This four-step implementation plan, through precise sensor arrangement, simultaneous measurement of multiple physical quantities, high-precision data acquisition, and intelligent data fusion, ensures the reliability and completeness of sparse monitoring data, providing a high-quality input data foundation for subsequent wake field reconstruction.

[0025] Optionally, the unification and fusion process in step S2 includes: Interpolation is used to unify sensor data from different frequencies to the same sampling time point; The differences in physical dimensions are eliminated by using the maximum-minimum normalization or Z-score normalization methods. The normalized multidimensional data is integrated into a unified input matrix according to the time series and divided into sliding windows of fixed length.

[0026] Optionally, the neural network model mentioned in step S2 is one or more combinations of a diffusion model, a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory network (LSTM), or a Transformer.

[0027] Optionally, the training data for the neural network model in step S2 includes simulation data or experimentally measured wake field data, and real wake field data labeled as PIV or CFD. Mean square error, correlation coefficient or vorticity error are used as verification indicators during the training process.

[0028] In the above embodiment, sensors arranged on the wind turbine blades monitor the pressure (P_x, P_y, P_z) and acceleration (A_x, A_y, A_z) in the X, Y, and Z directions, respectively. The input to neural networks (such as CNN, LSTM, or Transformer) typically requires a structured data format. Since different sensors have different acquisition frequencies, the raw data needs to be standardized. The specific steps are as follows: (1) Use linear interpolation or spline interpolation to unify sensor data of different frequencies to the same sampling time point to ensure data timestamp alignment. Set a uniform sampling frequency (e.g., 100Hz), downsample high-frequency data and upsample low-frequency data to ensure consistent data length.

[0029] (2) To eliminate the influence of different physical dimensions, the Min-MaxNormalization or Z-score normalization method is used to map the data to a uniform interval (such as [0,1] or standard normal distribution).

[0030] (3) The normalized multi-dimensional data (pressure, acceleration) are integrated into a unified input matrix according to the time series, and the continuous time series is divided into sliding windows of fixed length (e.g., each window contains 50 time points) to extract local dynamic features. By adding Gaussian noise or time offset, the training dataset is expanded to improve the model's generalization ability.

[0031] Optionally, the dynamic correction described in step S3 involves spatially transforming or compensating the reconstructed wake field by fusing the platform's six-degree-of-freedom motion data, in order to reflect the influence of the platform's motion on the wake field.

[0032] In the above embodiments, real-time collected sparse monitoring data is input into a trained neural network model, and the complete wake velocity field distribution is reconstructed through a condition generation process. The model employs a lightweight network architecture and parallel computing technology to ensure real-time computing capabilities on embedded devices. Simultaneously, the wake field is dynamically corrected by combining platform motion state data, outputting a wake evolution prediction result with spatiotemporal continuity, providing real-time data support for wind farm optimization control. Given sensor data at historical time steps (such as unified data D, grid point coordinates XYZ, time series t, etc.), the velocity components U, V, W and pressure P of the fluid wake field are directly predicted. The core idea of ​​constructing an end-to-end neural network model is as follows: (1) Choose a suitable neural network model, such as recurrent neural networks and convolutional neural networks.

[0033] (2) Set the structure according to the selected neural network and design key details such as input normalization and multi-task learning.

[0034] (3) Paired time-series input-wake-output datasets (simulation or experimental data) are required. Supervised training should be performed simultaneously, using real wake fields (such as PIV measurements or CFD results) as labels, and overfitting should be controlled. Finally, validation metrics should be selected, such as mean square error, correlation coefficient (R²), vorticity error, etc. A schematic diagram of the neural network is shown below. Figure 3 As shown.

[0035] This application also provides a system for monitoring and reconstructing the wake sparseness of a floating offshore wind turbine, used to implement a method for monitoring and reconstructing the wake sparseness of a floating offshore wind turbine, comprising: Flexible pressure measuring strips, pressure sensors, speed sensors, and six-axis accelerometers are arranged on the surface of the blades; Multi-physical quantity synchronous measurement and data acquisition system; Data processing and fusion module; Neural network inference module; Wake field visualization and output module.

[0036] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for monitoring and reconstructing the sparse wake of a floating offshore wind turbine, characterized in that, include: S1. Flexible pressure measurement strips are deployed at key locations on the wind turbine to acquire dynamic wind pressure distribution data on the surface of the wind turbine blades, real-time rotational speed of the wind turbine, and vibration acceleration data of the wind turbine platform in real time. A sparse monitoring dataset is constructed based on the dynamic wind pressure distribution data, the real-time rotational speed, and the vibration acceleration data. The key locations include the leading edge, trailing edge, and different cross sections along the spanwise direction of the blade. Multiple rows of pressure measurement holes are arranged along the chordwise direction in the leading edge region, and staggered pressure measurement holes are set in the trailing edge region. Pressure measurement strips are arranged along the spanwise direction at three characteristic cross sections: the root, middle, and tip of the blade. S2. Perform unified fusion processing on the sparse monitoring dataset, and input the processed unified data D, the grid point coordinates XYZ and the time sequence t into the trained neural network model to reconstruct the complete wake velocity field distribution and predict the velocity components U, V, W and pressure P of the fluid wake field. S3. Combine the platform motion state data to dynamically correct the reconstructed wake field and output a wake evolution prediction result with spatiotemporal continuity; the dynamic correction is to perform spatial transformation or motion compensation on the reconstructed wake field by fusing the platform's six degrees of freedom motion data to reflect the influence of the platform motion on the wake field.

2. The method for monitoring and reconstructing the wake sparseness of a floating offshore wind turbine according to claim 1, characterized in that, The unification and fusion process in step S2 includes: Interpolation is used to unify sensor data from different frequencies to the same sampling time point; The differences in physical dimensions are eliminated by using the maximum-minimum normalization or Z-score normalization methods. The normalized multidimensional data is integrated into a unified input matrix according to the time series and divided into sliding windows of fixed length.

3. The method for monitoring and reconstructing the wake sparseness of a floating offshore wind turbine according to claim 1, characterized in that, The neural network model mentioned in step S2 is one or more combinations of diffusion model, convolutional neural network, recurrent neural network, long short-term memory network or Transformer.

4. The method for monitoring and reconstructing the wake sparseness of a floating offshore wind turbine according to claim 1, characterized in that, The training data for the neural network model in step S2 includes simulation data or experimentally measured wake field data, and real wake field data labeled as PIV or CFD. During the training process, mean square error, correlation coefficient or vorticity error are used as verification indicators.

5. A marine floating wind turbine wake sparse monitoring and reconstruction system for implementing the method of any one of claims 1 to 4, characterized in that, include: Flexible pressure measuring strips, pressure sensors, speed sensors, and six-axis accelerometers are arranged on the surface of the blades; Multi-physical quantity synchronous measurement and data acquisition system; Data processing and fusion module; Neural network inference module; Wake field visualization and output module.

Images