Sea floating type wind turbine generator wave data acquisition device, monitoring system and method

Abstract

The invention discloses an offshore floating type wind turbine generator wave data acquisition device, a monitoring system and a monitoring method, and aims to solve the problems of non-contact wave sensor measurement deviation caused by floating type platform movement and fixed window statistics lag in a traditional wave monitoring method. Compensating six-degree-of-freedom motion interference in real time by using inertial navigation data; a self-adaptive time sliding window mechanism based on a wave height variable coefficient is provided, and the statistical duration and the sampling frequency are dynamically adjusted; and fusing prediction results of a plurality of prediction models, and outputting significant wave height, period, wave direction and 95% confidence interval through triple verification of an error threshold value, an extreme value range and physical consistency. According to the method, the accuracy, the real-time performance and the robustness of wave monitoring can be remarkably improved, and reliable data support is provided for safe operation of a unit, load reduction control, operation and maintenance decision making and hydrological environment analysis.

Description

Technical Field

[0001] This invention relates to the technical field of data monitoring, and in particular to a wave data acquisition device, monitoring system and method for offshore floating wind turbines. Background Technology

[0002] With the rapid development of deep-sea wind power, floating offshore wind turbines have become an important technological route for developing deep-sea wind energy resources (floating offshore wind turbines include a floating platform and wind turbines installed on the floating platform; the wind turbines include blades, hubs, pitch systems, nacelles, and towers, with a tower base guardrail platform on the tower). However, floating platforms are prone to significant movement in complex marine environments, which may lead to risks such as mooring cable slack, structural fatigue, and even system instability. Furthermore, the operation, maintenance, installation, and repair of floating offshore wind turbines are highly dependent on real-time sea conditions, especially wave parameters. The lack of accurate and real-time wave monitoring methods will not only significantly affect operation and maintenance efficiency but may also jeopardize operational safety.

[0003] Traditional weather forecasts are typically based on average data from large sea areas, making it difficult to accurately reflect the wave characteristics (such as wave height, period, and direction) of the local waters around wind farms. This often forces maintenance vessels to turn back after arriving at the site because actual sea conditions exceed the operational window, resulting in ineffective voyages and increased economic costs. Traditional contact wave monitoring equipment (such as wave buoys) is susceptible to seawater corrosion and biofouling, and carries risks such as capsizing and cable breakage, making it unsuitable for long-term stable monitoring of floating platforms. Furthermore, contact sensors installed on floating platforms provide a mixed result of the platform's own motion and the relative motion caused by waves, failing to accurately reflect the true characteristics of the waves.

[0004] Non-contact wave sensors are primarily based on radar ranging principles, transmitting signals to the sea surface and receiving echoes to retrieve wave elevation information. This technology can perform non-intrusive detection of the sea surface at a certain distance from a floating platform, directly acquiring the undisturbed original wave field in front of the platform and providing clean wave input data. However, existing technologies still face the following challenges: 1. Non-contact wave sensors are not easy to install, pose a high risk of working at heights, threaten the safety of maintenance personnel, cause long downtime due to equipment failure, and make it difficult to flexibly adjust the clearance distance.

[0005] 2. Offshore floating wind turbines have unique floating characteristics, and under the action of waves, they will produce six degrees of freedom of motion, including heave, pitch, and roll. Non-contact wave sensors are usually fixed to the platform of the offshore floating wind turbine and move synchronously with it, which makes their measurement accuracy susceptible to interference from the platform's movement, making it difficult to obtain the true parameters of the waves.

[0006] 3. Traditional wave monitoring relies on historical data collected over a certain period to output wave parameters, which inherently introduces a time delay. Offshore floating wind turbines need to adjust yaw and pitch maneuvers based on real-time wave conditions; delayed statistical values ​​can cause control commands to become disconnected from actual wave conditions. In the face of extreme and sudden waves, delayed statistical results cannot provide timely warnings, potentially missing the optimal opportunity for risk avoidance. Furthermore, the choice of historical data duration directly impacts the accuracy and reliability of the monitoring results. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a wave data acquisition device, monitoring system and method for offshore floating wind turbines.

[0008] To achieve the above objectives, the technical solution provided by this invention is as follows: A wave data acquisition device for offshore floating wind turbines includes a fixed support, a tilting support, a non-contact wave sensor with an integrated inertial navigation module, and sensor mounting components. The fixed bracket is fixed to the guardrail on the tower base guardrail platform; The flip-up bracket is rotatably connected to the fixed bracket; The sensor mounting bracket is installed on the flip-up bracket; The non-contact wave sensor with integrated inertial navigation module is mounted on the sensor fixture and arranged in an equilateral triangle. Its position is adjusted by the flip bracket.

[0009] Furthermore, the flipping support includes a flipping main frame, a telescopic frame, and a telescopic support arm assembly; The main frame for flipping is provided with a sliding groove; The telescopic frame is installed inside the slide groove and extends and retracts along the extension direction of the slide groove; The telescopic support arm assembly is connected between the tilting main frame and the telescopic frame.

[0010] Furthermore, to achieve the above objectives, the present invention also provides a wave data monitoring system for offshore floating wind turbines, which includes the aforementioned wave data acquisition device for offshore floating wind turbines, a data processing module, and a monitoring module. The data processing module processes the data collected by the wave data acquisition device of the offshore floating wind turbine and transmits the processed data to the monitoring module, which then monitors the ocean wave data. The data processing module includes a motion compensation unit, a data filtering unit, an adaptive statistics unit, a prediction unit, a fusion verification unit, and a data output unit. The motion compensation unit uses the Euler angle transformation method to transform the observation values ​​of the non-contact wave sensor in the carrier coordinate system to the global coordinate system, thereby eliminating the coupling interference of the six-degree-of-freedom motion of the floating platform on the wave measurement and obtaining the real wave data in the global coordinate system. The data filtering unit is used to remove local anomalies caused by sensor noise, sea surface stray echoes, or compensation residuals. The adaptive statistical unit is used to calculate the wave height variation coefficient CV and dynamically adjust the sliding window parameters; The prediction unit is used to train a prediction model based on historical wave data and output future wave prediction data through the trained prediction model. The fusion verification unit is used to perform consistency verification of statistical results and prediction results under multiple constraints. The data output unit is used to output real-time wave parameters, statistical feature reports, and abnormal alarm information; The monitoring module includes a remote wind farm service center and a local wind turbine control system.

[0011] Furthermore, to achieve the above objectives, the present invention also provides a wave data monitoring method for offshore floating wind turbines, which is implemented using the aforementioned wave data monitoring system for offshore floating wind turbines, including: S1. Collect raw wave observation data containing motion disturbances of the floating platform and six-degree-of-freedom pose data of the floating platform through the wave data acquisition device of the offshore floating wind turbine. S2. By using the Euler angle transformation method through the motion compensation unit, the observation values ​​of the non-contact wave sensor in the carrier coordinate system are transformed to the global coordinate system, thereby eliminating the coupling interference of the six-degree-of-freedom motion of the floating platform on the wave measurement and obtaining the real wave data in the global coordinate system. S3. The data filtering unit removes local anomalies caused by sensor noise, instantaneous interference from wave breakage, or compensation residuals from the real wave data to obtain purified wave data. S4. For the purified wave data, the wave height variation coefficient CV is calculated by adaptive statistical unit and the sliding window parameters are dynamically adjusted to extract historical wave features. S5. Train a prediction model based on historical wave characteristics using prediction units and predict future wave prediction data. S6. Perform a multi-constraint consistency check on the statistical results obtained in step S4 and the prediction results obtained in step S5 using the fusion verification unit. S7. Output real-time wave parameters, statistical characteristic reports, and abnormal alarm information through the data output unit; S8. Monitor ocean wave data through the monitoring module based on real-time wave parameters, statistical feature reports, and abnormal alarm information.

[0012] Further, step S2 includes: S2-1. Construct the rotation matrix from the carrier coordinate system to the global coordinate system. :

[0013] in, Indicates the roll angle of a floating platform. Indicates the pitch angle of a floating platform. Indicates the heading angle of the floating platform; S2-2, Combining rotation matrix Calculate the target position in the global coordinate system :

[0014] in, For the global location of the carrier, This represents the sensor's position offset in the carrier coordinate system. These are the raw measurements from the non-contact wave sensor. The displacement change of a non-contact wave sensor caused by the movement of a floating platform; S2-3. For the six-degree-of-freedom motion of the floating platform, the true wave data in the global coordinate system is obtained through inverse calculation compensation:

[0015]

[0016]

[0017] in, To the compensated effective wave height, The wave height is the original wave height obtained from observation. The heave of the floating platform is based on the target location. get; For the compensated wave period, The wave period obtained from the original observation; For the compensated wave direction, The wave direction obtained from the original observation. This is the initial heading angle.

[0018] Furthermore, step S3 filters using the following filtering mechanism: S3-1. The compensated wave height data is decomposed using the db4 wavelet basis into low-frequency signals corresponding to the real wave components and high-frequency signals corresponding to the noise components. S3-2, Using an adaptive threshold function Thresholding is performed on high-frequency coefficients to retain effective signal components; sensor electronic noise and transient interference caused by wave breakage are removed. The standard deviation of noise. For data length; S3-3. Calculate the mean of the denoised data sequence. and noise standard deviation ; S3-4. Based on the characteristics of normal distribution, […] Data points within a certain range are marked as outliers; S3-5. If the number of consecutive abnormal points does not exceed 5, linear interpolation is used for replacement; if the number of consecutive abnormal data exceeds 5, the data segment is determined to be subject to strong interference or sensor failure, and is marked as an invalid data segment. It will be automatically skipped in subsequent statistical analysis and feature extraction and will not be included in the calculation. S3-6, Output the purified wave data.

[0019] Further, step S4 includes: The wave height variation coefficient is continuously calculated for the most recent three sliding windows. :

[0020] in, The standard deviation of wave height. The mean value of the wave height; If CV > 0.3: the wave pattern is considered violent, and the wave history data statistics duration is switched to 5-8 minutes with a sampling frequency of 6-10Hz to capture high-frequency dynamic characteristics; if 0.2 < CV ≤ 0.3: the wave history data statistics duration is switched to 15-20 minutes with a sampling frequency of 3-5Hz to reflect stationary statistical characteristics; if CV ≤ 0.2: the wave history data statistics duration is switched to 20-30 minutes with a sampling frequency of 3-5Hz. Furthermore, the statistical duration and sampling frequency of historical wave data are optimized by combining the wave type of the sea area and the operating scenario of the generator unit: For sea areas dominated by wind and waves: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 15-20 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 5-8 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 20-25 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. For sea areas dominated by swells: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 18-20 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 8-10 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 25-30 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. For sea areas dominated by mixed waves: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 16-18 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 6-9 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 22-28 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. After optimization and adjustment, extract historical wave features: Significant wave height Take the arithmetic mean of the heights of 1 / 3 of the large wave within the window:

[0021] in, All are waves within the window. n is the total wave number within the window; maximum wave height : Maximum wave height within the window; Wave height standard deviation : Reflects the degree of dispersion in wave height distribution:

[0022] For the first in the window A wave, The arithmetic mean of the wave heights within the window; Peak period The wave data is converted to the frequency domain using Fourier transform, and the period corresponding to the frequency of maximum energy is extracted.

[0023] The peak frequency of the energy spectrum; Average period Arithmetic mean of all wave periods within the window:

[0024] For the first The period of a wave; Periodic variation coefficient :

[0025] The periodic standard deviation; Main wave direction The mode of the wave direction distribution within the statistical window, i.e., the wave direction that appears most frequently; Standard deviation of wave direction distribution : Reflects the degree of wave concentration The smaller the value, the more stable the wave direction.

[0026] Furthermore, the adaptive statistical unit satisfies the following constraints when performing data statistics: Minimum sample size: A single set of statistical data must contain at least 30 complete wave cycles; Recognition accuracy: After adjustment, the peak and trough recognition accuracy of the zero-point method is ≥90%; System synchronization: The statistical period matches the data update frequency of the wind turbine control system.

[0027] Furthermore, in step S5, different prediction models are selected for prediction based on different time series prediction needs; in, For forecasts with a duration of 30 minutes or more and a period of more than 1 hour, the TFT forecast model is selected. For the prediction of nonlinear complex fluctuations with a short duration window of 5-10 minutes, where the nonlinear complex fluctuations are defined as wave height variation coefficients CV>0.3, the TCN prediction model is selected. For stationary data, i.e., data with wave height variation coefficient CV ≤ 0.2, the N-BEATS prediction model is selected, and basis function decomposition is performed through deep learning to achieve highly automated prediction of stationary sequences.

[0028] Further, step S6 includes: S6-1. Associate the historical wave features of the latest window with the predicted data to construct a continuous "history-future" feature sequence with a dimension of 1×(M×2), where M is the number of features in a single window; S6-2. Calculate the deviation between the predicted value and the average value of the historical wave characteristics of the last 3 windows to verify the error threshold. If the following conditions are met: wave height deviation ≤ 20%, period deviation ≤ 15%, wave direction deviation ≤ 10, then proceed to step S6-3. Otherwise, mark it as an error exceeding the standard and return to step S4. S6-3. Calculate the standard deviation of the multi-model prediction results for consistency verification. If the standard deviation is ≤5%, the prediction is considered stable; otherwise, it is marked as unstable and the process returns to step S4.

[0029] Compared with existing technologies, the principles and advantages of this technical solution are as follows: 1. Adopting a non-contact monitoring method, the wave data acquisition device of the offshore floating wind turbine equipped with a non-contact wave sensor is installed on the tower base guardrail platform. It does not need to come into contact with seawater. Compared with traditional contact equipment, it can avoid problems such as seawater corrosion, biological adhesion, cable breakage, and equipment overturning that are prone to occur when in contact with seawater, fundamentally avoiding direct damage to the equipment caused by the marine environment.

[0030] The non-contact wave sensor with integrated inertial navigation module is mounted on the sensor fixture and arranged in an equilateral triangle for easy wave direction calculation.

[0031] Furthermore, in the wave data acquisition device for offshore floating wind turbines, a flip-up bracket is added between the fixed bracket and the sensor fixing component. When it is necessary to install and maintain the non-contact wave sensor, simply flip the bracket to the inside of the guardrail (the non-contact wave sensor moves to the inside of the guardrail accordingly). When the installation and maintenance of the non-contact wave sensor is completed, simply flip the bracket back to the outside of the guardrail (the non-contact wave sensor returns to its original acquisition position). This method is convenient to install, avoids the risks associated with high-altitude installation, reduces the safety risks for maintenance personnel, reduces equipment downtime, and lowers the equipment installation, maintenance, and total lifecycle costs.

[0032] 2. The telescopic frame is installed in the slide groove of the tilting main frame and extends and retracts along the extension direction of the slide groove, which makes it easy to adjust the distance of the non-contact wave sensor moving outward. Moreover, the telescopic support arm assembly is connected between the tilting main frame and the telescopic frame, which not only enhances the stability of the wave data acquisition device structure of the offshore floating wind turbine, but also allows for flexible adjustment of the clearance distance.

[0033] 3. The inertial navigation module integrated into the non-contact wave sensor possesses high dynamic response and overload resistance, enabling motion compensation and collaborative sensing. This solves the measurement challenge of coupling the floating platform's own motion with wave motion, effectively isolating platform motion interference from wave monitoring, restoring true wave characteristic parameters, and improving data accuracy and stability. Simultaneously, it achieves collaborative monitoring of "waves and platform," providing comprehensive data support for unit structural design, safety management, attitude control, and intelligent operation and maintenance. It also enhances sensor installation flexibility and reduces system deployment costs.

[0034] 4. Compared to the lag and accuracy degradation issues that easily occur in traditional fixed-time-window monitoring, the adaptive statistical unit employs an adaptive time-sliding method. Based on the wave height variation coefficient, combined with sea wave patterns, unit operating scenarios, and sampling frequency, the window duration is dynamically adjusted. This balances data representativeness and real-time performance during normal monitoring, rapidly captures extreme values ​​during extreme weather warnings, and reduces fluctuation interference during maintenance assessments. It improves the accuracy of wave height measurement and the reliability of the zero-point identification method, while also ensuring real-time response and data integrity. A wave feature database is constructed to solve the challenge of balancing real-time monitoring and data validity under complex wave conditions.

[0035] 5. By employing a method of "sliding window statistics + time-series prediction + error feedback correction," the system adapts to the duration requirements of adaptive time sliding windows, addressing both the statistical lag issue of traditional monitoring and improving the accuracy of prediction results. It combines multiple prediction models adapted to different conditions, using continuous historical window statistical feature values ​​as input and outputting predicted values ​​consistent with the window duration. Simultaneously, error feedback dynamically adjusts model parameters and supplements the training set, continuously optimizing prediction accuracy. This ensures the timeliness of statistical data, preventing control commands from deviating from actual wave conditions. It provides timely warnings in the face of extreme and sudden waves, buying time for turbine risk avoidance. By matching the different feature requirements of long and short windows and ensuring accurate timestamp synchronization, it further enhances the reliability of wave monitoring. This provides accurate data support for real-time control of turbine yaw and pitch, and helps optimize the monitoring system's adaptability to complex wave conditions. It provides high-quality data assurance for the safe operation, risk warning, and maintenance decisions of floating wind turbines, effectively compensating for insufficient monitoring accuracy caused by inappropriate selection of historical data duration. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the services required in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1This is a schematic diagram of the wave data acquisition device for offshore floating wind turbines installed on the tower base guardrail platform according to an embodiment of the present invention (A is the wave data acquisition device for offshore floating wind turbines). Figure 2 This is a structural diagram of the wave data acquisition device for an offshore floating wind turbine according to an embodiment of the present invention (the non-contact wave sensor with an integrated inertial navigation module is omitted). Figure 3 This is a connection block diagram of the wave data monitoring system for offshore floating wind turbines according to an embodiment of the present invention; Figure 4 This is a flowchart illustrating the principle of the wave data monitoring method for offshore floating wind turbines according to an embodiment of the present invention.

[0038] Figure label: 1-Fixed bracket; 2-Tilting bracket; 3-Sensor fixing component; 4-Tower base guardrail platform; 5-Floating platform; 6-Tilting main frame; 7-Telescopic frame; 8-Telescopic support arm assembly; 9-Data processing module; 10-Motion compensation unit; 11-Data filtering unit; 12-Adaptive statistics unit; 13-Prediction unit; 14-Fusion verification unit; 15-Data output unit; 16-Wind farm service center; 17-Wind turbine control system; 18-Adjusting threaded sleeve. Detailed Implementation

[0039] The present invention will be further described below with reference to specific embodiments: like Figure 1 and Figure 2 As shown, the wave data acquisition device for offshore floating wind turbines described in this embodiment includes a fixed bracket 1, a flip bracket 2, a non-contact wave sensor with an integrated inertial navigation module, and a sensor fixing component 3. Among them, the fixed bracket 1 is fixed to the guardrail on the tower base guardrail platform 4; the flip bracket 2 is rotatably connected to the fixed bracket 1; the sensor fixing part 3 is installed on the flip bracket 2; the non-contact wave sensor with an integrated inertial navigation module is installed on the sensor fixing part 3 and is distributed in an equilateral triangle, and its position is adjusted by the flip bracket 2.

[0040] Specifically, in this embodiment, the flipping bracket 2 includes a flipping main frame 6, a telescopic frame 7, and a telescopic support arm assembly 8; the flipping main frame 6 is provided with a sliding groove; the telescopic frame 7 is disposed in the sliding groove and extends and retracts along the extension direction of the sliding groove; the telescopic support arm assembly 8 is connected between the flipping main frame 6 and the telescopic frame 7. The telescopic support arm assembly 8 is adjusted for extension and retraction by adjusting the threaded sleeve 18.

[0041] Specifically, in this embodiment, the non-contact wave sensor employs a millimeter-wave radar wave meter with a range of 0-50m and a measurement accuracy of ±1%FS. Based on the microwave ranging principle, it receives reflected waves from the sea surface by transmitting microwave signals and calculates the round-trip time difference to obtain the wave height. The sampling frequency is adjustable from 1Hz to 10Hz to ensure the capture of high-frequency wave dynamics. The data output format is ASCII code, including timestamps, original observation values, and device status codes. The inertial navigation module uses an inertial measurement unit (IMU) with a range of ±45° roll / pitch, ±180° heading, and an angular velocity measurement range of ±200° / s. Based on MEMS technology, it senses the 5-angle and linear motion of the floating platform through accelerometers and gyroscopes, outputting real-time pose data.

[0042] Time synchronization is achieved using NTP (Network Time Protocol) v4, unifying all sensor timestamps to UTC time with a synchronization error ≤10ms. Sensor sampling trigger times are calibrated using hardware trigger signals (such as GPS second pulses) to avoid sampling delays. A transformation relationship between the local coordinate system and the inertial coordinate system is established, and a global coordinate system is defined. Carrier coordinate system Sensor coordinate system The sensor's mounting position in the carrier coordinate system is... This ensures spatial consistency between pose data and wave observation data.

[0043] In this embodiment, a non-contact monitoring method is adopted. The wave data acquisition device for the offshore floating wind turbine, equipped with a non-contact wave sensor, is installed on the tower base guardrail platform 4. This eliminates the need for contact with seawater, avoiding problems such as seawater corrosion, biofouling, cable breakage, and equipment overturning that are common with traditional contact equipment. This fundamentally prevents direct damage to the equipment from the marine environment. The non-contact wave sensor, integrated with an inertial navigation module, is mounted on the sensor mounting bracket 3 in an equilateral triangle arrangement for easy wave direction calculation.

[0044] Furthermore, in the wave data acquisition device for offshore floating wind turbines, a flip-up bracket 2 is added between the fixed bracket 1 and the sensor fixing component 3. When it is necessary to install and maintain the non-contact wave sensor, simply flip the flip-up bracket 2 to the inside of the guardrail (the non-contact wave sensor moves to the inside of the guardrail accordingly). When the installation and maintenance of the non-contact wave sensor is completed, flip the flip-up bracket 2 back to the outside of the guardrail (the non-contact wave sensor returns to its original acquisition position). This method is convenient to install, avoids the risks associated with high-altitude installation, reduces the safety risks for maintenance personnel, reduces equipment downtime, and lowers the equipment installation, maintenance, and total lifecycle costs.

[0045] The telescopic frame 7 is located in the slide groove of the tilting main frame 6 and extends and retracts along the extension direction of the slide groove, which facilitates the adjustment of the distance that the non-contact wave sensor moves outward. Moreover, the telescopic support arm assembly 8 is connected between the tilting main frame 6 and the telescopic frame 7, which not only strengthens the stability of the wave data acquisition device structure of the offshore floating wind turbine, but also allows for flexible adjustment of the clearance distance.

[0046] The inertial navigation module integrated into the non-contact wave sensor boasts high dynamic response and overload resistance, enabling motion compensation and collaborative sensing. This solves the measurement challenge of coupling the floating platform's own motion with wave motion, effectively isolating platform motion interference from wave monitoring, restoring true wave characteristic parameters, and improving data accuracy and stability. Simultaneously, it achieves collaborative wave-platform monitoring, providing comprehensive data support for unit structural design, safety management, attitude control, and intelligent operation and maintenance. Furthermore, it enhances sensor installation flexibility and reduces system deployment costs.

[0047] Specifically, this embodiment also includes a method such as Figure 3 The illustrated wave data monitoring system for offshore floating wind turbines includes the aforementioned wave data acquisition device, data processing module 9, and monitoring module. Data processing module 9 processes the data acquired by the wave data acquisition device and transmits the processed data to the monitoring module, which then monitors the ocean wave data.

[0048] In the specific structure, the data processing module 9 includes a motion compensation unit 10, a data filtering unit 11, an adaptive statistics unit 12, a prediction unit 13, a fusion verification unit 14, and a data output unit 15.

[0049] Among them, the motion compensation unit 10 uses the Euler angle transformation method to transform the observation values ​​of the non-contact wave sensor in the carrier coordinate system to the global coordinate system, thereby removing the coupling interference of the six-degree-of-freedom motion of the floating platform 5 on the wave measurement and obtaining the real wave data in the global coordinate system.

[0050] The data filtering unit 11 is used to remove local anomalies caused by sensor noise, sea surface stray echoes, or compensation residuals.

[0051] The adaptive statistical unit 12 is used to calculate the wave height variation coefficient CV and dynamically adjust the sliding window parameters.

[0052] Prediction unit 13 is used to train a prediction model based on historical wave data and output future wave prediction data through the trained prediction model.

[0053] The fusion verification unit 14 is used to perform consistency verification of statistical results and prediction results under multiple constraints.

[0054] Data output unit 15 is used to output real-time wave parameters, statistical characteristic reports and abnormal alarm information.

[0055] The monitoring module includes a remote wind farm service center 16 and a local wind turbine control system 17.

[0056] like Figure 4 As shown, the working principle of the wave data monitoring system for offshore floating wind turbines is as follows: S1. Collect raw wave observation data containing motion disturbances of floating platform 5 and six-degree-of-freedom pose data of floating platform 5 through the wave data acquisition device of offshore floating wind turbine. S2. The motion compensation unit 10 uses the Euler angle transformation method to transform the observation values ​​of the non-contact wave sensor in the carrier coordinate system to the global coordinate system, thereby removing the coupling interference of the six-degree-of-freedom motion of the floating platform 5 on the wave measurement and obtaining the real wave data in the global coordinate system. This step includes: S2-1. Construct the rotation matrix from the carrier coordinate system to the global coordinate system. :

[0057] in, This indicates the roll angle of the floating platform 5. This indicates the pitch angle of the floating platform 5. Indicates the heading angle of the floating platform 5; S2-2, Combining rotation matrix Calculate the target position in the global coordinate system :

[0058] in, For the global location of the carrier, This represents the sensor's position offset in the carrier coordinate system. These are the raw measurements from the non-contact wave sensor. The displacement change of the non-contact wave sensor due to the movement of the floating platform 5; S2-3. For the six-degree-of-freedom motion of the floating platform 5, the true wave data in the global coordinate system after compensation is obtained through inverse calculation:

[0059]

[0060]

[0061] in, To the compensated effective wave height, The wave height is the original wave height obtained from observation. The heave of the floating platform 5 is based on the target position. get; For the compensated wave period, The wave period obtained from the original observation; For the compensated wave direction, The wave direction obtained from the original observation. This is the initial heading angle.

[0062] S3. The data filtering unit 11 removes local anomalies caused by sensor noise, instantaneous interference caused by wave breakage, or compensation residuals from the real wave data to obtain purified wave data. This step uses the following filtering mechanism: S3-1. The compensated wave height data is decomposed using the db4 wavelet basis into low-frequency signals corresponding to the real wave components and high-frequency signals corresponding to the noise components. S3-2, Using an adaptive threshold function Thresholding is performed on high-frequency coefficients to retain effective signal components; sensor electronic noise and transient interference caused by wave breakage are removed. The standard deviation of noise. For data length; S3-3. Calculate the mean of the denoised data sequence. and noise standard deviation ; S3-4. Based on the characteristics of normal distribution, […] Data points within a certain range are marked as outliers; S3-5. If the number of consecutive abnormal points does not exceed 5, linear interpolation is used for replacement; if the number of consecutive abnormal data exceeds 5, the data segment is determined to be subject to strong interference or sensor failure, and is marked as an invalid data segment. It will be automatically skipped in subsequent statistical analysis and feature extraction and will not be included in the calculation. S3-6, Output the purified wave data.

[0063] S4. For the purified wave data, the wave height variation coefficient CV is calculated by adaptive statistical unit 12 and the sliding window parameters are dynamically adjusted to extract historical wave features. The specific process for this step is as follows: The wave height variation coefficient is continuously calculated for the most recent three sliding windows. :

[0064] in, The standard deviation of wave height. The mean value of the wave height; If CV > 0.3: the wave pattern is considered violent, and the wave history data statistics duration is switched to 5-8 minutes with a sampling frequency of 6-10Hz to capture high-frequency dynamic characteristics; if 0.2 < CV ≤ 0.3: the wave history data statistics duration is switched to 15-20 minutes with a sampling frequency of 3-5Hz to reflect stationary statistical characteristics; if CV ≤ 0.2: the wave history data statistics duration is switched to 20-30 minutes with a sampling frequency of 3-5Hz. Furthermore, the statistical duration and sampling frequency of historical wave data are optimized by combining the wave type of the sea area and the operating scenario of the generator unit: For sea areas dominated by wind and waves: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 15-20 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 5-8 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 20-25 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. For sea areas dominated by swells: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 18-20 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 8-10 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 25-30 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. For sea areas dominated by mixed waves: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 16-18 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 6-9 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 22-28 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. After optimization and adjustment, extract historical wave features: Significant wave height Take the arithmetic mean of the heights of 1 / 3 of the large wave within the window:

[0065] in, All are waves within the window. n is the total wave number within the window; maximum wave height : Maximum wave height within the window; Wave height standard deviation : Reflects the degree of dispersion in wave height distribution:

[0066] For the first in the window A wave, The arithmetic mean of the wave heights within the window; Peak period The wave data is converted to the frequency domain using Fourier transform, and the period corresponding to the frequency of maximum energy is extracted.

[0067] The peak frequency of the energy spectrum; Average period Arithmetic mean of all wave periods within the window:

[0068] For the first The period of a wave; Periodic variation coefficient :

[0069] The periodic standard deviation; Main wave direction The mode of the wave direction distribution within the statistical window, i.e., the wave direction that appears most frequently; Standard deviation of wave direction distribution : Reflects the degree of wave concentration The smaller the value, the more stable the wave direction.

[0070] In the above, the adaptive statistical unit 12 satisfies the following constraints when performing data statistics: Minimum sample size: A single set of statistical data must contain at least 30 complete wave cycles; Recognition accuracy: After adjustment, the peak and trough recognition accuracy of the zero-point method is ≥90%; System synchronization: The statistical period matches the data update frequency of the wind turbine control system 17.

[0071] By using the above constraints, we can avoid errors caused by insufficient sample size or conflicts between statistical time and data transmission cycle.

[0072] S5. Train the prediction model based on historical wave characteristics through prediction unit 13 and predict future wave prediction data. This step selects different prediction models for prediction based on different time series forecasting needs; in, For long-term windows of 30 minutes or more and long-period forecasts of over 1 hour, the Temporal Fusion Transformer (TFT) prediction model is selected. TFT, as a deep learning architecture based on an attention mechanism, demonstrates significant advantages in processing time-series data. Its core features are mainly reflected in four key components: First, a gated residual network (GRN), which intelligently filters and combines input features to ensure full utilization of effective features; second, a variable selection network, which automatically identifies which data among numerous historical data sets has the most critical impact on future predictions, improving prediction accuracy; third, a hybrid attention mechanism, which simultaneously captures long-term dependencies (such as tidal patterns) and short-term abrupt changes in the data, taking into account both long-term trends and short-term fluctuations; and fourth, quantile prediction functionality, which, unlike traditional methods, directly outputs the 95% confidence interval of the predicted value without pre-assuming that the data conforms to a normal distribution, enhancing the reliability of the prediction results. In practical applications, the TFT prediction model uses the historical feature sequence of the past 24 hours as input data to achieve accurate prediction of wave parameters for the next hour.

[0073] For predicting nonlinear complex fluctuations within a short time window of 5-10 minutes (i.e., fluctuations with a wave height variation coefficient CV > 0.3), the TCN prediction model is selected. The TCN prediction model effectively solves the prediction challenge of short-term complex fluctuations through a unique stacked structure of "causal convolution" and "dilated convolution." Causal convolution strictly follows the chronological order, ensuring that the prediction process relies only on past data, fundamentally avoiding the "future leakage" problem and guaranteeing the rationality of the prediction. Dilated convolution, on the other hand, has significant advantages, allowing the network to obtain a large "receptive field" with fewer layers. The network can simultaneously consider historical data from several hours ago and instantaneous abrupt changes from several minutes ago, comprehensively grasping the fluctuation patterns of the data. In this embodiment, the TCN prediction model is configured with the following parameters: a 5-layer stacked network, a convolution kernel size of 3, and dilation factors set to 1, 2, 4, 8, and 16 respectively. With this parameter configuration, the model is specifically designed to predict short-term wave characteristics within the next 5-15 minutes, providing an effective solution for dealing with short-term complex fluctuations.

[0074] For stationary data, i.e., data with a wave height coefficient of variation (CV) ≤ 0.2, the N-BEATS prediction model is selected. Through deep learning, basis function decomposition is performed to achieve highly automated prediction of stationary sequences. As a deep learning model that requires no manual parameter tuning, the N-BEATS prediction model greatly simplifies the model application process and reduces operational difficulty. Its core capability lies in its ability to automatically decompose time series data into three key components: a trend block, primarily used to fit the slowly changing trend in the data, reflecting long-term trends; a seasonality block, responsible for fitting the predictable periodic features in the data, capturing regular periodic fluctuations; and a residual block, which focuses on learning the remaining random fluctuations in the data, handling irregular changes that are difficult to explain by trends and seasons. By stacking multiple such block structures, N-BEATS can automatically complete the differencing and fitting process of time series data without manual intervention. This not only improves the convenience of model application but also makes the prediction results more robust, stably handling the prediction needs of linear stationary data.

[0075] In the entire wave data prediction process, the input data uses a sliding window feature sequence of the past 24 hours. Specifically, to predict wave parameters for the next 30 minutes, the wave feature data of the previous 10 30-minute windows are input. The data format is uniformly CSV or TXT, which includes key timestamp information and corresponding feature values ​​to ensure data integrity and readability. The output is designed according to actual application needs, mainly including wave parameter predictions for three different time scales: 10 minutes, 30 minutes, and 1 hour. It also includes a 95% confidence interval derived from a normal distribution estimate based on the model prediction error. This output setting not only meets the prediction needs at different time scales but also provides a reference for the reliability of the prediction results through the confidence interval, making the entire prediction output more comprehensive and practical.

[0076] S6. The statistical results obtained in step S4 and the prediction results obtained in step S5 are checked for consistency under multiple constraints by the fusion verification unit 14. The process for this step is as follows: S6-1. Associate the historical wave features of the latest window with the predicted data to construct a continuous "history-future" feature sequence with a dimension of 1×(M×2), where M is the number of features in a single window; S6-2. Calculate the deviation between the predicted value and the average value of the historical wave characteristics of the last 3 windows to verify the error threshold. If the following conditions are met: wave height deviation ≤ 20%, period deviation ≤ 15%, wave direction deviation ≤ 10, then proceed to step S6-3. Otherwise, mark it as an error exceeding the standard and return to step S4. S6-3. Calculate the standard deviation of the multi-model prediction results for consistency verification. If the standard deviation is ≤5%, the prediction is considered stable; otherwise, it is marked as unstable and the process returns to step S4.

[0077] S7. Output real-time wave parameters, statistical characteristic reports and abnormal alarm information through data output unit 15; S8. Monitor ocean wave data through the monitoring module based on real-time wave parameters, statistical feature reports, and abnormal alarm information.

[0078] The real-time wave parameters include effective wave height, peak period, and main wave direction, along with a data validity indicator ("valid" / "to be corrected"). Statistical Feature Report (output by window): Historical 10 / 30 minute wave height distribution histogram, periodic change trend chart, wave direction rose diagram; Forecast data (next 10 minutes - 1 hour): wave parameter forecast values ​​and 95% confidence intervals, forecast accuracy assessment report; Anomaly Alarm Information: When the data is invalid or the prediction deviation exceeds the standard, an alarm signal is output (including the anomaly type and suggested handling solution).

[0079] In this embodiment, it can be integrated with the remote operation and maintenance mode of wind farms. Operation and maintenance personnel can view wind farm wave data in real time at the wind farm service center 16 without on-site duty, which can reduce labor costs.

[0080] Compared to the lag and accuracy degradation issues that are common in traditional fixed-time-window monitoring, this embodiment employs an adaptive time-sliding method in the adaptive statistical unit 12. This method dynamically adjusts the window duration based on the wave height variation coefficient, combined with the sea wave pattern, generator operating scenario, and sampling frequency. During normal monitoring, it balances data representativeness and real-time performance; during extreme weather warnings, it quickly captures extreme values; and during maintenance assessments, it reduces fluctuation interference. This improves the accuracy of wave height measurement and the reliability of the zero-point identification method, while also ensuring real-time response and data integrity. A wave feature database is constructed to solve the challenge of balancing real-time monitoring and data validity under complex wave conditions.

[0081] By employing a method combining "sliding window statistics + time-series prediction + error feedback correction," the system adapts to the duration requirements of adaptive time sliding windows, addressing both the statistical lag issue of traditional monitoring and improving the accuracy of prediction results. It integrates multiple prediction models adapted to different conditions, using continuous historical window statistical features as input and outputting predicted values ​​consistent with the window duration. Simultaneously, error feedback dynamically adjusts model parameters and supplements the training set, continuously optimizing prediction accuracy. This ensures the timeliness of statistical data, preventing control commands from becoming out of sync with actual wave conditions. It provides timely warnings in the face of extreme and sudden waves, buying time for turbine risk avoidance. By matching the different feature requirements of long and short windows and ensuring accurate timestamp synchronization, it further enhances the reliability of wave monitoring. This provides precise data support for real-time control of turbine yaw and pitch, and helps optimize the monitoring system's adaptability to complex wave conditions. It provides high-quality data assurance for the safe operation, risk warning, and maintenance decisions of floating wind turbines, effectively compensating for insufficient monitoring accuracy caused by inappropriate selection of historical data duration.

[0082] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A wave data acquisition device for offshore floating wind turbines, characterized in that, Includes a fixed bracket, a flip bracket, a non-contact wave sensor with an integrated inertial navigation module, and sensor mounting hardware; The fixed bracket is fixed to the guardrail on the tower base guardrail platform; The flip-up bracket is rotatably connected to the fixed bracket; The sensor mounting bracket is installed on the flip-up bracket; The non-contact wave sensor with integrated inertial navigation module is mounted on the sensor fixture and arranged in an equilateral triangle. Its position is adjusted by the flip bracket.

2. The wave data acquisition device for offshore floating wind turbines according to claim 1, characterized in that, The flipping support includes a flipping main frame, a telescopic frame, and a telescopic support arm assembly; The main frame for flipping is provided with a sliding groove; The telescopic frame is installed inside the slide groove and extends and retracts along the extension direction of the slide groove; The telescopic support arm assembly is connected between the tilting main frame and the telescopic frame.

3. A wave data monitoring system for offshore floating wind turbines, characterized in that, Includes the wave data acquisition device, data processing module, and monitoring module for offshore floating wind turbines as described in claim 1 or 2; The data processing module processes the data collected by the wave data acquisition device of the offshore floating wind turbine and transmits the processed data to the monitoring module, which then monitors the ocean wave data. The data processing module includes a motion compensation unit, a data filtering unit, an adaptive statistics unit, a prediction unit, a fusion verification unit, and a data output unit. The motion compensation unit uses the Euler angle transformation method to transform the observation values ​​of the non-contact wave sensor in the carrier coordinate system to the global coordinate system, thereby eliminating the coupling interference of the six-degree-of-freedom motion of the floating platform on the wave measurement and obtaining the real wave data in the global coordinate system. The data filtering unit is used to remove local anomalies caused by sensor noise, sea surface stray echoes, or compensation residuals. The adaptive statistical unit is used to calculate the wave height variation coefficient CV and dynamically adjust the sliding window parameters; The prediction unit is used to train a prediction model based on historical wave data and output future wave prediction data through the trained prediction model. The fusion verification unit is used to perform consistency verification of statistical results and prediction results under multiple constraints. The data output unit is used to output real-time wave parameters, statistical feature reports, and abnormal alarm information; The monitoring module includes a remote wind farm service center and a local wind turbine control system.

4. A method for monitoring wave data of offshore floating wind turbines, characterized in that, This is achieved using the wave data monitoring system for offshore floating wind turbines as described in claim 3, comprising: S1. Collect raw wave observation data containing motion disturbances of the floating platform and six-degree-of-freedom pose data of the floating platform through the wave data acquisition device of the offshore floating wind turbine. S2. By using the Euler angle transformation method through the motion compensation unit, the observation values ​​of the non-contact wave sensor in the carrier coordinate system are transformed to the global coordinate system, thereby eliminating the coupling interference of the six-degree-of-freedom motion of the floating platform on the wave measurement and obtaining the real wave data in the global coordinate system. S3. The data filtering unit removes local anomalies caused by sensor noise, instantaneous interference from wave breakage, or compensation residuals from the real wave data to obtain purified wave data. S4. For the purified wave data, calculate the wave height coefficient of variation CV through an adaptive statistical unit and dynamically adjust the sliding window parameters to extract historical wave characteristics; S5. Through the prediction unit, train a prediction model based on historical wave characteristics and predict future wave prediction data; S6. Through the fusion verification unit, perform multi-constraint condition consistency verification on the statistical results obtained in step S4 and the prediction results obtained in step S5; S7. Output real-time wave parameters, statistical feature reports, and abnormal alarm information through the data output unit; S8. Through the monitoring module, monitor the ocean wave data based on the real-time wave parameters, statistical feature reports, and abnormal alarm information.

5. The wave data monitoring method for offshore floating wind turbines according to claim 4, characterized in that, Step S2 includes: S2-1. Construct the rotation matrix from the carrier coordinate system to the global coordinate system. : in, Indicates the roll angle of a floating platform. Indicates the pitch angle of a floating platform. Indicates the heading angle of the floating platform; S2-2, Combining rotation matrix Calculate the target position in the global coordinate system : in, For the global location of the carrier, This represents the sensor's position offset in the carrier coordinate system. These are the raw measurement values ​​from the non-contact wave sensor. The displacement change of a non-contact wave sensor caused by the movement of a floating platform; S2-3. For the six-degree-of-freedom motion of the floating platform, through reverse operation compensation, obtain the real wave data in the compensated global coordinate system: in, To the compensated effective wave height, The wave height is the original wave height obtained from observation. The heave of the floating platform is based on the target location. get; For the compensated wave period, The wave period obtained from the original observation; For the compensated wave direction, The wave direction obtained from the original observation. This is the initial heading angle.

6. The wave data monitoring method for offshore floating wind turbines according to claim 4, characterized in that, Step S3 is filtered through the following filtering mechanism: S3-1. Use the db4 wavelet basis to decompose the compensated wave height data into a low-frequency signal corresponding to the real wave component and a high-frequency signal corresponding to the noise component; S3-2, Using an adaptive threshold function Thresholding is performed on high-frequency coefficients to retain effective signal components; sensor electronic noise and transient interference caused by wave breakage are removed. The standard deviation of noise. For data length; S3-3. Calculate the mean of the denoised data sequence. and noise standard deviation ; S3-4. Based on the characteristics of normal distribution, […] Data points within a certain range are marked as outliers; S3-5. If the number of consecutive abnormal points does not exceed 5, use linear interpolation for replacement; If the number of consecutive abnormal data exceeds 5, it is determined that this data segment is affected by strong interference or sensor failure, marked as an invalid data segment, and automatically skipped during subsequent statistical analysis and feature extraction without participating in the calculation; S3-6. Output the purified wave data.

7. The wave data monitoring method for offshore floating wind turbines according to claim 4, characterized in that, Step S4 includes: The wave height variation coefficient is continuously calculated for the most recent three sliding windows. : in, The standard deviation of wave height. The mean value of the wave height; If CV>0.3: It is determined that the waves are violent, the statistical duration of the wave historical data is switched to 5-8 minutes, and the sampling frequency is 6-10 Hz to capture high-frequency dynamic characteristics; if 0.2<CV≤0.3: The statistical duration of the wave historical data is switched to 15-20 minutes, and the sampling frequency is 3-5 Hz to reflect stable statistical characteristics; if CV≤0.2: The statistical duration of the wave historical data is switched to 20-30 minutes, and the sampling frequency is 3-5 Hz; Then, optimize the statistical duration and sampling frequency of the wave historical data in combination with the wave type in the sea area and the unit operation scenario: For sea areas dominated by wind waves: If the unit operation scenario is normal operation monitoring, the statistical duration of the wave historical data is 15-20 minutes, and the sampling frequency is 3-5 Hz; if the unit operation scenario is extreme weather warning, the statistical duration of the wave historical data is 5-8 minutes, and the sampling frequency is 6-10 Hz; if the unit operation scenario is maintenance window period assessment, the statistical duration of the wave historical data is 20-25 minutes, and the sampling frequency is 5-10 Hz; if the unit operation scenario is long-term data accumulation, the statistical duration of the wave historical data is 60 minutes, and the sampling frequency is 3-10 Hz; For sea areas dominated by swell waves: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 18-20 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 8-10 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 25-30 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. For sea areas dominated by mixed waves: If the unit's operating scenario is normal operation monitoring, the historical wave data collection time is 16-18 minutes, and the sampling frequency is 3-5Hz; if the unit's operating scenario is extreme weather warning, the historical wave data collection time is 6-9 minutes, and the sampling frequency is 6-10Hz; if the unit's operating scenario is maintenance window assessment, the historical wave data collection time is 22-28 minutes, and the sampling frequency is 5-10Hz; if the unit's operating scenario is long-term data accumulation, the historical wave data collection time is 60 minutes, and the sampling frequency is 3-10Hz. After optimization and adjustment, extract historical wave features: Significant wave height Take the arithmetic mean of the heights of 1 / 3 of the large wave within the window: in, All are waves within the window. n is the total number of wavenumbers within the window; maximum wave height : Maximum wave height within the window; Wave height standard deviation : Reflects the degree of dispersion in wave height distribution: For the first in the window A wave, The arithmetic mean of the wave heights within the window; Peak period The wave data is converted to the frequency domain using Fourier transform, and the period corresponding to the frequency of maximum energy is extracted. The peak frequency of the energy spectrum; Average period Arithmetic mean of all wave periods within the window: For the first The period of a wave; Periodic variation coefficient : The periodic standard deviation; Main wave direction The mode of the wave direction distribution within the statistical window, i.e., the wave direction that appears most frequently; Standard deviation of wave direction distribution : Reflects the degree of wave concentration The smaller the value, the more stable the wave direction.

8. The wave data monitoring method for offshore floating wind turbines according to claim 7, characterized in that, The adaptive statistical unit satisfies the following constraints when performing data statistics: Minimum sample size: A single set of statistical data must contain at least 30 complete wave cycles; Recognition accuracy: After adjustment, the peak and trough recognition accuracy of the zero-point method is ≥90%; System synchronization: The statistical period matches the data update frequency of the wind turbine control system.

9. The wave data monitoring method for offshore floating wind turbines according to claim 7, characterized in that, In step S5, different prediction models are selected for prediction based on different time series prediction needs; in, For forecasts with a duration of 30 minutes or more and a period of more than 1 hour, the TFT forecast model is selected. For the prediction of nonlinear complex fluctuations with a short duration window of 5-10 minutes, where the nonlinear complex fluctuations are defined as wave height variation coefficients CV>0.3, the TCN prediction model is selected. For stationary data, i.e., data with wave height variation coefficient CV ≤ 0.2, the N-BEATS prediction model is selected, and basis function decomposition is performed through deep learning to achieve highly automated prediction of stationary sequences.

10. The wave data monitoring method for offshore floating wind turbines according to claim 9, characterized in that, Step S6 includes: S6-1. Associate the historical wave features of the latest window with the predicted data to construct a continuous "history-future" feature sequence with a dimension of 1×(M×2), where M is the number of features in a single window; S6-2. Calculate the deviation between the predicted value and the average value of the historical wave characteristics of the last 3 windows to verify the error threshold. If the following conditions are met: wave height deviation ≤ 20%, period deviation ≤ 15%, wave direction deviation ≤ 10, then proceed to step S6-3. Otherwise, mark it as an error exceeding the standard and return to step S4. S6-3. Calculate the standard deviation of the multi-model prediction results for consistency verification. If the standard deviation is ≤5%, the prediction is considered stable; otherwise, it is marked as unstable and the process returns to step S4.

Images