Fan tower damping method to reduce the effect of wake vortex

Abstract

The application discloses a fan tower shock absorption method for reducing the influence of tail vortex, and comprises the following steps: obtaining the operation state, structural high-frequency response, vibration monitoring data and tower inherent frequency parameters of upstream and downstream units in a wind power plant; based on a physical-data hybrid model, using the operation state and high-frequency response data, the vortex shedding frequency interval, the prediction confidence level and the vortex-induced force arrival time window are predicted; based on the vortex shedding frequency interval, the confidence level and the inherent frequency parameter evaluation, the lock frequency risk evaluation result is obtained; when the result meets the trigger condition, the fusion frequency modulation instruction is generated based on the confidence level and the vibration monitoring data; and within the time window, the apparent inherent frequency of the downstream tower is adjusted by driving the actuator. The application integrates prediction uncertainty into dynamic control, realizes robust feedforward-feedback active shock absorption under the premise of avoiding secondary resonance.

Description

Technical Field

[0001] This invention relates to the field of wind turbine generator control and structural vibration reduction technology, and in particular to a method for reducing the vibration of wind turbine tower columns to reduce the influence of wake vortices. Background Technology

[0002] As wind turbines become larger and more flexible, the height and length-to-diameter ratio of the towers are increasing, leading to a decrease in their natural structural frequency. In densely packed array formations of wind farms, the wake generated by upstream turbines creates strong wake vortices that detach behind them. When the periodic excitation force of these wake vortices propagates to the downstream towers and its frequency approaches a certain natural frequency of the tower, it can easily induce large-amplitude vortex-induced frequency-locked resonance, threatening the structural fatigue life and extreme load safety of the towers. Therefore, accurate perception of the wake vortex evolution characteristics and defense against the dynamic response of the towers are crucial technologies for ensuring the structural dynamic stability of large wind turbines.

[0003] Currently, techniques for suppressing vortex-induced vibration of wind turbine towers typically employ passive tuned mass dampers or simple feedback control strategies based on single-point wind speed measurements. Passive approaches pre-tune to a fixed natural frequency of the tower, utilizing the inertial force of the mass to dissipate resonant energy. In active or semi-active control schemes, existing technologies primarily rely on anemometers installed in the nacelle or on a separate wind tower to obtain point wind speeds, combining this with simple empirical constants to calculate the wake vortex shedding frequency. Once the actual tower amplitude exceeds a safe threshold, the system sends an action command to the bottom damper or pitch actuator to change the system damping or forcibly avoid the resonance zone.

[0004] Existing vibration reduction solutions face technical challenges when dealing with complex wake conditions, including insufficient prediction accuracy, the risk of secondary resonance from blind frequency tuning, and poor control robustness. Specifically, the evolution of wakes in large-scale spaces is severely affected by the superposition of multiple aircraft wakes. Purely physical empirical models have limited accuracy in characterizing wind speed deficits under complex multi-aircraft wake conditions, and single-point wind measurement cannot characterize the vortex decoherence of the flow field, leading to a decrease in the accuracy of vortex decoherence frequency prediction. Existing strategies typically only focus on the frequency band deviating from the current frequency band, ignoring the existence of multi-source excitations such as wave and blade passing frequencies in the real environment. When adjusting the structure's natural frequency, it is easy to fall into resonance traps caused by other environmental excitations. Existing control frameworks fail to calculate the uncertainty of feedforward predictions. Once the model fails under extreme flow fields, the lack of adaptive feedforward commands may lead to a decrease in control performance, while pure feedback control cannot establish a physical defense before the excitation force arrives due to response lag. Summary of the Invention

[0005] The purpose of this invention is to provide a method for reducing the vibration of wind turbine tower columns to mitigate the effects of wake vortices, thereby addressing one of the aforementioned problems in the existing technology.

[0006] Technical solution: A method for reducing the vibration of wind turbine tower columns to mitigate the effects of wake vortices, comprising:

[0007] Acquire multi-source operating status data of upstream wind turbines in the wind farm, high-frequency response data of upstream wind turbine structures, and vibration monitoring data of downstream target tower structures, and load pre-configured natural frequency parameters of downstream target tower structures;

[0008] Based on a pre-built physical-data hybrid model, using multi-source operating status data of upstream wind turbines and high-frequency response data of upstream wind turbine structures, the frequency range of vortex shedding and the prediction confidence level of the wake vortex propagating to the downstream target tower are predicted, and the arrival time window of vortex excitation force is calculated.

[0009] Risk assessment is conducted based on the vortex shedding frequency range, prediction confidence level, and the natural frequency parameters of the downstream target tower structure to obtain the frequency locking risk assessment results.

[0010] When the frequency locking risk assessment results meet the preset triggering conditions, a fusion frequency modulation command is generated based on the predicted confidence level and the vibration monitoring data of the downstream target tower structure.

[0011] Within the vortex-induced force arrival time window, the actuator is driven to adjust the apparent natural frequency of the downstream target tower column according to the fusion frequency modulation command.

[0012] Optionally, acquire multi-source operating status data of upstream wind turbines in the wind farm, high-frequency response data of upstream wind turbine structures, and vibration monitoring data of downstream target tower structures, including:

[0013] Collect wind speed, nacelle yaw angle, blade pitch angle, generator speed and thrust coefficient at the height of the upstream wind turbine hub to obtain multi-source operating status data of the upstream wind turbine;

[0014] The crosswind and downwind vibration acceleration signals at the bottom of the upstream wind turbine nacelle were collected to obtain high-frequency response data of the upstream wind turbine structure.

[0015] The triaxial structural vibration response signal of the upper part of the downstream target tower column was collected to obtain the vibration monitoring data of the downstream target tower column structure;

[0016] Using the sampling clock of the upstream wind turbine structure high-frequency response data as a reference, the synchronous timing signal obtained from the wind farm communication network is used to align the upstream wind turbine multi-source operating status data, the upstream wind turbine structure high-frequency response data, and the downstream target tower structure vibration monitoring data.

[0017] Optionally, based on a pre-built physical-data hybrid model, using multi-source operating status data and high-frequency response data of the upstream wind turbine structure, the prediction of the vortex shedding frequency range and prediction confidence level at the downstream target tower includes:

[0018] By using a pre-built engineering wake analysis model to process multi-source operating status data of upstream wind turbines, the basic predicted value of wake velocity loss is obtained.

[0019] Extract the vortex shedding intensity feature vector characterizing the flow field properties from the high-frequency response data of the upstream wind turbine structure;

[0020] The baseline predicted value of the wake velocity deficit and the vortex shedding intensity feature vector are jointly input into a pre-built data-driven residual compensation module for online correction to obtain the compensated downstream flow velocity.

[0021] Based on the compensated downstream flow velocity, and combined with the calculated model's cognitive uncertainty and input uncertainty, the vortex shedding frequency range and prediction confidence level are obtained.

[0022] Optionally, the process of using a pre-built engineering wake analysis model to process multi-source operating status data of upstream wind turbines to obtain the basic predicted value of wake velocity loss specifically includes:

[0023] Extract the current incoming wind speed, thrust coefficient, and lateral offset determined based on the nacelle yaw angle from the multi-source operating status data of the upstream wind turbine;

[0024] The inflow velocity at the downstream target tower section was calculated using the dimensionless modified Gaussian wake analytical model. The calculation results were used as the basic predicted value of wake velocity deficit.

[0025] Among them, the Gaussian wake analytical model constructs an analytical calculation formula that satisfies the conservation of physical dimensions based on the ratio of the thrust coefficient to the square of the dimensionless width.

[0026] Optionally, vortex shedding intensity feature vectors characterizing the flow field properties are extracted from the high-frequency response data of the upstream wind turbine structure, including:

[0027] The crosswind acceleration time history signal in the high-frequency response data of the upstream wind turbine structure is divided into multiple window segments with overlapping ratios by sliding window segmentation according to the preset time window length.

[0028] Short-time spectrum analysis is performed on the signal of each window segment to extract the dominant frequency value corresponding to the peak value of the spectrum in each window segment, the energy concentration degree composed of the ratio of spectral energy in the neighborhood of the peak frequency to the total spectral energy, and the rate of change of the dominant frequency value of adjacent window segments.

[0029] The extracted dominant frequency, energy concentration, and rate of change are combined in chronological order to form a vortex shedding intensity feature vector.

[0030] Optionally, the predicted wake velocity deficit and the vortex shedding intensity eigenvector are jointly input into a pre-built data-driven residual compensation module for online correction to obtain the compensated downstream flow velocity, including:

[0031] Extract the operating condition parameter vector containing the current incoming wind speed, thrust coefficient, and yaw angle from the multi-source operating status data of the upstream wind turbine;

[0032] The operating condition parameter vector and the vortex shedding intensity characteristic vector are jointly input into the pre-built data-driven residual compensation module to calculate the velocity correction caused by the current complex flow field effect.

[0033] The downstream inflow velocity after compensation is obtained by algebraically superimposing the basic predicted value of the wake velocity loss with the velocity correction.

[0034] Optionally, based on the compensated downstream flow velocity, and combining the calculated model cognitive uncertainty and input uncertainty, the vortex shedding frequency range and prediction confidence level are calculated, including:

[0035] Based on the pre-configured Strauhal number and the reference cross-section diameter of the downstream tower column, the compensated downstream inflow velocity is linearly mapped to the frequency space to obtain the estimated value of the vortex shedding frequency point.

[0036] In the inference phase of the data-driven residual compensation module, multiple random forward propagations with a random dropout mechanism are performed to statistically determine the model's cognitive uncertainty, and the input measurement uncertainty is calculated based on the pre-configured sensor accuracy specifications.

[0037] The combined model cognitive uncertainty and input measurement uncertainty constitute the comprehensive frequency standard deviation, and the estimated vortex shedding frequency point is expanded on both sides using the preset coverage factor and the comprehensive frequency standard deviation to generate the vortex shedding frequency range.

[0038] The probability value corresponding to the preset coverage factor is calculated based on the cumulative distribution function of the standard normal distribution, and is used as the prediction confidence level.

[0039] Optionally, a risk assessment is conducted based on the vortex shedding frequency range, the predicted confidence level, and the natural frequency parameters of the downstream target tower structure to obtain the frequency locking risk assessment results, including:

[0040] The upper and lower boundary values ​​of the vortex shedding frequency range are converted into the incoming flow velocity boundary. Combined with the natural frequencies of each mode in the natural frequency parameters of the downstream target tower structure and the diameter of the tower section, the fluctuation range of the corresponding reduced velocity values ​​is calculated respectively.

[0041] Calculate the intersection ratio between the fluctuation range of each order reduction velocity value and the preset frequency-locking reduction velocity range to determine the frequency-locking risk level;

[0042] When the frequency locking risk level meets the preset triggering conditions, the threatened mode order is dynamically identified in the natural frequency of each mode according to the principle of being closest to the center value of the preset frequency locking reduction speed interval.

[0043] The threatened mode order and its corresponding intrinsic frequency are used as the results of frequency locking risk assessment.

[0044] Optionally, based on the predicted confidence level and vibration monitoring data of the downstream target tower structure, a fused frequency modulation command is generated, including:

[0045] Extract the multi-source environmental excitation frequency, including the frequency of wind turbine blade passage, based on the multi-source operating status data of upstream wind turbines;

[0046] Calculate the minimum frequency offset required to shift the natural frequencies corresponding to the threatened mode order to higher and lower frequencies, respectively, to exit the preset frequency-locked reduction velocity range.

[0047] The minimum absolute distance between the natural frequency after the minimum offset frequency and the excitation frequency of each multi-source environment is calculated and defined as the safety margin in the high-frequency direction and the safety margin in the low-frequency direction, respectively.

[0048] The direction corresponding to a larger safety margin is selected as the target frequency modulation direction;

[0049] Based on the target frequency modulation direction, the predicted confidence level, and the vibration monitoring data of the downstream target tower structure, a fused frequency modulation command is generated.

[0050] Optionally, the event-triggered online correction step based on a pre-built physical-data hybrid model includes:

[0051] After the vortex-induced force reaches its arrival time window, the measured vortex shedding frequency is identified from the vibration monitoring data of the downstream target tower structure by detecting the spectral peak.

[0052] Calculate the amplitude deviation between the measured vortex shedding frequency and the center value of the vortex shedding frequency interval, and determine whether the measured vortex shedding frequency falls within the vortex shedding frequency interval in order to calculate the interval coverage index.

[0053] When the amplitude deviation exceeds the preset allowable deviation threshold, or when the interval coverage index shows an uncovered event, the operating condition parameters that triggered the current event, the vortex shedding intensity characteristics, and the measured vortex shedding frequency are extracted to form new training samples.

[0054] Using newly added training samples, the residual compensation parameters in the pre-built physical-data hybrid model are locally updated using an incremental learning approach.

[0055] Optionally, the arrival time window of the vortex-induced force is calculated, including:

[0056] Extract the inflow wind speed at the hub height at the current moment from the multi-source operating status data of the upstream wind turbine;

[0057] Obtain the pre-configured projection spacing of upstream and downstream turbines along the prevailing wind direction in the wind farm layout, and the pre-configured convection velocity ratio characterizing the wake structure with the average wind field propagation speed.

[0058] Divide the product of the pre-configured projection spacing and the pre-configured convection velocity ratio and the incoming wind speed at the hub height to estimate the time delay of the wake vortex propagating from the upstream wind turbine to the downstream tower section.

[0059] Based on the time delay, the predicted time lead from the current moment to the actual impact of the change on the downstream target tower column is determined, and this lead is used as the time window for the arrival of the vortex-induced force.

[0060] Optionally, within the vortex-induced force arrival time window, the actuator is driven to adjust the apparent natural frequency of the downstream target tower column according to the fused frequency modulation command, including:

[0061] A semi-actively tuned mass damper is used as the actuator;

[0062] Within the available time margin of the vortex-induced force arrival time window, the spring stiffness or the combined state of the additional mass of the semi-active tuned mass damper is automatically adjusted according to the fusion frequency modulation command.

[0063] By changing the combined state, the operating frequency of the semi-actively tuned mass damper is driven to shift along the target direction, causing the apparent natural frequency of the coupled system consisting of the downstream target tower and the semi-actively tuned mass damper to shift to the target value, thus preventing the establishment of vortex-induced locking conditions.

[0064] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer-readable storage medium, wherein the computer program is configured to execute the above-described method for deployable protection of wind turbine towers under extreme sea conditions when running.

[0065] Beneficial effects: This invention incorporates predictive uncertainty into dynamic control, achieving robust feedforward-feedback active vibration reduction while avoiding secondary resonance. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the basic implementation steps of the wind turbine tower vibration reduction method for reducing the impact of wake vortices in the embodiments of this application.

[0067] Figure 2 This is a flowchart illustrating the specific steps involved in establishing a multi-source data acquisition and synchronization benchmark in the embodiments of this application.

[0068] Figure 3 This is a flowchart illustrating the specific construction steps of the vortex shedding frequency prediction engine based on the physical-data hybrid model in this application embodiment.

[0069] Figure 4 This is a flowchart illustrating the steps involved in obtaining the frequency locking risk assessment results in this embodiment of the application.

[0070] Figure 5 This is a flowchart illustrating the steps for generating fusion frequency modulation commands in an embodiment of this application. Detailed Implementation

[0071] Example 1: This example details the system-level top-level architecture and basic implementation process of a wind turbine tower vibration reduction method for reducing wake vortex effects, such as... Figure 1 As shown.

[0072] Step 101: Obtain multi-source operating status data of upstream wind turbines, high-frequency response data of upstream wind turbine structures, and vibration monitoring data of downstream target tower structures in the wind farm, and load the pre-configured natural frequency parameters of downstream target tower structures.

[0073] Specifically, modern large-scale wind farms contain multiple wind turbine generators arranged in an array. When the natural flow passes through the rotor of the upstream turbine, it creates a wake region behind it, generating a wake vortex with velocity deficit and increased turbulence. As the wake vortex propagates downstream and acts on the downstream target tower, it can easily induce strong vortex-induced vibrations.

[0074] To achieve early detection and proactive defense against the aforementioned wake vortex effects, the system needs to establish a comprehensive data perception foundation. Specifically, multi-source operating status data of the upstream wind turbines are used to characterize the macroscopic boundary conditions for wake vortex generation; high-frequency response data of the upstream wind turbine structure are used to deduce the microscopic characteristics of the flow field; and vibration monitoring data of the downstream target tower structure are used to provide a real-time feedback benchmark for the control system. Simultaneously, by loading pre-configured natural frequency parameters of the downstream target tower structure, the system establishes a structural dynamics baseline for subsequent frequency comparison and deviation calculations.

[0075] In some optional implementations, the hardware system executing this embodiment may specifically include a processor, a memory, and various sensor communication interfaces. The processor continuously receives data streams sent by upstream and downstream sensor nodes through a communication network and caches them uniformly in the memory, providing a physical carrier for subsequent feedforward prediction and real-time control.

[0076] Step 102: Based on the pre-built physical-data hybrid model, using the multi-source operating status data of the upstream wind turbine and the high-frequency response data of the upstream wind turbine structure, predict the vortex shedding frequency range and prediction confidence level at the downstream target tower column, and calculate the vortex-induced force arrival time window.

[0077] In this embodiment, a pre-constructed physics-data hybrid model is introduced. This model combines analytical derivation from fluid dynamics with a data-driven online compensation mechanism, enabling it to handle complex atmospheric thermal stability variations and nonlinear flow field effects under the superposition of multiple turbine wakes. By jointly processing the two types of data acquired from upstream, the system outputs not only a single frequency point estimate, but also a vortex shedding frequency range containing probability distribution characteristics and the corresponding prediction confidence level. Furthermore, the system calculates the time delay required for the wake vortex to propagate from the upstream wind turbine to the downstream target tower by calculating the vortex induction arrival time window. This time lead is a prerequisite for active intervention, providing sufficient system response time for subsequent frequency tuning by the driving physical mechanism.

[0078] Furthermore, the prediction process explicitly calculates and transmits the uncertainty to the subsequent control links, so that the control conservatism can be automatically adjusted according to the reliability of the prediction, avoiding the divergence of the control system caused by the inaccuracy of the prediction under complex operating conditions.

[0079] Step 103: Based on the vortex shedding frequency range, the predicted confidence level, and the natural frequency parameters of the downstream target tower structure, a risk assessment is conducted to obtain the frequency locking risk assessment results.

[0080] Specifically, when the vortex shedding frequency of the flow field approaches the natural frequency of the structure, a destructive frequency-locking phenomenon will occur. This embodiment cross-compares the predicted flow field-side frequency characteristics, i.e., the vortex shedding frequency range, with the structural-side parameters, i.e., the natural frequency parameters of the downstream target tower structure. During the comparison process, the system, combined with the uncertainty boundary determined by the prediction confidence level, comprehensively assesses the potential probability and severity of resonance induced by the flow field in the structure, thereby generating a frequency-locking risk assessment result. This assessment result indicates whether the structure is in a dangerous state under the current operating conditions and the specific threatened modes.

[0081] In some specific implementations, frequency locking risk assessment results can be divided into different discrete levels. For example, the system can classify risks into three levels—low risk, medium risk, and high risk—based on the predicted overlap, thereby providing decision instructions for whether to initiate an active intervention mechanism.

[0082] Step 104: When the frequency locking risk assessment result meets the preset triggering conditions, a fusion frequency modulation command is generated based on the predicted confidence level and the vibration monitoring data of the downstream target tower column structure.

[0083] In this embodiment, once the structure is determined to face a substantial resonance threat, the system enters the decision generation phase. Since pure feedforward control relies excessively on prediction accuracy, while pure feedback control has inherent response lag, this embodiment employs a dynamic trade-off strategy.

[0084] Specifically, the system uses the prediction confidence level as an adaptive adjustment factor to organically fuse the target correction calculated based on feedforward prediction with the real-time feedback correction calculated based on vibration monitoring data of the downstream target tower structure. Through this fusion mechanism, when the prediction accuracy improves, the system fully leverages the initiative of the feedforward; when the prediction accuracy decreases, the system automatically increases the feedback weight to ensure control robustness, ultimately outputting a fused frequency modulation command that is both timely and safe. Preferably, the fused frequency modulation command includes drive signals mapped to the specific hardware execution level, rather than conventional frequency offset values.

[0085] Step 105: Within the time window of vortex-induced force arrival, drive the actuator to adjust the apparent natural frequency of the downstream target tower column according to the fusion frequency modulation command.

[0086] Specifically, the system utilizes the available time margin provided by the vortex-induced force arrival time window to complete the corresponding physical deployment before the destructive vortex-induced force actually impacts the downstream tower. After receiving the fusion frequency modulation command, the actuator actively causes the apparent natural frequency of the downstream target tower to deviate from the upcoming vortex shedding frequency by changing the mass or stiffness distribution of the coupled system. Through this mechanism of intervention before vibration occurs, the system disrupts the physical basis for the establishment of vortex-induced lock-in conditions, achieving significant vibration reduction protection for the wind turbine tower.

[0087] In another optional embodiment, for scenarios with sudden changes in wind conditions, if the time window for the arrival of vortex-induced force is short, the system can prioritize driving the electromagnetic actuator with the fastest response speed to perform primary frequency decoupling, and then supplement it with other easing actuators to consolidate the frequency modulation effect, thereby improving the reliability of the vibration reduction method under complex wind conditions in all weather conditions.

[0088] Example 2: This example details the specific implementation process of multi-source data acquisition and synchronization benchmark establishment, such as... Figure 2 As shown. As an optional implementation, acquiring multi-source operating status data of upstream wind turbines, high-frequency response data of upstream wind turbine structures, and vibration monitoring data of downstream target tower structures in a wind farm includes:

[0089] Step 201: Collect wind speed, nacelle yaw angle, blade pitch angle, generator speed and thrust coefficient at the height of the upstream wind turbine hub to obtain multi-source operating status data of the upstream wind turbine.

[0090] Specifically, the aforementioned parameters constitute a dataset characterizing the macroscopic operating conditions of the wind turbine generator. In this embodiment, the system extracts conventional parameters from the upstream wind turbine's data acquisition and monitoring system. These parameters are typically recorded as continuous time-series data at second-level sampling intervals. Among them, wind speed and thrust coefficient are used to determine the velocity deficit of the initial wake, while the nacelle yaw angle is used to determine the lateral offset trajectory of the wake centerline, providing a reliable baseline calculation input for subsequent physical engineering wake analytical models.

[0091] In some alternative implementations, the wind speed at hub height can be obtained either by a mechanical anemometer mounted on the top of the nacelle or by pre-scanning with a forward-looking lidar system. Using lidar for wind speed data avoids aerodynamic interference from rotor rotation, further improving the measurement accuracy of macroscopic boundary condition parameters.

[0092] Step 202: Collect crosswind and downwind dual-channel vibration acceleration signals at the bottom of the upstream wind turbine nacelle to obtain high-frequency response data of the upstream wind turbine structure.

[0093] Specifically, this embodiment utilizes a high-frequency accelerometer array installed at the bottom of the upstream wind turbine nacelle to acquire structural response signals. The system sets the sampling frequency of the high-frequency accelerometer to no less than 50 Hz to distinguish the spectral characteristics of the vortex shedding frequency in the flow field from the vibration signal. Simultaneous acquisition via crosswind and downwind channels is employed to capture the minute oscillation trajectory of the upstream tower column in a two-dimensional horizontal plane.

[0094] Furthermore, in harsh environments such as offshore wind farms, high-frequency acceleration sensors can be piezoelectric sensors with built-in constant current sources.

[0095] Step 203: Collect the triaxial structural vibration response signal of the upper part of the downstream target tower column to obtain the vibration monitoring data of the downstream target tower column structure.

[0096] In this embodiment, the system specifies that crosswind structural vibration response signals are continuously acquired from a triaxial accelerometer located approximately two-thirds of the height of the downstream target tower. This two-thirds distance from the tower height typically represents an area with significant mode displacements in the first and second crosswind bending modes of the wind turbine tower; acquiring signals at this location yields a superior signal-to-noise ratio for resonant response observations. Simultaneously, the system is configured to maintain the same sampling frequency as the upstream high-frequency sensor.

[0097] As a supplement to the above embodiments, if the downstream target tower is an ultra-high flexible tower, the system can also add multiple sets of triaxial acceleration sensors at equal intervals along the tower height direction, and extract more accurate real-time participation coefficients of each mode through multi-point array signal fusion technology.

[0098] Step 204: Using the sampling clock of the upstream wind turbine structure high-frequency response data as a reference, the upstream wind turbine multi-source operating status data, the upstream wind turbine structure high-frequency response data, and the downstream target tower structure vibration monitoring data are aligned using the pre-acquired synchronization time signal.

[0099] Specifically, since the three types of data mentioned above originate from different hardware acquisition systems, their data sampling frequencies and initial timestamps exhibit heterogeneity. Before clock alignment, the system independently performs preliminary quality cleaning processes on all received raw data, including missing value imputation, threshold screening for reasonable ranges of physical quantities, and trend term removal. Specifically, for missing values ​​caused by brief sensor outages, a moving average sliding window can be used for imputation; for temperature drift or zero-point drift trend terms in the signal, first-order difference or polynomial fitting can be used for removal.

[0100] After initial cleaning, the system initiates a multi-source data resampling and synchronization alignment engine. For the sampling rate gap between second-level routine operating parameters and high-frequency vibration signals of at least 50 Hz, the system uses the sampling clock of the high-frequency signal as the global time reference. For low-frequency routine operating parameters, the system employs a linear interpolation algorithm to densify them on the time axis and align them to the high-frequency clock node. The calculation formula for this linear interpolation alignment is as follows:

[0101] V interp =V1+(t interp -t1)*(V2-V1) / (t2-t1);

[0102] Among them, V interp V1 represents the target state parameter after interpolation and alignment, V2 represents the actual state parameter at the previous low-frequency sampling time, and t represents the actual state parameter at the next low-frequency sampling time. interp t1 is the timestamp for target alignment under high frequency reference, t2 is the timestamp for the previous low frequency sampling time, and t2 is the timestamp for the next low frequency sampling time.

[0103] Furthermore, to eliminate local area network communication delays caused by the different spatial physical locations of upstream and downstream components, the system extracts a synchronization signal based on a precise time protocol from the wind farm's communication network. This synchronization signal is used to compensate for and correct the local clock offset between upstream and downstream sensors, ensuring that all dynamic heterogeneous data achieves millisecond-level alignment under a unified absolute time base.

[0104] Example 3: This example details the construction and reasoning process of the vortex shedding frequency prediction engine based on a physics-data hybrid model, such as... Figure 3As shown. As an optional implementation, based on a pre-built physical-data hybrid model, using multi-source operating status data and high-frequency response data of the upstream wind turbine structure, the prediction of the vortex shedding frequency range and prediction confidence level at the downstream target tower column by the wake vortex propagation includes:

[0105] Step 301: Use the pre-built engineering wake analysis model to process the multi-source operating status data of the upstream wind turbine to obtain the basic predicted value of wake velocity loss.

[0106] Specifically, the first stage of predicting flow field characteristics is establishing a physical baseline that conforms to basic fluid dynamics laws. The engineering wake analytical model can roughly calculate the velocity attenuation of the wake vortex as it propagates downstream, based on the macroscopic operating boundary conditions of the upstream wind turbine. By establishing a physical baseline, the system can output initial estimates with correct physical causal direction when facing extremely rare operating conditions with zero samples. As an optional implementation method, the process of using a pre-constructed engineering wake analytical model to process multi-source operating state data from the upstream wind turbine to obtain the basic predicted value of wake velocity loss specifically includes:

[0107] Step 301a: Extract the current incoming wind speed, thrust coefficient, and lateral offset determined based on the nacelle yaw angle and pre-configured upstream and downstream unit spatial layout parameters from the multi-source operating status data of the upstream wind turbine.

[0108] In this embodiment, the above three parameters are the independent variables for calculating wake attenuation. The incoming wind speed determines the initial kinetic energy; the thrust coefficient reflects the degree of obstruction of the airflow by the wind turbine; and the lateral offset is calculated by combining the spatial geometry of the upstream and downstream wind turbines with the current nacelle yaw angle using trigonometric functions, and is used to determine the degree of deviation of the downstream target tower center from the wake centerline.

[0109] Step 301b: Calculate the incoming velocity at the downstream target tower section of the wake vortex using the dimensionless modified Gaussian wake analytical model, and use the calculation result as the basic predicted value of the wake velocity deficit.

[0110] Step 301c, wherein the Gaussian wake analytical model constructs an analytical calculation formula that satisfies the conservation of physical dimensions based on the ratio of the thrust coefficient to the square of the dimensionless width.

[0111] Specifically, this embodiment uses a dimensionless modified Gaussian wake analytical formula for physical baseline calculation. The specific analytical calculation formula is as follows:

[0112] U d =U0*(1-(1-sqrt(1-C T / (8*(σ w / D r ) 2)))*exp(-y t 2 / (2*σ w 2 )));

[0113] Among them, U d The baseline forecast value for wake velocity loss is given, where U0 is the current incoming wind speed, and C is the current wind speed. T σ is the thrust coefficient. w D is the radial spread width parameter of the wake. r Where y is the diameter of the wind turbine. t Let be the lateral offset, exp be the exponential function with the natural constant as the base, and sqrt be the square root function.

[0114] In practical applications, by substituting the current incoming wind speed, thrust coefficient, dimensionless wake width ratio, and lateral offset of the upstream wind turbine into the above formula, a basic predicted value characterizing the degree of wake attenuation can be obtained, providing a physical baseline for subsequent residual compensation.

[0115] Step 302: Extract the vortex shedding intensity feature vector characterizing the flow field properties from the high-frequency response data of the upstream wind turbine structure.

[0116] In this embodiment, the system uses the upstream tower as a sensor to perceive the coherence of the flow field. Since single-point measurements by anemometers suffer from insufficient spatial representativeness, measuring the high-frequency vibration response of the microstructure can invert the degree of disturbance of the vortex shedding process by atmospheric turbulence fluctuations. As an optional implementation, a vortex shedding intensity feature vector characterizing the flow field properties is extracted from the high-frequency response data of the upstream wind turbine structure, specifically including:

[0117] Step 302a: The crosswind acceleration time history signal in the high-frequency response data of the upstream wind turbine structure is divided into multiple window segments with overlapping ratios by sliding window segmentation according to the preset time window length.

[0118] In practice, the preset time window length can be set to 10 seconds, and the overlap ratio can be set to 50%. The system continuously extracts the crosswind acceleration time history signal along the time axis. This type of sliding window mechanism with overlapping characteristics makes the extracted flow field features have a smooth transition in the time domain, and can continuously capture the time-varying laws of the flow field eddy shedding characteristics.

[0119] Step 302b: Perform short-time spectrum analysis on the signal of each window segment to extract the dominant frequency value corresponding to the peak value of the spectrum in each window segment, the energy concentration degree composed of the ratio of spectral energy in the neighborhood of the peak frequency to the total spectral energy, and the rate of change of the dominant frequency value of adjacent window segments.

[0120] Specifically, short-time spectral analysis can reveal the frequency component distribution of a signal within a local time window. The dominant frequency corresponds to the actual vortex shedding frequency of the current flow field. The formula for calculating the energy concentration extraction logic is as follows:

[0121] E ratio =E peak / E total ;

[0122] Among them, E ratio E represents energy concentration. peak E represents the integral spectral energy in the neighborhood of the peak frequency. total This represents the total energy of the integrated spectrum of this window segment across the entire frequency band.

[0123] In other words, higher energy concentration indicates stronger vortex decoherence in the flow field, and a greater physical probability of frequency-locked resonance occurring in the downstream target tower; conversely, lower energy concentration indicates that the incoming turbulence disrupts the regular vortex shedding. The rate of change of the dominant frequency value in adjacent window segments characterizes the non-stationary evolution trend of the flow field.

[0124] Step 302c: Combine the extracted dominant frequency value, energy concentration and rate of change in chronological order to form the vortex shedding intensity feature vector.

[0125] In this embodiment, the three features mentioned above are concatenated into a three-dimensional vector structure at the same timestamp. This structure serves as the input parameter for subsequent artificial intelligence algorithms to perceive the degree of distortion in complex flow fields.

[0126] Step 303: The predicted wake velocity deficit and vortex shedding intensity feature vector are jointly input into a pre-built data-driven residual compensation module for online correction to obtain the compensated downstream flow velocity. As an optional implementation, the method of jointly inputting the predicted wake velocity deficit and vortex shedding intensity feature vector into a pre-built data-driven residual compensation module for online correction to obtain the compensated downstream flow velocity specifically includes:

[0127] Step 303a: Extract the operating condition parameter vector containing the current incoming wind speed, thrust coefficient and yaw angle from the multi-source operating status data of the upstream wind turbine.

[0128] Step 303b: Input the operating condition parameter vector and the vortex shedding intensity characteristic vector into the pre-built data-driven residual compensation module to calculate the velocity correction caused by the current complex flow field effect.

[0129] Specifically, the data-driven residual compensation module is implemented using a pre-trained multi-layer feedforward neural network. This multi-layer feedforward neural network consists of an input layer, multiple hidden layers, and an output layer. The input layer receives a joint feature vector formed by concatenating the operating condition parameter vector and the vortex shedding intensity feature vector. Each hidden layer uses a modified linear unit as the activation function, and each hidden layer includes a random dropout layer to support Monte Carlo random dropout sampling during the inference phase. The output layer is a single linear neuron that outputs a scalar velocity correction ΔU. During offline training, the system utilizes residual labels derived from the difference between the measured downstream flow velocity and the physical model's predicted values ​​from historical operating data, and optimizes the network parameters using a mean squared error loss function.

[0130] In some alternative implementations, the neural network does not directly predict the absolute wind speed, but instead learns the nonlinear residuals generated by the physical model when subjected to drastic changes in atmospheric thermal stability or superimposed interference from multiple aircraft wakes. The system's input interface concatenates a vector of operating parameters characterizing macroscopic boundary conditions with a vortex shedding intensity feature vector characterizing microscopic flow field distortion. Through the complementary sensing of these two heterogeneous vectors, the neural network outputs positive or negative velocity corrections.

[0131] Step 303c: Perform an algebraic superposition operation on the basic predicted value of the wake velocity loss and the velocity correction amount to obtain the compensated downstream flow velocity.

[0132] Specifically, the superposition process follows the calculation formula below:

[0133] U c =U d +ΔU;

[0134] Among them, U c To compensate for the downstream flow velocity, U d ΔU represents the baseline predicted value for the wake velocity loss, and ΔU represents the velocity correction amount output by the data-driven residual compensation module.

[0135] In some alternative implementations, if a communication failure occurs in the upstream high-frequency accelerometer, resulting in the loss of soft measurement features, the system provides an alternative prediction scheme. In this alternative scheme, the velocity correction is forcibly reset to zero, the system degenerates into a pure physical baseline prediction mode, and subsequent modules are simultaneously instructed to forcibly expand the prediction uncertainty range to ensure system safety.

[0136] Step 304: Based on the compensated downstream flow velocity, and combined with the calculated model cognitive uncertainty and input uncertainty, the vortex shedding frequency range and prediction confidence level are obtained.

[0137] As an optional implementation method, based on the compensated downstream flow velocity, and combining the calculated model cognitive uncertainty and input uncertainty, the vortex shedding frequency range and prediction confidence level are obtained, specifically including:

[0138] Optionally, based on the pre-configured Strauhal number and the reference cross-sectional diameter of the downstream tower column, the compensated downstream inflow velocity is linearly mapped to the frequency space to obtain the estimated vortex shedding frequency point.

[0139] It should be noted that the Reynolds number of modern megawatt-class wind turbine towers generally falls within the transcritical or supercritical flow range under normal wind speed conditions, rather than the traditional subcritical flow range. Therefore, the pre-configured Strouhal number is no longer a constant empirical value, but rather dynamically interpolated based on real-time calculated high Reynolds numbers between 0.2 and 0.3. Based on this corrected theory, the point estimate is calculated using the following formula:

[0140] f v =(St*U c ) / D t ;

[0141] Among them, f v Here, St is the estimated frequency of the vortex shedding point, and St is the dynamically obtained Strouhal number. c To compensate for the downstream flow velocity, D t This is the reference cross-sectional diameter of the downstream tower column.

[0142] Optionally, during the inference phase of the data-driven residual compensation module, multiple random forward propagations with a random dropout mechanism are performed to statistically determine the model's cognitive uncertainty and convert the input measurement uncertainty according to the pre-configured sensor accuracy specifications.

[0143] Specifically, the random dropout mechanism refers to randomly deactivating a predetermined proportion of neurons in the hidden layer of the neural network during each forward propagation inference. By performing multiple independent inference operations, the output set is collected and its statistical variance is calculated. This variance characterizes the model's cognitive uncertainty regarding the current unfamiliar operating condition. Simultaneously, based on the factory-calibrated accuracy specifications of the anemometer and attitude sensor, the input measurement uncertainty introduced by hardware measurement deviations is calculated using error propagation theory.

[0144] Optionally, the model cognitive uncertainty and input measurement uncertainty are combined to form a comprehensive frequency standard deviation, and the estimated vortex shedding frequency point is expanded on both sides using a preset coverage factor and the comprehensive frequency standard deviation to generate a vortex shedding frequency range.

[0145] The variances of the two uncertainties mentioned above are combined under the assumption that they are independent, and the combined formula is as follows:

[0146] σ f=sqrt(σ model 2 +σ sensor 2 );

[0147] Where, σ f σ represents the overall frequency standard deviation. model σ is the standard deviation of model cognitive uncertainty converted to the frequency domain. sensor This represents the standard deviation of the input measurement uncertainty converted to the frequency domain.

[0148] The system uses a preset coverage factor to calculate the upper and lower boundaries of the extended range, forming the eddy current frequency range. For example, the preset coverage factor can be set to 1.96.

[0149] Optionally, the probability value corresponding to the preset coverage factor can be calculated based on the cumulative distribution function of the standard normal distribution, and used as the prediction confidence level.

[0150] Specifically, the formula for calculating the probability mapping relationship is as follows:

[0151] η = 1 - 2 * Φ(-κ); where η is the prediction confidence level, Φ is the cumulative distribution function of the standard normal distribution, and κ is the preset coverage factor.

[0152] Furthermore, when the coverage factor is set to 1.96, the prediction confidence level calculated using the above formula is approximately 95%. This calculated confidence level reflects the reliability of the current prediction and provides a basis for the subsequent allocation of adaptive fusion control weights.

[0153] Example 4: This example details the process of estimating the prediction time lead using spatial topology and fluid kinematics principles. As an optional implementation, the calculation of the vortex-induced force arrival time window specifically includes:

[0154] Step 401: Extract the inflow wind speed at the hub height at the current moment from the multi-source operating status data of the upstream wind turbine.

[0155] Specifically, the transport process of the wake within the wind farm depends on the initial kinetic energy of the free flow field. The incoming wind speed at hub height, as a characteristic of the macroscopic meteorological boundary, determines the basic rate of the wake's downstream convective migration. In this embodiment, the system extracts the free incoming wind speed value, representing the current second-level operating condition, in real time from the synchronized multi-source operating status data of the upstream wind turbines.

[0156] Step 402: Obtain the pre-configured projection spacing of upstream and downstream units along the prevailing wind direction in the wind farm layout, and the pre-configured convection velocity ratio characterizing the wake structure with the average wind field propagation speed.

[0157] In this embodiment, the pre-configured projection spacing is not the absolute straight-line distance between the two wind turbine towers, but rather the projected length of the line connecting the upstream wind turbine and the downstream target tower on the current prevailing wind direction vector. This spacing dynamically changes with the wind direction, reflecting the actual physical space that the wake vortex needs to traverse. Furthermore, due to the intense turbulent mixing and energy dissipation within the wake, the migration speed of the wake structure itself will be slightly lower than the free-flow wind speed. Therefore, the system introduces a pre-configured convection velocity ratio. Based on extensive measured data from offshore wind farms and hydrodynamic wind tunnel experiments, the empirical range for this pre-configured convection velocity ratio is typically set between 0.7 and 0.9. The specific value can be calibrated based on the annual atmospheric turbulence intensity at the wind farm's geographical location; for example, 0.8 can be selected in highly turbulent sea areas to characterize the hysteresis effect of the wake structure.

[0158] Step 403: Divide the product of the pre-configured projection spacing and the pre-configured convection velocity ratio and the incoming wind speed at the hub height to estimate the time delay of the wake vortex propagating from the upstream wind turbine to the downstream tower section.

[0159] Specifically, this embodiment utilizes fundamental kinematic principles to convert spatial distance and effective propagation speed into time parameters. The calculation formula is as follows:

[0160] T p =L / (c w *U0);

[0161] Among them, T p The time delay for the wake vortex to propagate from the upstream wind turbine to the downstream tower section is given by L, where L is the pre-configured projection spacing, and c is the distance between the two sections. w For the pre-configured convection velocity ratio, U0 is the incoming wind speed at the hub height.

[0162] To further illustrate the estimation process, a specific numerical calculation example is provided. Assume the pre-configured projected spacing measurement under the current prevailing wind direction is 560 meters, the extracted inflow wind speed at hub height is 10 meters per second, and the pre-configured convection velocity ratio called by the system is 0.8. Substituting these parameters into the formula, the effective convection velocity in the denominator is 8 meters per second, and the final estimated time delay is 70 seconds. This value reflects the absolute time required for local flow field disturbances to cross the physical distance between the units.

[0163] Step 404: Based on the time delay, determine the predicted time advance from the current moment when the change of the wake vortex is predicted to the actual impact of the change on the downstream target tower column, and use it as the time window for the arrival of the vortex-induced force.

[0164] In this embodiment, the algorithm computation time consumed by the system to extract high-frequency vibration characteristics from upstream and complete the physical data hybrid model inference is relatively short compared to the fluid propagation time, which is on the order of tens of seconds. Therefore, the estimated time delay mentioned above is mainly converted into the prediction time advance of the system's resonance defense. The system defines this time period as the vortex-induced force arrival time window. This window constitutes a rigid time constraint for subsequent vibration reduction control. The underlying controller must complete the issuance of commands to the physical actuators, stiffness adjustment, and mechanical movement of the mass block before the countdown of this window reaches zero.

[0165] In some alternative implementations, if extreme gusts cause a sudden increase in incoming wind speed, resulting in a small calculated vortex-induced force arrival time window—even shorter than the minimum mechanical action time required for the tuned mass damper to complete the target frequency shift—as an alternative, the system will temporarily block physical frequency tuning commands for the semi-active actuators and instead urgently issue active yaw commands or unified pitch commands to the main control system. This rapid alteration of the wind turbine's aerodynamic damping surface serves as an auxiliary means of composite vibration reduction, ensuring the safety of the tower structure under extremely short-delay scenarios.

[0166] Example 5: This example details the specific implementation process of vortex-induced frequency locking risk assessment and dynamic mode switching. As an optional implementation, risk assessment is performed based on the vortex shedding frequency range, prediction confidence level, and the natural frequency parameters of the downstream target tower structure to obtain the frequency locking risk assessment result, such as... Figure 4 As shown, it specifically includes:

[0167] Step 501: Convert the upper and lower boundary values ​​of the vortex shedding frequency range into the incoming flow velocity boundary. Combine the natural frequencies of each mode in the natural frequency parameters of the downstream target tower structure with the tower section diameter, and calculate the fluctuation range of the corresponding reduced velocity values.

[0168] Specifically, reduced velocity is a dimensionless parameter used in wind engineering to characterize the relative relationship between the fluid flow frequency and the structure's natural frequency. Since the predicted result includes an uncertain vortex shedding frequency range, the system needs to utilize the inverse operation of the Strouhal relation to map the upper and lower limits of this frequency range back to the incoming flow velocity boundaries in the wind speed domain. The specific inverse mapping formula is as follows:

[0169] U bound =(f bound *D t ) / St;

[0170] Among them, U bound The incoming flow velocity boundary is obtained by transforming the vortex shedding frequency range boundary, f bound D represents the boundary value of the vortex shedding frequency range. tWhere is the diameter of the downstream target tower section, and St is the Strouhal number.

[0171] Furthermore, after obtaining the incoming flow velocity boundary, the system retrieves the pre-loaded natural frequency parameters of the downstream target tower structure. Modern megawatt-class wind turbine towers typically exhibit flexibility, and both their first-order and second-order crosswind bending modes can potentially be excited within the operating wind speed range. Therefore, the system calculates the reduced velocity boundary for each mode under the current wind speed boundary. The formulas for calculating each reduced velocity boundary are as follows:

[0172] V r_bound =U bound / (f n_i *D t ); where V r_bound For the reduced velocity boundary value, U bound For the incoming flow velocity boundary, f n_i Let D be the natural frequency of the i-th mode. t The diameter of the downstream target tower section.

[0173] Through the above calculations, for the first and second modes, the system obtained a set of fluctuation ranges for each reduced velocity value, consisting of lower and upper boundary values. These fluctuation ranges transmit the uncertainty of aerodynamic prediction to the structural response assessment stage.

[0174] Step 502: Calculate the intersection ratio of the fluctuation range of each order reduction speed value with the preset frequency-locking reduction speed range to determine the frequency-locking risk level.

[0175] In this embodiment, the preset frequency-locking reduction velocity range is an engineering empirical constant range determined based on wind tunnel tests or historical long-term monitoring data. When the actual reduction velocity falls within this range, the fluid eddy shedding frequency will be locked near the structure's natural frequency, triggering significant eddy-induced resonance. Optionally, the preset frequency-locking reduction velocity range can be set to 5~7.

[0176] Accordingly, the system performs an intersection operation between the fluctuation ranges of the reduced velocity values ​​calculated in the previous step and the aforementioned preset frequency-locked reduced velocity intervals. The intersection ratio is calculated by dividing the width of the intersection interval by the total width of the fluctuation range. Based on this intersection ratio, the system establishes a risk classification mapping mechanism. Specifically, when the intersection ratio is greater than 50%, the system determines that the current operating condition is prone to severe resonance and sets the frequency-locked risk level to high risk; when the intersection ratio is between 20% and 50%, the frequency-locked risk level is set to medium risk; and when the intersection ratio is less than 20%, the frequency-locked risk level is set to low risk.

[0177] Step 503: When the frequency locking risk level meets the preset triggering conditions, the threatened mode order is dynamically identified in the natural frequency of each mode according to the principle of being closest to the center value of the preset frequency locking reduction speed interval.

[0178] Specifically, the preset triggering conditions are typically configured so that the frequency locking risk level reaches medium or high risk. Once the condition is triggered, the system needs to determine the target object for frequency modulation of the subsequent physical actuators. Because the first-order natural frequency of large wind turbine towers is usually low, it is easily excited in low wind speed ranges; while the second-order natural frequency is higher, it is easily excited in medium to high wind speed ranges. Under certain specific transitional wind speeds or conditions with high turbulence, the fluctuation ranges of both modes may simultaneously intersect with the preset frequency locking reduction velocity range.

[0179] To achieve a hardware optimization strategy that reuses two modes with a single actuator, this embodiment employs the principle of proximity of the center value. The system calculates the center point of the first-order and second-order reduced velocity fluctuation ranges respectively, and compares the distances of the two center points to the absolute center point of the preset frequency-locked reduced velocity interval. The closer the distance, the higher the physical probability and energy coupling strength of deep frequency locking of that mode. The system dynamically identifies the closest first-order mode as the most urgently threatened mode order that needs to be defended. Through time-division switching logic between this type of first-order and second-order modes, limited damping resources are allocated to the most dangerous vibration mode.

[0180] In some alternative implementations, if the distance determination results of the two modes are equal, the system will assign a higher defense priority to the lower-order mode as an alternative breakthrough strategy.

[0181] Step 504: The threatened mode order and its corresponding intrinsic frequency are used as the results of the frequency locking risk assessment.

[0182] In this embodiment, after completing the above calculations and discrimination, the system encapsulates the selected threatened mode orders and their corresponding structural natural frequencies. This encapsulated data serves as the final frequency-locking risk assessment result and is output to the subsequent decision control layer.

[0183] Example 6: This example details the specific implementation process of multi-source anti-secondary resonance optimization and adaptive offset calculation. As an optional implementation, based on the predicted confidence level and downstream target tower structure vibration monitoring data, a fused frequency modulation command is generated, such as... Figure 5 As shown, it specifically includes:

[0184] Step 601: Extract multi-source environmental excitation frequencies, including the frequency of wind turbine blade passage, based on the multi-source operating status data of the upstream wind turbine.

[0185] Specifically, large wind turbines operate in complex marine or terrestrial environments. Their tower structures, in addition to bearing the wake vortex excitation force, are also exposed to various periodic environmental excitations. Blindly adjusting the tower frequency solely to avoid vortex shedding frequencies could lead to falling within the resonance zones of other excitation sources. Therefore, the system needs to construct a comprehensive frequency exclusion map.

[0186] In this embodiment, the system uses the generator speed and pre-configured gearbox ratio from the acquired multi-source operating status data of the upstream wind turbine to calculate the fundamental rotational frequency of the wind turbine, i.e., the 1P frequency, and multiplies it by the number of blades to obtain the turbine blade passage frequency, i.e., the 3P frequency. Furthermore, for offshore wind power scenarios, the system can also extract the main period frequency of ocean waves using an external wave radar. The aforementioned 1P frequency, 3P frequency, and wave frequency together constitute the multi-source environmental excitation frequency.

[0187] Step 602: Calculate the minimum frequency offset required to shift the natural frequencies corresponding to the threatened mode order to higher and lower frequencies respectively, thus exiting the preset frequency-locked reduction speed range.

[0188] In this embodiment, the operation of actuators such as the actively tuned mass damper requires energy, and their physical adjustment range has mechanical limits. The system uses the upper and lower boundaries of the preset frequency-locked reduced speed range as references to calculate the positive displacement required to cross the danger zone towards the high-frequency end and the negative displacement required to cross the danger zone towards the low-frequency end. These two values ​​represent the minimum frequency offset required to deviate from the preset frequency-locked reduced speed range towards the high-frequency and low-frequency directions, respectively.

[0189] Step 603: Calculate the minimum absolute distance between the natural frequency after offsetting the minimum frequency and the excitation frequency of each multi-source environment, and define them as the safety margin in the high-frequency direction and the safety margin in the low-frequency direction, respectively.

[0190] Specifically, in the virtual computing space, the system adds and subtracts the corresponding minimum frequency offset to the natural frequencies of the threatened mode orders, simulating two alternative frequency modulation results. The system then calculates the distances between these two alternative results and all known multi-source environmental excitation frequencies. The formula for calculating this minimum absolute distance is as follows:

[0191] d plus =min(|f n_i +Δf min_plus -f e_j |);

[0192] Where, d plus For high-frequency directions, the safety margin is given; min is the operation that takes the minimum value of the set; f n_i Δf represents the natural frequency corresponding to the threatened mode order. min_plusf is the minimum frequency shift towards higher frequencies. e_j Let be the j-th multi-source environmental excitation frequency in the set, and || be the absolute value operator.

[0193] Similarly, the calculation formula for the low-frequency direction is:

[0194] d minus =min(|f n_i -Δf min_minus -f e_j |);

[0195] Where, d minus For the safety margin in the low-frequency direction, Δf min_minus This represents the minimum frequency offset towards lower frequencies.

[0196] The process can be illustrated with an example: Assume the threatened first-order natural frequency is 0.30 Hz, the minimum shift to a higher frequency is 0.05 Hz, and the alternative frequency is 0.35 Hz. If the environment contains a wave excitation frequency of 0.38 Hz and a wind turbine rotation frequency of 0.15 Hz, the system calculates the distance between 0.35 Hz and 0.38 Hz to be 0.03 Hz, and the distance to 0.15 Hz to be 0.20 Hz. The minimum value of 0.03 Hz is taken as the safety margin in the high-frequency direction. The larger this margin, the farther the new frequency after the shift is from any known dangerous excitation source, and the lower the risk of secondary resonance.

[0197] Step 604: Select the direction corresponding to the larger safety margin as the target frequency modulation direction.

[0198] In this embodiment, the system compares the calculated safety margins in the high-frequency and low-frequency directions and determines the one with the larger value as the target frequency modulation direction. Through this comparison mechanism, the system finds a wider safe path in the global frequency space while avoiding the current vortex shedding frequency.

[0199] In some alternative implementations, if the wind turbine is shut down for maintenance or located in calm sea areas, resulting in insignificant multi-source environmental excitation frequencies, the system provides an alternative optimization scheme. In this alternative scheme, the system substitutes the natural frequencies of adjacent orders in the natural frequency parameters of the downstream target tower structure as virtual excitation sources into the aforementioned distance calculation formula. That is, it preferentially selects the direction away from other natural frequencies of the structure itself as the target frequency modulation direction to prevent modal coupling aliasing.

[0200] Step 605: Generate a fused frequency modulation command based on the target frequency modulation direction, the predicted confidence level, and the vibration monitoring data of the downstream target tower structure.

[0201] Specifically, after determining the absolutely safe target frequency modulation direction, the system begins to calculate the actual offset value that needs to be issued in preparation for generating instructions.

[0202] As an optional implementation, a fused frequency modulation command is generated based on the target frequency modulation direction, the predicted confidence level, and the vibration monitoring data of the downstream target tower structure, specifically including:

[0203] Optionally, the upper and lower limits of the vortex shedding frequency range are cross-mapped with the boundary values ​​of the preset frequency-locked reduction velocity range to determine the frequency-locked band that the threatened mode order needs to avoid under the current uncertainty.

[0204] In this embodiment, the system utilizes the uncertainty of the forward prediction output to adjust the dynamic expansion and contraction of the defense width. Since the flow field prediction involves a probabilistic vortex shedding frequency range, the system needs to cross-couple this velocity range with a fixed reduced velocity criterion. The specific cross-mapping calculation logic is as follows: the upper limit of the predicted incoming flow velocity is obtained by dividing the upper limit of the criterion by the lower limit of the criterion, and the lower limit of the predicted incoming flow velocity is obtained by dividing the lower limit of the criterion by the upper limit of the criterion.

[0205] Thus, the static, single-point frequency defense line is expanded into a wide frequency band. The higher the prediction uncertainty, the wider this frequency band, and the greater the range that the actuator needs to circumvent.

[0206] Optionally, based on the target frequency modulation direction, a feedforward target frequency offset is calculated that causes the natural frequency corresponding to the threatened mode order to unidirectionally cross the boundary of the frequency-locked band, with an additional preset safety redundancy.

[0207] Specifically, considering the already determined target frequency modulation direction, the system requires that the frequency modulation action not only shift the frequency out of the locked frequency band but also add an additional safety distance. When the target frequency modulation direction is a high-frequency direction, the calculation formula is:

[0208] Δf ff =f lock_edge -f n_i +m s ;

[0209] When the target frequency modulation direction is low frequency, the calculation formula is:

[0210] Δf ff =f lock_edge -f n_i -m s ;

[0211] Where, Δf ff f is the feedforward target frequency offset. lock_edge f is the frequency value furthest from the side boundary of the frequency band selected based on the target frequency modulation direction.n_i m is the natural frequency corresponding to the threatened mode order. s This is a preset safety redundancy. According to the above formula, regardless of whether the frequency is tuned to a higher or lower frequency, the tuned natural frequency will always be located outside the frequency band of the frequency lock and will maintain a safety distance from the boundary that is not less than the preset safety redundancy.

[0212] Furthermore, the preset safety redundancy is a very small positive number, used to compensate for the steady-state tracking error that may occur in the mechanical actuator after long-term operation. By adding this redundancy, the final actual frequency remains within the absolute safety range even if the actuator fails to fully reach the commanded position.

[0213] Optionally, a fused frequency modulation command is generated based on the feedforward target frequency offset, the predicted confidence level, and the vibration monitoring data of the downstream target tower structure.

[0214] In this embodiment, the feedforward target frequency offset obtained through safety margin optimization and uncertainty mapping calculation possesses high engineering safety attributes. The system deeply integrates and matches this feedforward command with the real-time vibration feedback from the backend under dynamic adjustment of the prediction confidence level to generate a fused frequency modulation command that drives the underlying hardware.

[0215] Example 7: This example details the specific calculation process of the feedforward and feedback dynamic fusion control law and the operation mechanism of the underlying physical actuator. As an optional implementation, based on the feedforward target frequency offset, prediction confidence level, and downstream target tower structure vibration monitoring data, a fusion frequency modulation command is generated, specifically including:

[0216] Step 701: Calculate the feedback frequency offset based on the deviation of the current crosswind vibration acceleration from the preset vibration threshold in the vibration monitoring data of the downstream target tower structure.

[0217] Specifically, the system extracts vibration monitoring data of the downstream target tower structure in real time and calculates the root mean square or peak value of the current crosswind vibration acceleration. When the actual vibration level exceeds the preset vibration threshold, it is judged as resonance energy accumulation. The system inputs this deviation degree into a preset proportional-integral-derivative controller, or uses a nonlinear feedback gain matrix based on structural response sensitivity to calculate the feedback frequency offset required to suppress the current amplitude.

[0218] Specifically, the system calculates the root mean square value of the crosswind vibration acceleration of the downstream target tower column within the current time window, uses the difference between it and the preset vibration threshold as the deviation input, and linearly maps the deviation into the feedback frequency offset through the preset proportional gain coefficient.

[0219] Step 702: Using a preset logical transition function, the prediction confidence level is mapped to feedforward weight coefficients. The preset logical transition function exhibits an S-shaped nonlinear smooth transition characteristic with a preset center confidence threshold as the inflection point.

[0220] Specifically, the system uses the prediction confidence level as the independent variable and substitutes it into a function that exhibits an S-shaped nonlinear smooth transition characteristic. The formula for calculating this logical transition function is as follows:

[0221] w=1 / (1+e^(-α*(η-η0)));

[0222] Where w is the feedforward weight coefficient, e is the natural constant, α is the transition steepness coefficient, η is the current prediction confidence level, and η0 is the center confidence threshold for weight transition.

[0223] For example, the following is the numerical calculation process of the physical mapping effect of the function: Assuming that it is calibrated through a large amount of historical running data, the system's preset center confidence threshold is set to 0.8 and the transition steepness coefficient is set to 20.

[0224] Under the first operating condition, the flow field is relatively stable, and the current prediction confidence level output by the hybrid prediction model is as high as 0.95. Substituting this into the above formula, the result of the exponential part is -3, and the calculated feedforward weight coefficient is approximately 0.95. At this point, the system highly trusts the prediction results, and feedforward control dominates.

[0225] In the second operating condition, encountering intense and complex turbulence, the current prediction confidence level output by the hybrid prediction model plummeted to 0.65. Substituting this into the formula, the calculated result of the exponential part is +3, indicating that the feedforward weight coefficient dropped sharply to approximately 0.05. At this point, the system automatically reduced the control weight of the feedforward prediction, degenerating into a conservative defensive mode dominated by feedback control. Through this type of S-shaped smooth transition, the system avoids high-frequency switching of the control law caused by small fluctuations in confidence level near the critical point, thus extending the service life of the mechanical actuator.

[0226] Step 703: Use the feedforward weighting coefficient to perform dynamic weighted fusion calculation on the feedforward target frequency offset and the feedback frequency offset to generate a fused frequency modulation command.

[0227] Specifically, the system uses the obtained weighting coefficients to perform a linear weighted superposition of the two control signals. The fusion calculation formula is as follows:

[0228] Δf cmd =w*Δf ff +(1-w)*Δf fb ;

[0229] Where, Δf cmdThe integrated target frequency offset represented by the fused frequency modulation command, w is the feedforward weighting coefficient, and Δf ff Δf is the feedforward target frequency offset. fb This is the feedback frequency offset.

[0230] This fusion command integrates trend predictions based on physics and data with current correction requirements based on structural response. The system converts the frequency value of this fusion command into underlying physical control electrical signals and sends them to the specific actuators.

[0231] Optionally, a semi-actively tuned mass damper can be used as the actuator.

[0232] In this embodiment, the specific physical hardware carrier of the actuator is clearly defined. The semi-active tuned mass damper is typically installed at the top of the wind turbine tower, below the nacelle, or at a specific flange platform in the upper middle part of the tower. Unlike purely passive dampers that rely on a fixed frequency to passively dissipate energy, the semi-active tuned mass damper retains the basic physical structure of the mass block, spring, and damper, but connects a servo control device in parallel to its stiffness element or mass element, which can be adjusted in real time by an external electrical signal.

[0233] Optionally, within the available time margin during the vortex-induced force arrival time window, the spring stiffness or the combined state of the additional mass of the semi-active tuned mass damper is automatically adjusted according to the fusion frequency modulation command.

[0234] Specifically, according to the principles of classical structural dynamics, the natural frequency of a spring oscillator is proportional to the square root of the ratio of stiffness to mass. The system precisely initiates the operation of the underlying servo mechanism before the vortex-induced force reaches its limit within the specified time window.

[0235] In one specific implementation, the system continuously and steplessly adjusts the equivalent spring stiffness of the semi-actively tuned mass damper by changing the magnitude of the excitation current in the electromagnetic spring coil.

[0236] In another specific implementation, the system controls the mechanical locking state of multiple discretely configured additional mass blocks through a micro servo motor, and adjusts the effective mass parameters in stages by changing the number of actual mass blocks participating in the vibration.

[0237] Optionally, by changing the combined state, the operating frequency of the semi-actively tuned mass damper is driven to shift along the target direction, causing the apparent natural frequency of the coupled system consisting of the downstream target tower and the semi-actively tuned mass damper to shift to the target value, thus preventing the vortex-induced locking condition from being established.

[0238] In this embodiment, the system cannot change the physical mass or stiffness of the steel structure of the downstream tower column itself. The downstream tower column body and the internally suspended semi-active tuned mass damper constitute a two-degree-of-freedom or multi-degree-of-freedom vibration coupling system.

[0239] When the system actively shifts the operating frequency of the damper according to the fusion frequency modulation command by changing the stiffness or mass combination, the apparent natural frequency of the entire macroscopic coupled system will also shift in the same direction due to the coupling effect of the mass and stiffness matrices, reaching the set target value. At this time, the vortex shedding frequency of the flow field will not be able to find a matching energy absorption peak point on the structural dynamic response curve. Through the forced misalignment of the physical frequency, the energy exchange channel of the vortex-induced locking condition is cut off, realizing a highly efficient feedforward active damping effect that completes frequency decoupling before vortex-induced vibration actually occurs.

[0240] In some alternative implementations, if the system detects that the frequency offset required by the fusion frequency modulation command is too large, causing physical saturation of the adjustable stiffness range of the semi-actively tuned mass damper, or if the prediction time window is too short, preventing the servo motor from completing the mechanical switching of the additional mass block in time, as a system-level redundancy alternative, the underlying controller will temporarily freeze the current state parameters of the semi-actively tuned mass damper and issue an independent pitch control command to the downstream unit through the wind farm main control network. By actively increasing the blade pitch angle to change the aerodynamic damping characteristics of the wind turbine, the energy dissipation at the aerodynamic level is used to compensate for the frequency decoupling failure at the structural level, thus preventing the wind turbine from becoming unstable and failing under extreme operating conditions.

[0241] Example 8: This example details a system offline evolution mechanism independent of the real-time control closed loop. As an optional implementation, the event-triggered online correction step based on a pre-built physical-data hybrid model specifically includes:

[0242] Step 801: After the vortex-induced force reaches the time window, the measured vortex shedding frequency is identified from the vibration monitoring data of the downstream target tower structure by means of spectrum peak detection.

[0243] Specifically, after the predicted vortex-induced force event has ended, the system extracts vibration monitoring data of the downstream target tower structure that occurred during that historical time period. The vibration response signal is then frequency-domain transformed using a Fast Fourier Transform algorithm, and the peak coordinates with the highest energy concentration are searched in the spectrum. The frequency coordinates corresponding to this peak represent the physical dominant frequency of the flow field wake vortex acting on the downstream tower. Through this type of spectral peak detection based on post-hoc facts, the system obtains baseline fact labels for evaluating the accuracy of the feedforward prediction model.

[0244] Step 802: Calculate the amplitude deviation between the measured vortex shedding frequency and the center value of the vortex shedding frequency interval, and determine whether the measured vortex shedding frequency falls within the vortex shedding frequency interval in order to calculate the interval coverage index.

[0245] In this embodiment, the system evaluates the quality of the initial prediction from two dimensions. The first dimension is absolute accuracy, where the system calculates the geometric distance between the actual frequency and the absolute center point of the prediction interval. The second dimension is probabilistic reliability, where the system uses Boolean logic to check whether the frequency coordinates are enveloped within the upper and lower limits of the frequency interval output by the system in the early stage. If they fall within the interval, it is determined to be a covered event; if they are outside the interval, it is determined to be an uncovered event.

[0246] The accuracy assessment logic and calculation formula are as follows:

[0247] E f =|f actual -f pred_center |;

[0248] Among them, E f For amplitude deviation, f actual To measure the vortex shedding frequency, f pred_center is the center value of the vortex shear frequency range, and || is the absolute value operator.

[0249] Step 803: When the amplitude deviation exceeds the preset allowable deviation threshold, or when the interval coverage index shows an uncovered event, extract the operating condition parameters that triggered the current event, the vortex shedding intensity characteristics, and the measured vortex shedding frequency to form new training samples.

[0250] When constructing supervisory labels for newly added training samples, the system uses the Strouhal relation to back-calculate the measured eddy current frequency into the actual downstream inflow velocity, i.e., U. actual =(f actual *D t ) / S t Calculate the difference between the actual incoming flow velocity and the baseline predicted wake velocity deficit output by the engineering wake analysis model under the current operating conditions, i.e., ΔU. label =U actual -U d The difference is used as the speed correction label for the data-driven residual compensation module, and together with the corresponding operating condition parameter vector and vortex shedding intensity feature vector, it forms a complete new training sample.

[0251] Specifically, this embodiment employs an event-triggered mechanism for model self-learning. Unlike traditional online learning frameworks that update the model at each time step, this embodiment uses an event-triggered strategy, initiating the learning mechanism only when the model's prediction deviation exceeds a preset threshold. The preset deviation tolerance threshold can be set to 0.05 Hz based on specific engineering tolerances. When a learning event is triggered, the system backtracks from the historical database to obtain the input feature vector that led to the misjudgment, uses the subsequently identified frequency as a supervision label to construct training samples, and stores them in a dynamic experience replay buffer.

[0252] Step 804: Using the newly added training samples, the residual compensation parameters in the pre-built physical-data hybrid model are locally updated using an incremental learning approach.

[0253] In this embodiment, the system extracts a small batch of data, including newly added training samples, from the dynamic experience replay buffer and starts the background offline training engine. During the update process, the system freezes the underlying parameters of the physical engineering analytical model and only allows the neural network weights of the data-driven residual compensation module to participate in the backpropagation calculation. A momentum-based stochastic gradient descent algorithm with a very small incremental learning rate is used to fine-tune the local weights of the neural network.

[0254] Optionally, the attenuation ratio of the downstream target tower structure vibration monitoring data before and after the actuator adjusts the apparent natural frequency of the downstream target tower can be extracted as an effectiveness evaluation indicator.

[0255] Furthermore, in addition to revising the aerodynamic prediction model, the system also performs closed-loop self-calibration of the effectiveness of the structural dynamics defense strategy. The system extracts historical baseline vibration amplitudes under the same wind speed conditions when active damping is not activated and compares them with the actual vibration amplitudes after the actuators are activated for frequency offset vibration damping. The measurement results generated in this comparison process are defined as effectiveness evaluation indicators, and their specific calculation formula is as follows:

[0256] R eff =(A baseline -A actual ) / A baseline ;

[0257] Among them, R _eff As an indicator for effectiveness evaluation, A baseline A represents the baseline vibration amplitude when frequency modulation is not performed. actual The actual vibration amplitude after frequency modulation intervention.

[0258] Optionally, the redundancy of the actual frequency offset corresponding to the fusion frequency modulation command relative to the minimum offset required to exit the preset frequency locking reduction speed range can be extracted as an economic evaluation index.

[0259] In this embodiment, the economic assessment aims to prevent over-defense and energy waste in the system. Physical actuators consume power and experience mechanical wear to reach the target frequency. The system calculates the redundancy by dividing the final output actual execution offset by the minimum physical distance necessary to exit the danger zone. The calculation formula is as follows:

[0260] R eco =Δf actual / Δf min ;

[0261] Among them, R eco As an economic evaluation indicator, Δf actual Δf represents the actual frequency offset of the actuator. min The minimum offset required to exit the preset frequency-locked reduced speed range.

[0262] Optionally, when the effectiveness evaluation index characterizes the attenuation ratio as lower than a preset benchmark, the upper and lower bounds of the preset frequency-locked reduced speed range are extended outward to increase the conservatism of the criterion.

[0263] For example, if, after consuming actuator resources for frequency modulation, the final vibration attenuation ratio is found to be less than 30%, it indicates that the originally set fixed frequency locking range is too narrow, causing the system's calculated safe zone to still be subject to edge intrusion from vortex-induced resonance energy. In this case, the system uses a step-gradient expansion algorithm to automatically move the upper limit of the preset frequency locking reduced velocity range upward by a preset step size and the lower limit downward by a preset step size, so that subsequent frequency modulation commands adjust the structure to a wider safe range.

[0264] Optionally, when the redundancy represented by the economic evaluation index is higher than the preset benchmark, the upper and lower bounds of the preset frequency-locked reduced speed range are narrowed inward to optimize the vibration prevention economy, thereby achieving closed-loop self-calibration of the upper and lower bounds of the preset frequency-locked reduced speed range.

[0265] In this embodiment, if the economic evaluation index shows excessive redundancy, such as exceeding 1.5 times, it indicates that the system has incurred a high mechanical adjustment cost to ensure safety. Provided the effectiveness evaluation index meets the standard, the system activates the economic recovery mechanism, converging the upper and lower boundaries of the preset frequency-locked reduction speed range towards the center, thus reducing the frequency bandwidth deemed dangerous.

[0266] In some optional implementations, to avoid drastic oscillations in the closed-loop self-calibration process due to a single extreme anomaly, a forgetting factor that decays over time is introduced into the calculation of expanding or contracting the threshold boundaries. The calculated target correction is then weighted and averaged with historically accumulated steady-state thresholds before being updated to the system's activity configuration file.

[0267] This invention senses the flow field through the high-frequency vibration response of the upstream wind turbine, extracts vortex shedding intensity characteristics such as energy concentration, and combines a dimensionless corrected Gaussian wake physical baseline with residual compensation using a data-driven module. This solves the problems of insufficient prediction accuracy of purely physical empirical models and the inability of single-point wind measurement to characterize the vortex shedding coherence of the flow field.

[0268] Furthermore, a panoramic safety margin optimization mechanism based on multi-source environmental excitation is introduced to avoid the drawback of blindly avoiding current frequency band locking, which can easily lead to secondary resonance. By proactively calculating the minimum absolute distance between the target frequency and environmental excitation sources such as wave and blade passing frequencies, the system adaptively selects the safest direction with the largest margin for physical offset.

[0269] By employing a logic function with a smooth S-shaped transition, the control weights of feedforward prediction and real-time vibration feedback are dynamically allocated. This enables the system to adopt a conservative defense mode based on feedback when predictions are inaccurate due to extreme flow fields. This mode allocates weights between the timeliness of active vibration reduction and the safety of the underlying system's defense.

[0270] It should be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A method for reducing the vibration of wind turbine tower columns to mitigate the effects of wake vortices, characterized in that, include: Acquire multi-source operating status data of upstream wind turbines in the wind farm, high-frequency response data of upstream wind turbine structures, and vibration monitoring data of downstream target tower structures, and load pre-configured natural frequency parameters of downstream target tower structures; Based on a pre-built physical-data hybrid model, using multi-source operating status data of upstream wind turbines and high-frequency response data of upstream wind turbine structures, the frequency range of vortex shedding and the prediction confidence level of the wake vortex propagating to the downstream target tower are predicted, and the arrival time window of vortex excitation force is calculated. Risk assessment is conducted based on the vortex shedding frequency range, prediction confidence level, and the natural frequency parameters of the downstream target tower structure to obtain the frequency locking risk assessment results. When the frequency locking risk assessment results meet the preset triggering conditions, a fusion frequency modulation command is generated based on the predicted confidence level and the vibration monitoring data of the downstream target tower structure. Within the vortex-induced force arrival time window, the actuator is driven to adjust the apparent natural frequency of the downstream target tower column according to the fusion frequency modulation command.

2. The method according to claim 1, characterized in that, Acquire multi-source operating status data of upstream wind turbines in the wind farm, high-frequency response data of upstream wind turbine structures, and vibration monitoring data of downstream target tower structures, including: Collect wind speed, nacelle yaw angle, blade pitch angle, generator speed and thrust coefficient at the height of the upstream wind turbine hub to obtain multi-source operating status data of the upstream wind turbine; The crosswind and downwind vibration acceleration signals at the bottom of the upstream wind turbine nacelle were collected to obtain high-frequency response data of the upstream wind turbine structure. The triaxial structural vibration response signal of the upper part of the downstream target tower column was collected to obtain the vibration monitoring data of the downstream target tower column structure; Using the sampling clock of the upstream wind turbine structure high-frequency response data as a reference, the synchronous timing signal obtained from the wind farm communication network is used to align the upstream wind turbine multi-source operating status data, the upstream wind turbine structure high-frequency response data, and the downstream target tower structure vibration monitoring data.

3. The method according to claim 1, characterized in that, Based on a pre-built physical-data hybrid model, using multi-source operating status data and high-frequency response data of upstream wind turbines, the prediction of the vortex shedding frequency range and confidence level at the downstream target tower is specifically included: By using a pre-built engineering wake analysis model to process multi-source operating status data of upstream wind turbines, the basic predicted value of wake velocity loss is obtained. Extract the vortex shedding intensity feature vector characterizing the flow field properties from the high-frequency response data of the upstream wind turbine structure; The baseline predicted value of the wake velocity deficit and the vortex shedding intensity feature vector are jointly input into a pre-built data-driven residual compensation module for online correction to obtain the compensated downstream flow velocity. Based on the compensated downstream flow velocity, and combined with the calculated model's cognitive uncertainty and input uncertainty, the vortex shedding frequency range and prediction confidence level are obtained.

4. The method according to claim 3, characterized in that, The process of using a pre-built engineering wake analysis model to process multi-source operating status data of upstream wind turbines to obtain the basic predicted value of wake velocity loss specifically includes: Extract the current incoming wind speed, thrust coefficient, and lateral offset determined based on the nacelle yaw angle from the multi-source operating status data of the upstream wind turbine; The inflow velocity at the downstream target tower section was calculated using the dimensionless modified Gaussian wake analytical model. The calculation results were used as the basic predicted value of wake velocity deficit. Among them, the Gaussian wake analytical model constructs an analytical calculation formula that satisfies the conservation of physical dimensions based on the ratio of the thrust coefficient to the square of the dimensionless width.

5. The method according to claim 3, characterized in that, Extracting vortex shedding intensity feature vectors characterizing the flow field properties from high-frequency response data of upstream wind turbine structures, including: The crosswind acceleration time history signal in the high-frequency response data of the upstream wind turbine structure is divided into multiple window segments with overlapping ratios by sliding window segmentation according to the preset time window length. Short-time spectrum analysis is performed on the signal of each window segment to extract the dominant frequency value corresponding to the peak value of the spectrum in each window segment, the energy concentration degree composed of the ratio of spectral energy in the neighborhood of the peak frequency to the total spectral energy, and the rate of change of the dominant frequency value of adjacent window segments. The extracted dominant frequency, energy concentration, and rate of change are combined in chronological order to form a vortex shedding intensity feature vector.

6. The method according to claim 3, characterized in that, The predicted wake velocity deficit and the vortex shedding intensity eigenvector are jointly input into a pre-built data-driven residual compensation module for online correction, resulting in the compensated downstream flow velocity, including: Extract the operating condition parameter vector containing the current incoming wind speed, thrust coefficient, and yaw angle from the multi-source operating status data of the upstream wind turbine; The operating condition parameter vector and the vortex shedding intensity characteristic vector are jointly input into the pre-built data-driven residual compensation module to calculate the velocity correction caused by the current complex flow field effect. The downstream inflow velocity after compensation is obtained by algebraically superimposing the basic predicted value of the wake velocity loss with the velocity correction.

7. The method according to claim 3, characterized in that, Based on the compensated downstream flow velocity, and combining the calculated model's cognitive uncertainty and input uncertainty, the vortex shedding frequency range and prediction confidence level are obtained, including: Based on the pre-configured Strauhal number and the reference cross-section diameter of the downstream tower column, the compensated downstream inflow velocity is linearly mapped to the frequency space to obtain the estimated value of the vortex shedding frequency point. In the inference phase of the data-driven residual compensation module, multiple random forward propagations with a random dropout mechanism are performed to statistically determine the model's cognitive uncertainty, and the input measurement uncertainty is calculated based on the pre-configured sensor accuracy specifications. The combined model cognitive uncertainty and input measurement uncertainty constitute the comprehensive frequency standard deviation, and the estimated vortex shedding frequency point is expanded on both sides using the preset coverage factor and the comprehensive frequency standard deviation to generate the vortex shedding frequency range. The probability value corresponding to the preset coverage factor is calculated based on the cumulative distribution function of the standard normal distribution, and is used as the prediction confidence level.

8. The method according to claim 1, characterized in that, A risk assessment was conducted based on the vortex shedding frequency range, predicted confidence level, and the natural frequency parameters of the downstream target tower structure, resulting in a frequency locking risk assessment, including: The upper and lower boundary values ​​of the vortex shedding frequency range are converted into the incoming flow velocity boundary. Combined with the natural frequencies of each mode in the natural frequency parameters of the downstream target tower structure and the diameter of the tower section, the fluctuation range of the corresponding reduced velocity values ​​is calculated respectively. Calculate the intersection ratio between the fluctuation range of each order reduction velocity value and the preset frequency-locking reduction velocity range to determine the frequency-locking risk level; When the frequency locking risk level meets the preset triggering conditions, the threatened mode order is dynamically identified in the natural frequency of each mode according to the principle of being closest to the center value of the preset frequency locking reduction speed interval. The threatened mode order and its corresponding intrinsic frequency are used as the results of frequency locking risk assessment.

9. The method according to claim 1, characterized in that, Based on the predicted confidence level and vibration monitoring data of the downstream target tower structure, a fused frequency modulation command is generated, including: Based on the multi-source operating status data of upstream wind turbines, multi-source environmental excitation frequencies, including the frequency of wind turbine blade passage, are extracted. Calculate the minimum frequency offset required to shift the natural frequencies corresponding to the threatened mode order to higher and lower frequencies, respectively, to exit the preset frequency-locked reduction velocity range. The minimum absolute distance between the natural frequency after the minimum offset frequency and the excitation frequency of each multi-source environment is calculated and defined as the safety margin in the high-frequency direction and the safety margin in the low-frequency direction, respectively. The direction corresponding to a larger safety margin is selected as the target frequency modulation direction; Based on the target frequency modulation direction, the predicted confidence level, and the vibration monitoring data of the downstream target tower structure, a fused frequency modulation command is generated.

10. The method according to claim 3, characterized in that, The event-triggered online correction steps based on a pre-built physical-data hybrid model include: After the vortex-induced force reaches its arrival time window, the measured vortex shedding frequency is identified from the vibration monitoring data of the downstream target tower structure by detecting the spectral peak. Calculate the amplitude deviation between the measured vortex shedding frequency and the center value of the vortex shedding frequency interval, and determine whether the measured vortex shedding frequency falls within the vortex shedding frequency interval in order to calculate the interval coverage index. When the amplitude deviation exceeds the preset allowable deviation threshold, or when the interval coverage index shows an uncovered event, the operating condition parameters that triggered the current event, the vortex shedding intensity characteristics, and the measured vortex shedding frequency are extracted to form new training samples. Using newly added training samples, the residual compensation parameters in the pre-built physical-data hybrid model are locally updated using an incremental learning approach.

11. The method according to claim 1, characterized in that, Calculate the arrival time window of the vortex-induced force, including: Extract the inflow wind speed at the hub height at the current moment from the multi-source operating status data of the upstream wind turbine; Obtain the pre-configured projection spacing of upstream and downstream turbines along the prevailing wind direction in the wind farm layout, and the pre-configured convection velocity ratio characterizing the wake structure with the average wind field propagation speed. Divide the product of the pre-configured projection spacing and the pre-configured convection velocity ratio and the incoming wind speed at the hub height to estimate the time delay of the wake vortex propagating from the upstream wind turbine to the downstream tower section. Based on the time delay, the predicted time lead from the current moment to the actual impact of the change on the downstream target tower column is determined, and this lead is used as the time window for the arrival of the vortex-induced force.

12. The method according to claim 1, characterized in that, Within the vortex-induced force arrival time window, the actuator is driven to adjust the apparent natural frequency of the downstream target tower column according to the fused frequency modulation command, including: A semi-actively tuned mass damper is used as the actuator; Within the available time margin of the vortex-induced force arrival time window, the spring stiffness or the combined state of the additional mass of the semi-active tuned mass damper is automatically adjusted according to the fusion frequency modulation command. By changing the combined state, the operating frequency of the semi-actively tuned mass damper is driven to shift along the target direction, causing the apparent natural frequency of the coupled system consisting of the downstream target tower and the semi-actively tuned mass damper to shift to the target value, thus preventing the establishment of vortex-induced locking conditions.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 12.

Images