A method for constructing a wind power blade clearance nonlinear dynamic prediction model

Abstract

The application discloses a kind of wind power blade clearance nonlinear dynamic prediction model construction methods, it is related to wind power blade monitoring technical field, including the following steps: deployment multi-frequency phased array laser radar matrix, the spatial wind speed data of wind farm front row unit wake region is collected, constructs three-dimensional instantaneous speed voxel library;Based on three-dimensional instantaneous speed voxel library, the profile phase convolution processing is executed to speed voxel under continuous time, extract wake rotation phase information, and map as standard vortex phase spectrum.The application introduces multi-frequency phased array laser radar, phase convolution and high-dimensional tensor analysis, accurately identifies the spatial coherence characteristics of wake vortex, combined with layered integration and multi-scale resonance detection, extracts wake phase superposition path, and constructs wake phase superposition exponential dynamic correction prediction model wind pressure boundary, significantly improves the prediction sensitivity and adaptability of wind power blade yawing risk, effectively solves the problem of underestimating and inaccurate prediction of wake coupling in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine blade monitoring, and particularly relates to a wind turbine blade clearance nonlinear dynamic prediction model construction method. BACKGROUND

[0002] The wind turbine blade clearance nonlinear dynamic prediction model construction refers to establishing a mathematical modeling and reasoning mechanism that can predict the minimum distance (i.e. clearance) between the blade and the surrounding structure (such as the tower, the ground, the power transmission line, etc.) at each moment based on the nonlinear motion characteristics of the wind turbine blade under the influence of multiple factors such as wind speed, wind direction, rotating speed, blade angle, etc. during operation. The model usually integrates multi-source dynamic perception data (such as wind speed sensors, blade attitude sensors, radar monitoring, etc.), combines nonlinear dynamics theory, time series prediction algorithm or machine learning technology, and constructs a dynamic trajectory prediction function reflecting the change of the spatial position of the blade with time, so as to realize early warning and decision support for potential collision risk under extreme working conditions, and improve the safety and operation stability of the wind power system.

[0003] The prior art has the following disadvantages:

[0004] In the dense arrangement scene of the wind farm, multiple front-row wind turbine generators may form a wake vortex structure with a specific rotational phase relationship during operation. When these vortices are phase superimposed in space, periodic high-frequency wind pressure disturbances are easily induced in the operating height range of the rear-row generator blades, and then the lateral resonance yawing phenomenon of the rear-row blades occurs. Since the prior art generally does not model the spatial wind pressure coupling effect caused by the interaction of the wakes between the generators in the process of constructing the dynamic prediction model of the wind turbine blade clearance, the prediction result of the maximum yawing amplitude of the blade is systematically underestimated, it is difficult to accurately identify the risk that the blade may cross the clearance safety boundary under extreme working conditions, and there is a safety hazard of touching the tower, the ground or other adjacent structures.

[0005] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a wind turbine blade clearance nonlinear dynamic prediction model construction method, by introducing a multi-frequency phased array laser radar to obtain high-resolution three-dimensional wind speed voxel data, combining continuous frame phase convolution to extract vortex phase spectrum, and using high-dimensional phase difference tensor and sparse spectrum aggregation strategy, the spatial coherence characteristics of the wake vortex are accurately identified; further, by integrating the layered height, the phase energy density profile is generated, combined with multi-scale resonance detection to identify the potential resonance path, the wake phase superposition index is constructed, the wind pressure boundary condition of the wind turbine blade clearance prediction model is dynamically corrected, and the response ability of the model to extreme yawing risk is effectively improved, to solve the problems in the above background technology.

[0007] In order to achieve the above purpose, the present application provides the following technical scheme: a wind turbine blade clearance nonlinear dynamic prediction model construction method, comprising the following steps:

[0008] S100, deploy a multi-frequency phased array laser radar matrix to cover the wake influence area of the front row of wind turbines in the wind farm, and collect spatial wind speed data of the wake influence area; based on the collected spatial wind speed data, a three-dimensional instantaneous speed voxel library is constructed to represent the speed distribution characteristics of the wake vortex structure at different spatial positions;

[0009] S200, based on the three-dimensional instantaneous speed voxel library, a continuous frame space correlation solver is called to perform profile phase convolution processing on the speed voxel at continuous time, extract the rotational phase information of the vortex structure in the wake of each wind turbine, and map the extracted rotational phase information to a standardized vortex phase spectrum;

[0010] S300, based on the vortex phase spectrum corresponding to adjacent wind turbines, a high-dimensional phase difference tensor is constructed, which is used to represent the phase coupling relationship of the wake vortex structures of multiple wind turbines at different spatial positions and height ranges;

[0011] S400, perform sparse spectrum aggregation processing on the high-dimensional phase difference tensor, extract the phase coupling channels with the same frequency characteristics in the frequency domain, identify the wake vortex structures with spatial coherence characteristics between multiple wind turbines, and obtain a set of wake coherent structures;

[0012] S500, project the set of wake coherent structures to a preset layered height index surface, and perform adaptive phase weight integration processing on the vortex phase spectrum information in each layered height to calculate the phase energy density profile corresponding to each height layer;

[0013] S600, based on the phase energy density profile, construct a phase coupling curve across the layered height, and input the phase coupling curve into a multi-scale resonance detector to identify the resonance response behavior caused by the wake phase superposition, and output the wake phase superposition response amplitude;

[0014] S700, based on the wake phase superposition response amplitude and phase coupling curve, generates a wake phase superposition index through a normalized peak operator, and dynamically inputs the wake phase superposition index as a correction factor into the wind turbine blade clearance nonlinear dynamic prediction model to adjust the wind pressure input boundary conditions in the wind turbine blade clearance nonlinear dynamic prediction model, so as to improve the response sensitivity and prediction coverage of the prediction model to extreme yaw behavior.

[0015] Preferably, step S100 includes:

[0016] A multi-frequency phased array lidar is deployed in the wind farm, consisting of multiple lidar units whose transmission frequency and direction can be independently controlled.

[0017] The lidar array is controlled to perform a three-dimensional spatial synchronous scan within the wake influence area, collect wind speed vector data, and generate a spatial wind speed vector field.

[0018] The spatial wind speed vector field is mapped to a unified three-dimensional coordinate system and divided into multiple three-dimensional voxel units containing wind speed vector information to form a three-dimensional instantaneous velocity voxel set.

[0019] The three-dimensional instantaneous velocity voxel set is time-series numbered and stored to construct a three-dimensional dynamic velocity voxel data sequence that reflects the wake evolution process.

[0020] Preferably, step S200 includes:

[0021] The velocity voxel data of consecutive time frames are extracted from the three-dimensional instantaneous velocity voxel library to form a velocity voxel time series, and spatial registration is performed to ensure inter-frame consistency.

[0022] A set of profile slices was constructed in the wake axis region of the wind turbine based on time series, and the rotational phase distribution of the profile was extracted by two-dimensional Fourier transform and Hilbert algorithm.

[0023] The multi-frame profile rotation phase map is jointly processed to generate the phase evolution trajectory at a continuous time scale, and normalization and bandwidth limiting processing are performed.

[0024] The phase trajectory is mapped to a standardized vortex phase spectrum with the wind turbine rotor surface as a reference.

[0025] Preferably, step S300 includes:

[0026] Standardized vortex phase spectra are constructed for adjacent wind turbine units, and unified spatial grids and height layers are processed.

[0027] Within the same height index layer, calculate the phase difference between corresponding voxel regions and construct a data structure for joint mapping of three-dimensional space and time.

[0028] Based on spatial location, height layer, time frame, and unit pair dimensions, a high-dimensional phase difference tensor is constructed to record the quantized values ​​of phase difference.

[0029] Normalization and noise filtering are performed on the high-dimensional phase difference tensor to improve the stability and accuracy of phase coupling representation.

[0030] Preferably, step S400 includes:

[0031] Perform a multi-scale fast Fourier transform on the high-dimensional phase difference tensor to construct a normalized frequency domain phase difference spectrum tensor;

[0032] Sparse reconstruction based on L1 regularization term constraint is performed on the frequency domain phase difference spectrum tensor to preserve stable main frequency characteristics and filter out interference frequency components.

[0033] Spatial regions are clustered based on frequency co-occurrence and phase similarity to identify phase-coupled channels that satisfy the same frequency characteristics;

[0034] Phase coupling channels that satisfy frequency consistency, spatial connectivity, and phase smoothness are merged to form a wake coherent structure set.

[0035] Preferably, step S500 includes:

[0036] Construct a layered height index surface that matches the operating range of wind turbine blades, and project the wake coherent structure set onto each height layer;

[0037] Extract the vortex phase spectrum information from each height index surface and generate a weighted vector set containing position, frequency and phase features;

[0038] Perform adaptive phase weight integration within each altitude layer to calculate the phase energy density;

[0039] The obtained phase energy density is normalized and sorted to construct a phase energy density profile that reflects the intensity of the disturbance.

[0040] Preferably, step S600 includes:

[0041] Phase coupling curves across different layer heights are constructed based on the phase energy density profile to represent the phase energy correlation paths between layers at different heights;

[0042] The phase coupling curves are organized into a time series curve cluster in the time dimension, and the phase feature parameters are extracted and their evolution trend is modeled.

[0043] The phase coupling curve is input into a multi-scale resonant detector to perform frequency domain analysis and compare it with the blade's natural frequency to identify potential resonance paths.

[0044] Based on the identification results, frequency and stability indicators are extracted, and the wake phase superposition response amplitude is generated to quantify the disturbance intensity.

[0045] Preferably, step S700 includes:

[0046] By fusing the wake phase superposition response amplitude and phase coupling curve, a set of disturbance characteristics is constructed.

[0047] The normalized peak operator is invoked to normalize the wake phase superposition response amplitude in the disturbance feature set, generating the wake phase superposition index.

[0048] The wake phase superposition index is used as a correction factor to dynamically input the prediction model, which is used to adjust the wind pressure boundary conditions and turbulence input parameters.

[0049] The model performance is evaluated based on the prediction results, and the evaluation feedback is used for parameter optimization and iterative updates of the wake phase superposition index.

[0050] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0051] This invention introduces a multi-frequency phased array lidar matrix to acquire high-resolution three-dimensional wind speed voxel data. It then extracts the vortex rotation phase spectrum using a continuous frame profile phase convolution method, and employs a high-dimensional phase difference tensor construction and sparse spectrum aggregation strategy to accurately identify the spatial coherence characteristics of the wake vortex structure. Based on this, adaptive phase integration is performed using a layered height index surface to generate a phase energy density profile. A multi-scale resonance detection mechanism is then used to identify wake phase superposition paths that may induce lateral yaw resonance in wind turbine blades. Finally, a wake phase superposition index is constructed and input as a dynamic correction factor into the nonlinear dynamic prediction model of wind turbine blade clearance. This effectively enhances the model's ability to perceive extreme yaw risks and its dynamic adaptability, overcoming the common problems in existing technologies of underestimating the coupling effect of wake disturbances and inaccurate prediction of yaw amplitude. This significantly improves the operational safety and prediction reliability of wind turbines in complex wake interference environments. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0053] Figure 1 This is a flowchart of a method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to the present invention. Detailed Implementation

[0054] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0055] This invention provides, for example Figure 1 The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance, as shown, includes the following steps:

[0056] S100: Deploy a multi-frequency phased array lidar matrix to cover the wake influence area of ​​the front wind turbine units in the wind farm and collect spatial wind speed data of the wake influence area; Based on the collected spatial wind speed data, construct a three-dimensional instantaneous velocity voxel library to characterize the velocity distribution characteristics of the wake vortex structure at different spatial locations.

[0057] In this embodiment, to achieve high-precision perception of the spatial velocity distribution characteristics of the wake region of the front-row wind turbines in a wind farm, a three-dimensional instantaneous velocity voxel library is constructed, specifically including the following steps:

[0058] A multi-frequency phased array lidar is deployed in the wind farm. The lidar array consists of multiple lidar units whose transmission frequency and direction can be independently controlled. Each lidar unit is installed on a mounting platform on the ground around the wind farm or on top of the tower. Each lidar unit can perform synchronous scanning in three-dimensional space at high frequency. By adjusting its transmission angle, wavelength, and phased beam direction, multi-view coverage of different spatial levels within the influence area of ​​the wake of the preceding wind turbines is achieved, forming an observation field with spatial overlap, thus constructing a complete and resolution-adjustable three-dimensional scanning network structure. During operation, the lidar array continuously transmits scanning signals and receives scattered echo data. Based on the laser flight time and frequency shift effect, wind speed vector data at different spatial locations within the target area are obtained. Specifically, by jointly processing the data collected by multiple lidar units, performing spatiotemporal matching of reflection points and viewpoint fusion, the wind speed components in the vertical, tangential, and radial directions of each measurement point are calculated, generating a spatial wind speed vector field. This spatial wind speed vector field not only includes information on wind speed magnitude, but also retains the trend of local wind speed direction changes, which helps to identify the core features of the vortex rotation structure in the wake.

[0059] The obtained spatial wind speed vector fields are mapped to a unified three-dimensional spatial coordinate system. Based on the established voxel partitioning strategy, the entire wake influence area is divided into several regular spatial unit voxels. Each voxel contains wind speed vector information from one or more laser scanning points. Spatial filling operations are performed on uncovered areas using a data interpolation algorithm to ensure the continuity and integrity of the spatial velocity data. After this step, a three-dimensional instantaneous velocity voxel set containing velocity vector attributes is formed. Its resolution can be dynamically adjusted as needed, for example, by setting densely distributed areas according to the sweep height of different wind turbine blades to improve vortex identification accuracy. This three-dimensional instantaneous velocity voxel set is time-series numbered and stored to establish a high-frequency dynamic data foundation for subsequent phase extraction and wake interferometry analysis. Specifically, wind speed voxels are snapshotted at fixed time intervals, with each frame of voxel data corresponding to a timestamp, thereby constructing a three-dimensional dynamic data sequence reflecting the wake's evolution over time. This data sequence can not only be used for real-time instantaneous assessment of wake disturbances, but also be input into subsequent spatiotemporal solvers for rotating phase identification and coherent structure extraction, significantly improving the modeling accuracy of the entire wind turbine blade clearance prediction model for high-frequency disturbances and resonance risks.

[0060] S200, based on the three-dimensional instantaneous velocity voxel library, calls the continuous frame spatiotemporal correlation solver to perform profile phase convolution processing on the velocity voxels at continuous time points, extracts the rotation phase information of the vortex structure in the wake of each wind turbine, and maps the extracted rotation phase information into a standardized vortex phase spectrum.

[0061] To achieve accurate extraction and standardized representation of the rotational phase information of the wake vortex structure of wind turbines in wind farms, based on the established three-dimensional instantaneous velocity voxel library, continuous frame spatiotemporal correlation calculation and profile phase convolution processing are performed, specifically including the following steps:

[0062] Velocity voxel frames from multiple consecutive moments are extracted in chronological order from the constructed 3D instantaneous velocity voxel library to form a velocity voxel time series with a defined time step. Each frame of velocity voxels contains the wind speed vector distribution across the entire wake space at a specific time point. Based on this, a continuous frame spatiotemporal correlation solver is invoked to perform spatial registration processing on velocity voxels between adjacent time frames to eliminate local positional offsets caused by radar scanning angle errors and generator vibrations, ensuring the consistency of the time series in spatial structure.

[0063] A profile region within the spatial area containing the wake axis of each wind turbine is selected. Based on continuous frame velocity voxel data, a set of profile slices aligned with the wake direction is constructed. For each set of profile slices, a derivative filtering method based on the rate of change of the wind speed vector field is used to identify the periodic rotating structure of velocity perturbations in the profile. By introducing a two-dimensional Fourier transform and a Hilbert phase extraction algorithm, convolution calculations are performed on the velocity distribution in the profile slices to obtain the local rotating phase distribution map of the wake profile at each time frame.

[0064] The rotating phase maps of each time frame profile are jointly processed to construct the phase evolution trajectory over a continuous time scale. By employing a temporal envelope extraction algorithm and a phase unwrapping mechanism, numerical discontinuities caused by phase jumps are eliminated, thereby recovering the spatial-temporal rotational behavior trajectory of each wake vortex. For the rotating phase data extracted from wake profiles of different wind turbines, a normalized amplitude modulation function and a spectral bandwidth limiting mechanism are further applied to eliminate influencing factors such as equipment noise and observation angle differences, enhancing the comparability and stability of phase characteristics.

[0065] The resulting rotating phase spectrum is uniformly mapped to a standardized vortex phase spectrum. This vortex phase spectrum uses the wind turbine rotor surface as a reference plane and corresponds to the dominant frequency component, phase center, and phase amplitude of the wake rotation structure for each unit of time, forming a characteristic spectral representation of the wake rotation behavior. This standardized representation facilitates subsequent phase difference analysis, coupled structure identification, and spatial interference modeling among multiple wind turbine wakes. It is a key intermediate variable for modeling high-frequency disturbances in the nonlinear dynamic prediction model of wind turbine blade clearance, significantly improving the model's sensitivity and accuracy in detecting lateral yaw resonance risks.

[0066] S300. A high-dimensional phase difference tensor is constructed based on the vortex phase spectrum corresponding to adjacent wind turbine units. The high-dimensional phase difference tensor is used to characterize the phase coupling relationship of the wake vortex structure of multiple wind turbine units in different spatial locations and height ranges.

[0067] To express the phase interference and coupling characteristics among the wake vortex structures of multiple wind turbines in terms of spatial dimension and height, a high-dimensional phase difference tensor needs to be constructed based on the extracted standardized vortex phase spectrum to characterize the wake phase correlation between wind turbines. Specifically, the following steps are included:

[0068] Two or more spatially adjacent wind turbines within a wind farm are selected, and standardized vortex phase spectra are established for their respective wake regions. Each vortex phase spectrum is extracted from the aforementioned steps and includes the time-varying rotating phase dominant frequency component, phase angle, and amplitude index. It is then uniformly meshed according to a pre-defined spatial voxel division and height hierarchy. This processing ensures the comparability of wake data from different wind turbines within the same spatial reference frame and establishes a data foundation for phase difference calculation.

[0069] For the corresponding voxel regions within the same height index layer of each wind turbine, the differences in their phase spectra are calculated sequentially over time to obtain the phase difference value between multiple voxel pairs per unit time. This phase difference calculation employs a periodic mapping algorithm to automatically handle the 2π jump problem, ensuring the continuity and true physical meaning of the phase difference value. The phase differences of all voxel pairs are indexed by coordinate position, time series, and height number, thereby constructing a data structure that jointly maps three-dimensional space and time.

[0070] Building upon the aforementioned data structure, multiple dimensions, including height layer, spatial location, time frame number, and wind turbine pair number, are introduced as tensor construction dimensions to generate a high-dimensional phase difference tensor describing the phase difference relationship of the wakes of adjacent wind turbines. Each tensor unit of this high-dimensional phase difference tensor records the quantized value of the phase difference of the vortex structure in the wakes of different wind turbines at a specified location and time. This tensor structure possesses strong expressive power and can be used to identify highly correlated phase resonance modes and the spatial propagation paths of lateral periodic disturbances.

[0071] Finally, the constructed high-dimensional phase difference tensor is normalized and noise filtered to improve the robustness and discriminativeness of the phase difference representation. Specifically, this includes using a high-pass time-series filter to remove short-period disturbances, introducing a spatial consistency constraint function to suppress constant phase difference values, and performing spatiotemporal interpolation-based repair operations on missing data dimensions in the tensor. The resulting high-dimensional phase difference tensor, as a mathematical expression of the coupling relationship between wake phase structures, can be directly used in subsequent wake coherence structure extraction, phase convergence analysis, and disturbance gain calculation for yaw prediction models, providing crucial modeling support for improving the sensitivity of wind turbine blade clearance dynamic prediction models to coupled disturbances.

[0072] S400: Perform sparse spectrum aggregation processing on the high-dimensional phase difference tensor, extract phase coupling channels with the same frequency characteristics in the frequency domain, identify wake vortex structures with spatial coherence characteristics between multiple wind turbine units, and obtain a set of wake coherent structures.

[0073] To extract vortex structures with co-frequency characteristics and spatial coherence from the phase relationships of wakes from multiple wind turbine generators, and to further improve the modeling capability of the nonlinear dynamic prediction model for wind turbine blade clearance for complex wake interference modes, sparse spectrum aggregation processing is performed based on the constructed high-dimensional phase difference tensor, ultimately obtaining a set of coherent wake structures. This process includes the following steps:

[0074] A frequency domain transformation is performed on the high-dimensional phase difference tensor to identify implicit periodic phase coupling modes. Specifically, a multi-scale fast Fourier transform algorithm is used to perform frequency domain transformation on the time-series phase difference data at each height level and spatial location in the tensor, extracting its spectral distribution characteristics. To ensure the comparability of the transformed spectral data, the spectral amplitudes at different spatial locations are normalized, and standardized frequency coordinates are unified to construct a frequency domain phase difference spectral tensor with consistent frequency resolution. This spectral tensor preserves the dominant frequency information of the wake disturbance in the frequency domain and is a key intermediate data structure for discovering co-frequency coupling phenomena.

[0075] A sparse representation reconstruction of the spectral tensor is performed to eliminate non-coupled interference frequency components. Specifically, a sparse coding algorithm based on L1 regularization constraints is applied to perform sparse reconstruction on each height-level spectral slice, retaining the dominant frequency components with concentrated spectral energy, significant amplitude, and recurrence across multiple units, while filtering out random high-frequency noise and interference from local unstructured phase jumps on the frequency domain characteristics. To enhance robustness, a spectral time stability index is introduced, scoring and filtering frequency channels with relatively stable frequency and phase over continuous time periods, retaining stable co-frequency characteristic signals, and providing a foundation for subsequent spatial coupling identification.

[0076] Based on the sparsely processed spectral tensor, a co-frequency coupling channel map is constructed in the frequency dimension. Specifically, cluster analysis is performed on spatial regions where a dominant frequency component coexists in the wakes of different wind turbines, and the clustering results are weighted and evaluated using a phase difference similarity matrix. Clustering results that satisfy the requirements of continuous phase difference distribution, relatively consistent frequency amplitude, and coherent spatial distribution are marked as potential phase coupling channels. By constructing a co-occurrence heatmap in both frequency and spatial dimensions, a visualized set of channels for co-frequency resonance of wake phases among wind turbines is formed. This channel set can characterize the spatial synchronicity and coherence of wake vortices among multiple turbines, and is an important basis for judging the potential resonance impact between wakes.

[0077] All identified phase-coupled channel results are fused to generate a wake coherent structure set. Specifically, structures that are located within the same frequency channel, have spatial connectivity, smooth phase transitions, and appear in multiple height layers are considered as a single wake coherent structure unit. Each wake coherent structure unit includes its corresponding frequency identifier, spatial envelope, phase distribution function, and duration index, and is stored as a callable structure dataset.

[0078] Through the above processing, the coherent vortex structure with resonant characteristics between wind turbine wakes in the frequency domain can be accurately extracted from the original complex high-dimensional phase difference data, and represented in a structured manner, providing a quantifiable and callable source of disturbance input for the dynamic headroom prediction model. This processing mechanism significantly improves the modeling accuracy and prediction completeness of nonlinear wake coupling risks under densely distributed wind farm conditions.

[0079] S500: Project the wake coherent structure set onto the preset layer height index surface, and perform adaptive phase weight integration processing on the vortex phase spectrum information in each layer height to calculate the phase energy density profile of each layer height.

[0080] To achieve a detailed modeling of the phase coupling energy distribution characteristics of wake coherent structures at different heights in the vertical direction, the identified wake coherent structure set needs to be projected spatially onto a defined layered height index surface. Then, within each height layer, an adaptively adjustable phase weight integration operation is performed on the vortex phase spectrum information to obtain the phase energy density profile reflecting the phase perturbation intensity at each height layer. This process specifically includes the following steps:

[0081] A hierarchical height indexing system matching the operating range of wind turbine blades is constructed. Based on the sweep height distribution of wind turbine blades and the principle of dividing the clearance area between the tower and the ground, the entire vertical space is divided into several height index surfaces using either equal or finely unequal height methods. Each height index surface is defined as a two-dimensional spatial region, corresponding to a fixed altitude coordinate, used to carry the wake disturbance characteristic data projected within that height layer. By performing geometric sectioning calculations on the spatial envelope of the wake coherent structure set, its volumetric shape is peeled off layer by layer according to vertical height, obtaining its horizontal projection pattern on each height index surface, forming a hierarchical projection section set of the wake coherent structure.

[0082] For each height index surface, vortex phase spectrum information within the corresponding region is extracted. Specifically, a spatial mapping algorithm is used to assign the coordinate points in the normalized vortex phase spectrum contained in each wake coherent structural unit to the corresponding index surface according to their height, and the phase dominant frequency, phase angle value, and spectral amplitude of all valid points on the index surface are extracted. Based on this, each valid data point is assigned spatial location weight and frequency response weight, forming a weight vector set containing three-dimensional features of position, frequency, and phase, which supports the differentiated evaluation of different types of disturbances in subsequent integration operations.

[0083] Within each height index plane, adaptive phase weight integration is performed to calculate the phase energy density of the corresponding height layer. Specifically, a weighted integration strategy based on a density distribution kernel function is used to weight and accumulate all valid phase spectrum points according to their weight vectors. A phase inflection factor is introduced during integration, assigning higher weights to points with similar phase angles to ensure that phase-consistent perturbations have a greater impact on the overall energy assessment result. Simultaneously, an adaptive adjustment factor is set to dynamically adjust the integration kernel width when phase transition regions or regions of drastic frequency fluctuations appear in a certain height layer, thereby reducing the sensitivity of the calculation process to anomalous perturbation points. This integration method not only reflects the spatial energy density distribution of phase perturbations but also preserves the continuity of the dominant frequency characteristics of the vortex structure along the height direction.

[0084] The phase energy density values ​​obtained from each height index surface are normalized and organized to construct a phase energy density profile. This profile, with height as the horizontal axis and phase energy density as the vertical axis, forms an energy distribution curve that varies with height, clearly reflecting the strength of the disturbance that the wake coherent structure may cause to the wind turbine blades in different height regions. This profile can not only be used to identify disturbance hotspot height regions, but also serve as an input feature in lateral yaw prediction, supporting the subsequent construction of a phase-coupled response model across heights. To enhance its real-time application capabilities, the phase energy density profile can be updated according to a time series and linked with real-time wind turbine operating data to achieve continuous perception and prediction support of disturbance characteristics under dynamic wind fields.

[0085] Through the above steps, the phase disturbance intensity of the wake coherent structure at different height layers can be captured with high precision, realizing the effective mapping of spatial distribution to vertical structure, and providing a key modeling foundation for subsequent resonance detection and dynamic prediction in the form of high-resolution and robust data, which significantly improves the hierarchical response capability of the wind turbine blade clearance dynamic prediction model to spatial coupling disturbances.

[0086] S600: Construct a phase coupling curve across the layer height based on the phase energy density profile, and input the phase coupling curve into a multi-scale resonant detector to identify the resonant response behavior caused by the superposition of wake phases, and output the wake phase superposition response amplitude.

[0087] To further reveal the phase energy transfer characteristics of the wake vortex structure in the vertical direction and identify whether spatial resonance phenomena caused by wake phase superposition exist, a phase coupling curve across the layer height needs to be constructed based on the aforementioned phase energy density profile. This phase coupling curve is then input into a multi-scale resonant detector for spectral response analysis to output the wake phase superposition response amplitude, reflecting the wake disturbance intensity. The specific processing steps include the following:

[0088] A phase coupling curve across different height layers is constructed. This curve expresses the degree of correlation between the phase energies of the wake coherent structure in the vertical direction. Specifically, based on each phase energy density profile, phase energy correlation indices are calculated between any two height index surfaces, including energy spectrum similarity, phase dominance frequency difference, and phase angle synchronization coefficient. These indices are combined along the height dimension to form multiple coupling paths across different height layers. Directional constraints and intensity filtering are applied to all paths, retaining only those curves with consistent phase change trends and good energy continuity, forming the final set of phase coupling curves. Each phase coupling curve consists of a set of physically continuous height points, representing the coherent evolution trajectory of vortex disturbances propagating from one height to another.

[0089] To quantitatively assess the phase-energy coupling relationship of wake perturbations between index surfaces at different heights, three indicators are introduced as criteria: energy spectrum similarity, phase dominance frequency difference, and phase angle synchronization coefficient. These indicators can be measured using the following existing mathematical definitions or algorithms:

[0090] Energy spectral similarity: Cosine similarity or Pearson correlation coefficient can be used to compare the energy spectral distribution vectors of two height layers. Cosine similarity assesses the consistency of spectral morphology by calculating the cosine of the angle between the two spectral vectors (between 0 and 1); while the correlation coefficient reflects the degree of linear correlation of spectral amplitudes at different frequencies. Both can effectively represent the similarity of energy distribution in the frequency domain.

[0091] Phase dominance frequency difference: This can be defined by peak frequency offset, which is calculated by extracting the most significant dominant frequency point (i.e., the frequency at the maximum energy) from the energy spectrum at two different altitude levels and calculating their numerical difference, usually expressed as a normalized frequency difference. This index reflects the consistency of the dominant frequency of the disturbance in spatial propagation; the smaller the dominance frequency difference, the more stable the propagation chain.

[0092] Phase angle synchronization coefficient: This can be measured using metrics such as Phase Locking Value (PLV) or Mean Phase Coherence (MPC). Taking PLV as an example, it is calculated by projecting the phase angle difference between corresponding frequency points of two height layers onto a unit vector and averaging the values, resulting in a value between 0 and 1. A higher value indicates stable phase synchronization behavior between the two layers at that frequency point, which is suitable for identifying the cooperative rotation characteristics of wake disturbances.

[0093] Secondly, the phase coupling curves are organized into a time-series curve cluster in the time dimension to support the dynamic analysis of resonance trends. Specifically, each phase coupling curve is sampled within a given time window, its change trajectory at multiple time points is extracted, and curve fitting and modeling are performed on characteristic parameters such as phase dominance frequency, phase amplitude, and phase slope to generate an evolution curve of phase over time. This method can identify the gradually increasing wake disturbance trend over time, providing an early warning basis for identifying potential resonance accumulation. Furthermore, a phase slope change matrix can be constructed to statistically analyze the phase energy transfer rate between each pair of height layers, thereby identifying whether the disturbance exhibits intensified propagation characteristics along the height direction.

[0094] The aforementioned phase coupling curves and their time evolution data are input into a multi-scale resonant detector to perform cross-scale frequency response analysis. The multi-scale resonant detector includes multiple predefined spatial frequency kernels and height filters, capable of simulating the resonant response capability of wind turbine blades to disturbances at different spatial scales. During the detection process, the input phase coupling curves are frequency-domain transformed at multiple spatial scales to extract the dominant frequency component and secondary frequency envelope, which are then compared with the blade's natural vibration frequency. When the detector identifies an input signal frequency close to or matching the blade's natural frequency at any scale, and the phase response shows an increasing trend, it is determined that this path poses a risk of inducing lateral yaw resonance.

[0095] Finally, based on the detection results, a wake phase superposition response amplitude is generated to quantify the potential disturbance intensity of the wake coherent structure on the blade's operating state. Specifically, the generation method involves extracting the dominant frequency amplitude, propagation connectivity, and time stability weights from all phase coupling curves marked as potential resonant paths, and then fusing them using a multi-factor weighted average function to form a unique response amplitude index. Physically, this index represents the energy concentration of the wake disturbance forming coupled resonance in the vertical spatial direction per unit time; a higher value indicates a greater disturbance intensity and a wider impact range.

[0096] To quantitatively evaluate the phase coupling curves of all potential resonant paths, propagation connectivity and temporal stability weights need to be extracted from both spatial and temporal dimensions, as calculated below:

[0097] Propagation connectivity is used to characterize whether a phase coupling curve forms a continuous and coherent perturbation propagation chain in the vertical spatial dimension. This index can be achieved by statistically analyzing the proportion of continuous effective points of phase energy density between adjacent height index surfaces. Specifically, in each height layer corresponding to a phase coupling curve, an energy density threshold (e.g., a multiple of the global mean) is set, and all effective phase points exceeding the energy density threshold are marked. Then, the proportion of these effective phase points continuously distributed in the height direction is calculated, i.e., the ratio of the length of effective continuous height segments to the total length of height segments, representing the propagation connectivity. The closer the value is to 1, the more coherent the propagation of the curve in space, and the more likely the perturbation is to form a stable transmission channel.

[0098] The time stability weight is used to measure the consistency of the dominant frequency perturbation characteristics of a phase-coupled path within a given time window. The specific extraction method is as follows: within a fixed-length time sliding window (e.g., 5 seconds or 10 seconds), the amplitude and phase angle sequences corresponding to the dominant frequency points of the path are continuously tracked, and their standard deviations (or coefficients of variation) are calculated to characterize the volatility of the path's perturbation characteristics. Then, an inverse proportional normalization function (e.g., weight = 1 / (1 + standard deviation)) is used to convert this volatility into a stability score. The smaller the standard deviation, the more stable the path is in the time domain, and the higher its corresponding time stability weight, which can be used to improve the reliability of the path in the final response amplitude calculation.

[0099] Through the above steps, it is possible to model the propagation path of wake disturbances in the height dimension, perform dynamic evolution analysis and quantitative assessment of resonance risk, and finally generate wake phase superposition response amplitude with real-time and callable capability. This provides quantitative input for the entire airspace prediction system to enhance disturbances under extreme conditions, effectively improving the operational safety assurance capability of wind turbines in complex wake environments.

[0100] S700: Based on the wake phase superposition response amplitude and phase coupling curve, a wake phase superposition index is generated by a normalized peak operator. The wake phase superposition index is then dynamically input as a correction factor into the wind turbine blade clearance nonlinear dynamic prediction model to adjust the wind pressure input boundary conditions in the wind turbine blade clearance nonlinear dynamic prediction model, thereby improving the response sensitivity and prediction coverage of the prediction model to extreme yaw behavior.

[0101] To achieve high sensitivity and enhanced prediction capability of the wind turbine blade clearance nonlinear dynamic prediction model for extreme yaw behavior, it is necessary to normalize the obtained wake phase superposition response amplitude and phase coupling curve, and construct a wake phase superposition index as the basis for dynamic adjustment of wind pressure boundary conditions. The introduction of this index enables the wind turbine blade clearance dynamic prediction model to have higher responsiveness and prediction coverage under complex wake interference environments. Specifically, the following steps are included:

[0102] A composite disturbance feature representation structure is constructed by integrating the wake phase superposition response amplitude with the corresponding phase coupling curve. Specifically, the local peak point that resonates in the time dimension within each cross-height phase coupling curve is paired with the corresponding wake phase superposition response amplitude. Information such as frequency, phase amplitude, response intensity, and disturbance duration are uniformly encoded to form a disturbance feature set. Each entry in this set can be mapped to a spatiotemporal disturbance peak event, used to characterize the intensity evolution trend of wake disturbances along their spatial propagation path. This structure provides a rich quantitative foundation for the construction of the wake phase superposition index.

[0103] Based on the aforementioned set of disturbance characteristics, the normalized peak operator is invoked to normalize the amplitudes of all wake phase superposition responses, generating a dimensionless wake phase superposition index. The normalized peak operator is an arithmetic operator that standardizes the peak response of dynamic signals. Its core idea is to normalize all peak signals according to the maximum value in the region, while introducing an amplitude boundary compression function based on quantiles to prevent extreme values ​​from dominating the global representation. To preserve the spatial distribution characteristics of the disturbance, the operator also introduces a highly distributed weighting and phase similarity weighting mechanism, giving higher weight to highly coupled and highly consistent disturbance characteristics in the wake phase superposition index. After the normalization process is completed, a unified index reflecting the spatial synchronicity and intensity integration effect of the current wake disturbance is formed, defined as the wake phase superposition index.

[0104] The wake phase superposition index is used as a disturbance sensitivity correction factor and is input in real time into the nonlinear dynamic prediction model of wind turbine blade clearance to dynamically adjust the wind pressure input boundary conditions in the model. Specifically, in the wind pressure field modeling module of the prediction model, the initial amplitude parameters and turbulence intensity factor of the wind pressure disturbance distribution within the prediction area are adjusted based on the wake phase superposition index value. In the load calculation stage, the wake phase superposition index is converted into an amplitude adjustment factor for the lateral load to amplify the load simulation effect under extreme conditions. In the wind speed input probability distribution modeling, the wake phase superposition index is used as a skew correction parameter for the non-Gaussian disturbance factor to guide the model to generate higher disturbance energy input sequences in regions with strong wake coupling. Through this series of dynamic adjustment mechanisms, the prediction model, which originally relied solely on conventional boundary data, gains the ability to respond in advance to potential extreme disturbance paths.

[0105] The prediction performance of the adjusted wind turbine blade clearance nonlinear dynamic prediction model was evaluated, and the wake phase superposition index was back-optimized and iteratively updated based on the prediction results. Evaluation methods included: comparing the error range, maximum yaw amplitude prediction accuracy, and warning lead time of models using and without the wake phase superposition index when simulating lateral yaw trajectories; statistically analyzing the prediction success rate and false alarm rate of the model under different wake coupling intensities; and recording the model's response delay time to joint disturbances from multiple wind turbines. If the model exhibits over- or under-prediction under specific operating conditions, the wake phase superposition index at the corresponding moment is fed back into the normalized peak operator parameters to perform iterative optimization of the amplitude compression range or phase weight function. Through this closed-loop adjustment mechanism, the wake phase superposition index continuously self-adjusts as time and the disturbance environment evolves, ensuring the prediction model remains within the response range most sensitive to high-risk disturbances.

[0106] This implementation method constructs a wake phase superposition index and introduces it into the nonlinear dynamic prediction model of wind turbine blade clearance. This not only achieves the technical goals of enhancing disturbance and improving prediction sensitivity, but also establishes a dynamic coupling feedback channel between disturbance identification and prediction response, providing a solution for extreme conditions in high-density wind farms.

[0107] The aforementioned method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance achieves comprehensive modeling and dynamic response compensation for the complex disturbance mechanism caused by the interaction of wakes from multiple wind turbine units in a wind farm, demonstrating significant beneficial effects. Specifically, this method introduces a multi-frequency phased array lidar matrix for the first time to acquire high-resolution three-dimensional wind speed voxel data, and extracts the vortex rotation phase spectrum through continuous frame profile phase convolution. Combined with a high-dimensional phase difference tensor and sparse spectrum aggregation strategy, it achieves accurate identification of the spatial coherence characteristics of the wake vortex structure. On this basis, it further generates a phase energy density profile through layered height integration and uses a multi-scale resonance detection mechanism to identify wake phase superposition paths that may induce blade yaw resonance. Finally, it forms a wake phase superposition index as a dynamic correction factor input to the prediction model, effectively improving the model's ability to identify extreme yaw risks and its dynamic adaptability. This solves the problems of underestimating wake disturbance coupling and inaccurate prediction of yaw amplitude commonly found in existing technologies, significantly improving the safe operation guarantee level of wind turbine units in complex wake environments.

[0108] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance, characterized in that, Includes the following steps: S100: Deploy a multi-frequency phased array lidar matrix to collect spatial wind speed data in the wake area of ​​the front-row wind turbines in the wind farm and construct a three-dimensional instantaneous velocity voxel library. S200, based on a three-dimensional instantaneous velocity voxel library, performs profile phase convolution processing on velocity voxels at continuous time points to extract wake rotation phase information and map it into a standardized vortex phase spectrum; S300, constructing a high-dimensional phase difference tensor based on the vortex phase spectrum of adjacent units; S400: Perform sparse spectrum aggregation on the phase difference tensor, extract phase coupling channels with the same frequency characteristics, identify wake vortex structures with spatial coherence characteristics, and obtain a set of wake coherent structures. S500: Project the wake coherent structure set onto the layer height index plane, and perform phase weight integration on the corresponding height layer to obtain the phase energy density profile; S600 constructs a phase coupling curve across height based on the phase energy density profile and inputs it to a multi-scale resonant detector to identify the resonant response behavior and output the wake phase superposition response amplitude. S700 generates a wake phase superposition index based on the wake phase superposition response amplitude and phase coupling curve, and dynamically inputs it as a correction factor into the wind turbine blade clearance nonlinear dynamic prediction model to adjust the wind pressure input boundary conditions.

2. The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to claim 1, characterized in that, Step S100 includes: A multi-frequency phased array lidar is deployed in the wind farm, consisting of multiple lidar units whose transmission frequency and direction can be independently controlled. The lidar array is controlled to perform a three-dimensional spatial synchronous scan within the wake influence area, collect wind speed vector data, and generate a spatial wind speed vector field. The spatial wind speed vector field is mapped to a unified three-dimensional coordinate system and divided into multiple three-dimensional voxel units containing wind speed vector information to form a three-dimensional instantaneous velocity voxel set. The three-dimensional instantaneous velocity voxel set is time-series numbered and stored to construct a three-dimensional dynamic velocity voxel data sequence that reflects the wake evolution process.

3. The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to claim 1, characterized in that, Step S200 includes: Extract velocity voxel data from continuous time frames from a 3D instantaneous velocity voxel library, form a velocity voxel time series, and perform spatial registration; A set of profile slices was constructed in the wake axis region of the wind turbine based on time series, and the rotational phase distribution of the profile was extracted by two-dimensional Fourier transform and Hilbert algorithm. The multi-frame profile rotation phase map is jointly processed to generate the phase evolution trajectory at a continuous time scale, and normalization and bandwidth limiting processing are performed. The phase trajectory is mapped to a standardized vortex phase spectrum with the wind turbine rotor surface as a reference.

4. The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to claim 1, characterized in that, Step S300 includes: Standardized vortex phase spectra are constructed for adjacent wind turbine units, and unified spatial grids and height layers are processed. Within the same height index layer, calculate the phase difference between corresponding voxel regions and construct a data structure for joint mapping of three-dimensional space and time. A high-dimensional phase difference tensor is constructed to record the quantized values ​​of the phase difference, and then the high-dimensional phase difference tensor is normalized and noise filtered out.

5. The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to claim 1, characterized in that, Step S400 includes: Perform a multi-scale fast Fourier transform on the high-dimensional phase difference tensor to construct a normalized frequency domain phase difference spectrum tensor; Sparse reconstruction based on L1 regularization term constraint is performed on the frequency domain phase difference spectrum tensor to preserve stable main frequency characteristics and filter out interference frequency components. Spatial regions are clustered based on frequency co-occurrence and phase similarity to identify phase-coupled channels that satisfy the same frequency characteristics; Phase coupling channels that satisfy frequency consistency, spatial connectivity, and phase smoothness are merged to form a wake coherent structure set.

6. The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to claim 1, characterized in that, Step S500 includes: Construct a layered height index surface that matches the operating range of wind turbine blades, and project the wake coherent structure set onto each height layer; Extract the vortex phase spectrum information from each height index surface and generate a weighted vector set containing position, frequency and phase features; Perform adaptive phase weight integration within each altitude layer to calculate the phase energy density; The obtained phase energy density is normalized and sorted to construct a phase energy density profile that reflects the intensity of the disturbance.

7. The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to claim 1, characterized in that, Step S600 includes: Phase coupling curves across different layer heights are constructed based on the phase energy density profile to represent the phase energy correlation paths between layers at different heights; The phase coupling curves are organized into a time series curve cluster in the time dimension, and the phase feature parameters are extracted and their evolution trend is modeled. The phase coupling curve is input into a multi-scale resonant detector to perform frequency domain analysis and compare it with the blade's natural frequency to identify potential resonance paths. Based on the identification results, frequency and stability indicators are extracted, and the wake phase superposition response amplitude is generated to quantify the disturbance intensity.

8. The method for constructing a nonlinear dynamic prediction model for wind turbine blade clearance according to claim 1, characterized in that, Step S700 includes: By fusing the wake phase superposition response amplitude and phase coupling curve, a set of disturbance characteristics is constructed. The normalized peak operator is invoked to normalize the wake phase superposition response amplitude in the disturbance feature set, generating the wake phase superposition index. The wake phase superposition index is used as a correction factor to dynamically input the prediction model, which is used to adjust the wind pressure boundary conditions and turbulence input parameters. The model performance is evaluated based on the prediction results, and the evaluation feedback is used for parameter optimization and iterative updates of the wake phase superposition index.

Images