A large wind turbine monitoring and sensing system and method

Abstract

The present application belongs to the technical field of wind power generation, and discloses a large wind turbine monitoring and sensing system and method, aiming to solve the problem of load prediction deviation of downwind units caused by wake interference in the array effect of wind turbine groups. The present application obtains real-time operation state data of the wind turbine group, constructs a dynamic topology graph of wake propagation, and accurately represents the spatial influence range and intensity distribution of the wake. Based on the set of blade turbulent separation feature points and the topology graph of the wake, the sensitive area of the downwind unit to the wake interference is identified, and the dynamic gain effect of the wake interference is analyzed through the vibration response transfer function to generate a real-time load prediction correction model. The model combines adaptive optimization of prediction deviation to ensure prediction accuracy. At the same time, a dynamic risk assessment model of the wind turbine group is constructed to optimize the operation scheduling strategy and reduce the influence of wake interference. The present application improves the load prediction accuracy, the operation safety of the unit and the overall efficiency of the wind turbine group.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and more specifically, to a monitoring and sensing system and method based on large wind turbines. Background Technology

[0002] With the rapid growth of global demand for renewable energy, wind power, as an important component of clean energy, has become one of the core technologies for energy structure transformation. In particular, the widespread application of large-scale wind turbine clusters has significantly improved the installed capacity and power generation efficiency of wind farms. However, in the actual operation of wind farms, the array effect of wind turbine clusters brings many technical challenges, the most prominent of which is the impact of the wake of upstream wind turbines on downwind turbines. The wake refers to the low-speed, high-turbulence wind field region formed downstream after the blades of the upstream turbines rotate. Its propagation path and intensity are affected by various factors such as wind speed, wind direction, terrain, and turbulence intensity, exhibiting significant spatial heterogeneity and temporal dynamics. Wake interference can lead to abnormal blade stress, increased vibration, and uneven load distribution in downwind turbines, thereby affecting the operating efficiency, safety, and service life of the turbines.

[0003] Specifically, wind turbine clusters are typically deployed in complex terrain or near-shore environments. The wake generated by upstream turbines propagates dynamically and heterogeneously to downwind turbines, causing significant changes in the blade stress, vibration response, and load distribution of the downwind turbines. However, existing technologies often rely on static or quasi-static wake models (such as the Jensen model based on empirical formulas or simple turbulence attenuation assumptions). These models struggle to accurately capture the spatial heterogeneity and temporal dynamics of wake propagation, especially under complex conditions involving wind speed fluctuations, wind direction changes, or dramatic fluctuations in turbulence intensity. For example, under the influence of strong wind shear or terrain-induced turbulence, the wake propagation path may exhibit nonlinear and non-uniform characteristics. Existing methods typically ignore the spatial gradient differences in wake energy transfer and the local concentration effect of turbulence intensity, leading to significant biases in the assessment of the wake interference range and intensity of downwind turbines. Furthermore, existing technologies lack accurate identification of wake interference-sensitive areas in load prediction and fail to effectively combine the turbulence separation characteristics of the blade surface with the dynamic response characteristics of the turbine vibration modes, further exacerbating prediction biases. For example, in actual operation, downwind units may experience turbulent separation or pressure anomalies in localized areas of the blades due to wake interference. However, existing methods cannot extract these features in real time and incorporate them into the load prediction model, resulting in significant deviations between the predicted results and the actual loads, especially in high wake interference regions where the deviation can exceed 20%. This deviation not only affects the safety and stability of unit operation but may also lead to accelerated accumulation of fatigue damage, shorten the unit's service life, and even, in extreme cases, trigger structural failure.

[0004] In view of this, the present invention proposes a monitoring and sensing system and method based on large wind turbines to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a monitoring and sensing method based on a large wind turbine, comprising:

[0006] The real-time operating status data sequence of each wind turbine in the wind turbine group is obtained. The real-time operating status data sequence includes the pressure distribution of the turbulent boundary layer on the blade surface, the velocity vector field of the upstream wake, and the vibration mode parameters of the unit.

[0007] Based on the upstream wake velocity vector field, a wake propagation dynamic topology map is constructed. The wake propagation dynamic topology map characterizes the spatial influence range and intensity distribution of the upstream wind turbine wake on the downwind wind turbine.

[0008] Based on the pressure distribution of the turbulent boundary layer on the blade surface, the set of local turbulent separation feature points of the blade is extracted, and combined with the wake propagation dynamic topology map, the wake interference sensitive area of ​​the downwind wind turbine is calculated.

[0009] Using the unit's vibration mode parameters, a vibration response transfer function under wake interference is constructed, and the dynamic gain factor of wake interference on the downwind wind turbine load is identified through the vibration response transfer function.

[0010] Based on the wake interference sensitive area and dynamic gain factor, a real-time load prediction correction model for downwind wind turbine generators is generated.

[0011] Based on the real-time load prediction correction model, the load prediction value of the downwind wind turbine is output, and adaptive optimization is performed by combining the deviation between the prediction value and the actual load.

[0012] Preferably, constructing a wake propagation dynamic topology graph includes:

[0013] The upstream wake velocity vector field is discretized in three dimensions to generate a wake velocity vector grid. Each grid node of the wake velocity vector grid contains velocity components and turbulence intensity components.

[0014] Based on the wake velocity vector grid, the wake energy transfer coefficient between adjacent grid nodes is calculated. The wake energy transfer coefficient is determined by the gradient direction of the velocity component and the spatial attenuation rate of the turbulence intensity.

[0015] Using the wake energy transfer coefficient, a directed weighted graph of wake propagation is constructed, where the edge weights of the directed weighted graph are characterized by the wake energy transfer coefficient, and the nodes represent the grid nodes in the wake velocity vector grid.

[0016] Dynamic clustering analysis is performed on the directed weighted graph to identify high-energy concentrated regions in the wake propagation path, and the spatial boundaries of the high-energy concentrated regions are marked as the wake influence core area;

[0017] Geometrically map the spatial locations of the wake influence core area and the downwind wind turbine to generate a dynamic topology map of wake propagation.

[0018] Preferably, the set of local turbulence separation feature points extracted from the blade includes:

[0019] Wavelet decomposition was performed on the pressure distribution of the turbulent boundary layer on the blade surface to extract high-frequency pressure fluctuation components. These high-frequency pressure fluctuation components characterize the local disturbance features of the turbulent separation on the blade surface.

[0020] Based on the high-frequency pressure fluctuation component, the turbulence separation probability of each sampling point on the blade surface is calculated. The turbulence separation probability is determined by the ratio of the local variance to the global variance of the pressure fluctuation component.

[0021] Identify sampling points whose turbulence separation probability exceeds a preset probability threshold and mark them as candidate turbulence separation feature points;

[0022] Spatial connectivity analysis is performed on candidate turbulence separation feature points to eliminate isolated points, and the geometric center of the connected region is taken as the final turbulence separation feature point to form a set of local turbulence separation feature points of the blade.

[0023] The time series of local turbulence separation feature points on the blade is tracked to remove transient noise points and retain turbulence separation feature points that appear periodically or quasi-periodically.

[0024] Preferably, the calculation of the wake interference sensitive area of ​​the downwind wind turbine generator includes:

[0025] Project the wake influence core area in the wake propagation dynamic topology diagram onto the blade spatial coordinate system of the downwind wind turbine to generate a wake interference projection diagram.

[0026] Based on the wake interference projection map, the spatial overlap between the wake influence core area and the set of characteristic points of local turbulence separation of the blade is calculated. The spatial overlap is determined by the weighted sum of the area ratio of the overlapping area and the turbulence intensity.

[0027] The spatial overlap is graded and quantified to identify blade regions whose spatial overlap exceeds a preset overlap threshold and mark them as candidate wake interference sensitive regions.

[0028] Based on the pressure distribution of the turbulent boundary layer on the blade surface, the pressure gradient anomaly index in the candidate wake disturbance sensitive region is calculated. The pressure gradient anomaly index is determined by the deviation between the local extremum and the global mean of the pressure gradient.

[0029] Candidate wake interference sensitive regions whose pressure gradient anomaly index exceeds a preset anomaly threshold are identified as the final wake interference sensitive regions.

[0030] Preferably, the vibration response transfer function under wake disturbance is constructed as follows:

[0031] The vibration mode parameters of the unit are decomposed by spectrum, and the dominant frequency and damping ratio of each mode are extracted. The dominant frequency and damping ratio characterize the dynamic response characteristics of the wind turbine generator under wake interference.

[0032] Based on the dominant frequency, the frequency domain response basis function of the vibration mode is constructed. The frequency domain response basis function is determined by the harmonic components of the dominant frequency and the attenuation factor of the damping ratio.

[0033] By utilizing the wake influence core region in the wake propagation dynamic topology diagram, the modulation coefficient of the wake interference on the frequency domain response basis function is calculated. The modulation coefficient is determined by the spatial distribution of wake turbulence intensity and the spatial coupling degree of vibration mode.

[0034] The vibration response transfer function is generated by convolving the modulation coefficients with the frequency domain response basis function.

[0035] The vibration response transfer function is verified in the time domain, and the parameters of the transfer function are iteratively optimized using the response error of the actual vibration signal.

[0036] Preferably, identifying the dynamic gain factor of wake interference on the load of downwind wind turbine generators includes:

[0037] Based on the vibration response transfer function, the spectral characteristics of the wake interference input signal are calculated. The wake interference input signal is determined by the wake turbulence intensity sequence in the wake propagation dynamic topology diagram.

[0038] By using the vibration response transfer function, the wake interference input signal is mapped to the vibration output signal of the downwind wind turbine, and a vibration output spectrum is generated.

[0039] Extract the gain peak value from the vibration output spectrum. The gain peak value characterizes the amplification effect of the wake interference on the vibration response.

[0040] Based on the spatial distribution of the wake interference sensitive area, the weighted average value of the gain peak value in different blade areas is calculated, with the spatial overlap of the wake interference sensitive area as the weight.

[0041] The weighted average value is normalized and used as the dynamic gain factor.

[0042] Preferably, the real-time load prediction correction model for downwind wind turbine generators is generated, including:

[0043] An initial load prediction model is constructed, which takes wind speed, blade angle of attack and unit speed as inputs and blade root stress as output.

[0044] Based on the wake interference sensitive area, the disturbance component of the wake interference on the input parameters of the initial load prediction model is extracted. The disturbance component is determined by the coupling effect of the wake turbulence intensity and the set of local turbulence separation feature points of the blade.

[0045] By using a dynamic gain factor, the output of the initial load prediction model is corrected to generate the corrected load prediction value.

[0046] A time delay compensation mechanism is introduced, which determines the propagation time delay of the wake interference by the wake propagation path length and the average velocity of the velocity vector field in the wake propagation dynamic topology graph.

[0047] By combining the corrected load prediction values ​​with a time delay compensation mechanism, a real-time load prediction correction model is generated.

[0048] Preferably, adaptive optimization is performed by combining the deviation between the predicted value and the actual load, including:

[0049] Real-time acquisition of actual load data of downwind wind turbine generators; the actual load data is obtained by measuring strain sensors at the blade root.

[0050] Calculate the deviation sequence between the load prediction values ​​output by the real-time load prediction correction model and the actual load data;

[0051] Time-frequency analysis is performed on the deviation sequence to extract the dominant frequency component and amplitude component. The dominant frequency component characterizes the periodicity of the deviation, and the amplitude component characterizes the intensity of the deviation.

[0052] Based on the dominant frequency component, the damping parameter of the vibration response transfer function in the real-time load prediction correction model is adjusted.

[0053] Based on the amplitude component, adjust the weight coefficient of the dynamic gain factor in the real-time load prediction correction model.

[0054] The stability of the adjusted real-time load prediction correction model is tested. When the deviation amplitude of multiple consecutive frames is lower than the preset deviation threshold, the optimization is confirmed to be complete.

[0055] Preferably, the method further includes:

[0056] A dynamic risk assessment model for the wind turbine array effect is constructed. The dynamic risk assessment model takes the output of the wake propagation dynamic topology map and the real-time load prediction correction model as input to evaluate the fatigue damage accumulation rate of downwind wind turbine generators.

[0057] Based on the fatigue damage accumulation rate, high-risk areas of wake interference in the wind turbine group are identified, and a spatial distribution map of the risk areas is generated.

[0058] By utilizing the spatial distribution map of the risk area, the operation and scheduling strategy of the wind turbine group is optimized. The operation and scheduling strategy includes adjusting the yaw angle and blade pitch angle of the upstream wind turbine to reduce the impact of wake interference on the downwind wind turbine.

[0059] The optimized operation and scheduling strategy was validated in a closed loop by comparing the actual load changes of downwind wind turbines before and after optimization to assess the effectiveness of the strategy.

[0060] A monitoring and sensing system based on large wind turbines, which implements a monitoring and sensing method based on large wind turbines, includes:

[0061] The data acquisition module is used to acquire real-time operating status data sequences of each wind turbine in the wind turbine group. The real-time operating status data sequences include the pressure distribution of the turbulent boundary layer on the blade surface, the velocity vector field of the upstream wake, and the vibration mode parameters of the unit.

[0062] The wake topology construction module is used to construct a dynamic topology diagram of wake propagation based on the upstream wake velocity vector field.

[0063] The turbulence separation feature extraction module is used to extract the set of local turbulence separation feature points of the blade based on the pressure distribution of the turbulent boundary layer on the blade surface, and to calculate the wake interference sensitive area of ​​the downwind wind turbine by combining the wake propagation dynamic topology map.

[0064] The vibration response analysis module is used to construct the vibration response transfer function under wake interference using the unit's vibration mode parameters, and to identify the dynamic gain factor of wake interference on the downwind wind turbine load through the vibration response transfer function.

[0065] The load prediction model generation module is used to generate a real-time load prediction correction model for downwind wind turbines based on the wake interference sensitive area and dynamic gain factor.

[0066] The adaptive optimization module is used to output the load prediction value of the downwind wind turbine based on the real-time load prediction correction model, and perform adaptive optimization by combining the deviation between the prediction value and the actual load.

[0067] The technical effects and advantages of the present invention regarding a monitoring and sensing system and method for large wind turbines are as follows:

[0068] This invention effectively reduces the uncertainties caused by changes in environmental and operating conditions, enabling downwind turbines to maintain a more stable operating state when facing complex disturbances. This significantly extends the service life of the turbines and reduces fatigue damage and potential structural failure risks caused by abnormal loads. Simultaneously, by optimizing the operating strategy of the entire wind turbine group, this invention maximizes overall power generation efficiency, especially in high-interference areas, effectively alleviating the operating pressure on local turbines and thus enhancing the overall energy output capacity of the wind farm. Furthermore, this invention reduces the operation and maintenance costs of wind farms by minimizing unplanned downtime and maintenance needs caused by misjudgments or delayed responses, bringing higher economic returns to wind farm operators. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of a monitoring and sensing method based on a large wind turbine according to the present invention;

[0070] Figure 2 This is a schematic diagram of a monitoring and sensing system based on a large wind turbine according to the present invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] This application provides a monitoring and sensing system and method based on large wind turbines. The main execution entities of the system include, but are not limited to, wind farm monitoring systems, wind turbine condition monitoring platforms, wake interference analysis systems, and intelligent dispatch centers for wind turbine groups, which can be regarded as general computing nodes of this application. The monitoring and sensing system includes, but is not limited to, at least one of the following: a wind turbine condition monitoring engine, a wake interference assessment system, and a wind turbine load predictor.

[0073] Please see Figure 1 In this embodiment of the invention, a specific implementation process of a monitoring and sensing method based on large wind turbines includes:

[0074] Real-time operational status data sequences of each wind turbine in the wind turbine cluster are acquired. These data sequences include key indicators such as the turbulent boundary layer pressure distribution on the blade surface, the upstream wake velocity vector field, and the turbine vibration modal parameters, all collected in real-time through a distributed sensor network. The turbulent boundary layer pressure distribution on the blade surface directly reflects the stress state of the blade, the upstream wake velocity vector field characterizes the spatial distribution characteristics of wake disturbances, and the turbine vibration modal parameters describe the dynamic response characteristics of the wind turbine during operation. These data provide a foundation for wake disturbance analysis and load prediction, ensuring the accuracy and effectiveness of monitoring.

[0075] Based on the upstream wake velocity vector field, a dynamic topology map of wake propagation is constructed. This dynamic topology map characterizes the spatial range and intensity distribution of the wake's influence on the downwind wind turbines, directly determining the spatial distribution characteristics of wake interference. The dynamic topology map is derived through three-dimensional discretization of the velocity vector field and energy transfer analysis, ensuring that the wake impact assessment matches the actual wind field environment.

[0076] Based on the pressure distribution of the turbulent boundary layer on the blade surface, this module extracts a set of local turbulent separation feature points. First, it performs wavelet decomposition on the pressure distribution to extract high-frequency pressure fluctuation components, identifying regions where the turbulent separation probability exceeds a preset threshold, thus forming a set of turbulent separation feature points. This lays the foundation for calculating wake interference-sensitive areas. Combining the wake propagation dynamic topology map and the set of local turbulent separation feature points on the blade, the module calculates the wake interference-sensitive areas of the downwind wind turbine. By projecting the wake influence core area onto the blade spatial coordinate system, this module analyzes the spatial overlap and pressure gradient anomaly index to identify the blade regions most susceptible to wake interference, ensuring the targeted nature of subsequent load predictions.

[0077] By utilizing the unit's vibration modal parameters, a vibration response transfer function under wake disturbance is constructed, and the dynamic gain factor of wake disturbance on the downwind wind turbine load is identified. This module establishes a mapping relationship between wake disturbance input and vibration output through spectral decomposition and convolution operations, quantifying the amplification effect of wake disturbance on the unit load and providing key parameters for load prediction.

[0078] Based on the wake interference sensitive area and dynamic gain factor, a real-time load prediction correction model for downwind wind turbine generators is generated. This module accurately describes the impact of wake interference on generator load by constructing an initial load prediction model and introducing a time delay compensation mechanism, achieving high-precision load prediction.

[0079] Based on a real-time load prediction correction model, the module outputs predicted load values ​​for downwind wind turbine generators and performs adaptive optimization by considering the deviation between the predicted values ​​and the actual loads. Through time-frequency analysis and parameter adjustment, this module continuously optimizes the prediction model, improves prediction accuracy, and ensures the system's adaptability to changes in the wind field environment.

[0080] In this embodiment of the invention, the detailed implementation steps for constructing a dynamic topology map of wake propagation based on the upstream wake velocity vector field include:

[0081] The upstream wake velocity vector field is discretized in three dimensions to generate a wake velocity vector mesh. Each grid node in the wake velocity vector mesh contains both velocity and turbulence intensity components. Three-dimensional discretization is a fundamental step in constructing the topology graph, transforming the continuous vector field into a discrete mesh structure for easier subsequent calculations. An adaptive mesh generation method is employed, with higher mesh density in the wake core region and relatively sparser mesh in the edge regions, ensuring accuracy in critical areas while controlling computational complexity. Each grid node records the three-dimensional velocity components (u, v, w) and turbulence intensity τ, comprehensively characterizing the local wake properties. The mesh accuracy is typically set to 5-10 meters and dynamically adjusted based on wind field size and computational resources to ensure both reasonable mesh generation and computational efficiency.

[0082] Based on the wake velocity vector grid, the wake energy transfer coefficient between adjacent grid nodes is calculated. This coefficient is determined by the gradient direction of the velocity components and the spatial attenuation rate of the turbulence intensity. The energy transfer coefficient is a key parameter describing the wake propagation characteristics, quantifying the flow law of wake energy in space. The calculation formula is:

[0083] ;

[0084] in, For nodes To the node The energy transfer coefficient, The velocity gradient vector, Let be the position vector between the two nodes. and These represent the turbulence intensities at the two nodes, and These are the weighting coefficients, and .

[0085] The weight settings are dynamically adjusted according to wind characteristics; when the turbulence intensity is high, An increase in the value highlights the effect of turbulence attenuation; when the wind speed gradient is significant, An increasing value emphasizes the contribution of the velocity gradient. The energy transfer coefficient considers both the directionality of velocity changes and the attenuation law of turbulence intensity, comprehensively characterizing the spatial transfer characteristics of wake energy.

[0086] A directed weighted graph of wake propagation is constructed using wake energy transfer coefficients. The edge weights of the directed weighted graph are characterized by the wake energy transfer coefficients, and the nodes represent grid nodes in the wake velocity vector grid. The directed weighted graph is a mathematical model representing the wake propagation path, intuitively displaying the wake energy flow network. The construction process is based on graph theory, with each grid node as a vertex, the energy transfer relationships between nodes as directed edges, and the edge weights as energy transfer coefficients. The graph construction considers computational efficiency by employing a threshold filtering strategy, retaining only edges with transfer coefficients greater than a preset threshold. This ensures the integrity of critical propagation paths while controlling the graph's complexity. The directed weighted graph not only describes the static structure of wake propagation but also reflects the dynamic characteristics of energy transfer, providing an ideal data structure for subsequent cluster analysis.

[0087] Dynamic clustering analysis is performed on directed weighted graphs to identify high-energy concentration regions in the wake propagation path, and the spatial boundaries of these high-energy concentration regions are marked as wake influence core areas. Dynamic clustering is the core step in identifying key influence regions, aiming to find highly energy-concentrated subgraph structures in complex networks. The analysis process employs an improved modularity clustering algorithm, which identifies community structures in the graph by optimizing the modularity function.

[0088] ;

[0089] in, For modularity, It is the sum of the weights of all edges in the graph. For nodes The degree of exit, For nodes in-degree, For nodes and An indicator function for whether they belong to the same community.

[0090] The algorithm maximizes modularity through iterative optimization. The system automatically determines the optimal community division without requiring a preset number of communities. In the clustering results, communities with edge weight density exceeding a preset threshold are marked as high-energy concentration areas, and their spatial boundaries are defined as the wake influence core area. This network topology-based clustering method can accurately identify key areas in wake energy propagation, providing a precise spatial range for the impact assessment of downstream wind turbines.

[0091] A dynamic topology map of wake propagation is generated by geometrically mapping the spatial locations of the wake's core influence area and the downwind wind turbines. Geometric mapping is a crucial step in transforming an abstract graph structure into a real spatial distribution, establishing a spatial relationship between wake influence and turbine locations. The mapping process employs spatial coordinate transformation, converting the wake's core influence area from a grid coordinate system to the global wind field coordinate system. This coordinate system is then overlaid with the location information of the downwind turbines to form a complete dynamic topology map of wake propagation. The topology map adopts a multi-layered structure: the bottom layer represents the wind field layout and turbine locations; the middle layer represents the wake propagation path network; and the top layer represents high-energy concentration areas and the core influence area. This hierarchical topology visually illustrates the entire process of the wake, from its generation and propagation from upstream turbines to its impact on downstream turbines, providing a global perspective and local details for wake interference assessment.

[0092] In this embodiment of the invention, the detailed implementation steps for extracting the set of local turbulent separation feature points of the blade based on the turbulent boundary layer pressure distribution on the blade surface include:

[0093] Wavelet decomposition is performed on the pressure distribution of the turbulent boundary layer on the blade surface to extract high-frequency pressure fluctuation components. These high-frequency pressure fluctuation components characterize the local disturbance features of turbulent separation on the blade surface. Wavelet decomposition is a fundamental step in identifying turbulent separation features and can effectively separate pressure fluctuation characteristics at different frequency scales. The decomposition process uses discrete wavelet transform, recursively filtering to decompose the pressure signal into sub-signals of different frequency bands. Choosing appropriate wavelet basis functions (usually Daubechies wavelets) and the number of decomposition levels (typically 3-5 levels depending on the sampling frequency) ensures accurate extraction of turbulent disturbance features. The wavelet decomposition can be expressed as:

[0094] ;

[0095] in, For the original pressure distribution, For the first Layer details (high-frequency components) For the first Layer approximation information (low-frequency components).

[0096] High-frequency components (typically 1-3 layers of detailed information) correspond to rapid pressure fluctuations caused by turbulent disturbances and are key signal features for identifying turbulence separation. Wavelet decomposition not only provides frequency separation but also preserves the spatiotemporal local characteristics of the signal, providing an accurate feature representation for subsequent turbulence separation probability calculations.

[0097] Based on high-frequency pressure fluctuation components, the turbulent separation probability at each sampling point on the blade surface is calculated. This probability is determined by the ratio of the local variance to the global variance of the pressure fluctuation component. The turbulent separation probability is a key indicator for quantifying the local flow state, reflecting the likelihood of turbulent separation at each point. The calculation employs the variance ratio method, based on the characteristic that high-frequency fluctuations are more intense in turbulent separation regions. The local variance is calculated using a sliding window (the window size is set according to the blade size, typically 5-10% of the blade chord length) to determine the degree of high-frequency component fluctuation around each sampling point, while the global variance represents the average fluctuation level across the entire blade surface. The formula for calculating the turbulent separation probability is:

[0098] ;

[0099] in, Sampling points The probability of turbulent separation, for Local variance around the point For global variance, The correction function is based on the pressure gradient, taking into account the correlation between the pressure gradient and turbulent separation.

[0100] ;

[0101] in, This is the normalized pressure gradient, expressed as the rate of change of pressure along the normal direction of the blade surface. It is the correction amplitude coefficient, which usually ranges from 0.3 to 0.6; It is the sensitivity coefficient, which typically ranges from 2.0 to 4.0. It is a hyperbolic tangent function used to smoothly map the effect of pressure gradients. The correction function is based on fluid dynamics principles, taking into account the characteristics that adverse pressure gradients (positive values) promote turbulent separation, while compromising pressure gradients (negative values) inhibit turbulent separation. When a strong adverse pressure gradient appears on the blade surface, the correction function value increases, increasing the probability of turbulent separation; when a compromising pressure gradient appears, the correction function value decreases, reducing the probability of turbulent separation.

[0102] The probability of turbulent separation ranges from [0,1], with higher values ​​indicating a greater likelihood of turbulent separation. This probability calculation method based on variance ratio can effectively identify regions with abnormal flow states, providing a reliable basis for feature point selection.

[0103] Sampling points with a turbulence separation probability exceeding a preset probability threshold are identified and marked as candidate turbulence separation feature points. Threshold screening is a crucial step in the initial determination of feature points, filtering out a large number of low-probability non-feature points and improving the efficiency and accuracy of subsequent processing. The preset probability threshold is typically set between 0.7 and 0.85, dynamically adjusted according to wind conditions and blade characteristics. The more complex the wind conditions, the lower the threshold is set to capture more possible separation features. The screening process employs an adaptive threshold strategy, applying different thresholds to different blade regions. The thresholds are relatively lowered at the leading and trailing edges of the blade, taking into account the more sensitive characteristics of turbulence separation in these regions. The screening result is a set of discrete candidate feature points distributed in areas with high turbulence separation probability on the blade surface, providing input for subsequent spatial connectivity analysis.

[0104] Spatial connectivity analysis is performed on candidate turbulence separation feature points to eliminate isolated points. The geometric centers of connected regions are then used as the final turbulence separation feature points, forming a set of local turbulence separation feature points for the blade. Connectivity analysis is a crucial step in improving the reliability of feature points, based on the physical characteristic that turbulence separation is typically regionally distributed. The analysis process employs a region growing algorithm, using each candidate point as a seed to aggregate adjacent points according to a preset spatial distance threshold (usually 1-3% of the blade chord length) to form connected regions. Regions smaller than the preset threshold (usually 3-5 points) are considered isolated points and eliminated to avoid interference from noise points. For each effective connected region, its geometric center (weighted average, with turbulence separation probability as the weight) is calculated as the representative feature point of that region. The final feature point set avoids redundant representations while preserving the spatial distribution characteristics of key separation regions, providing accurate turbulence separation characterization for subsequent calculations of wake-sensitive regions.

[0105] A time-series tracking method was used to analyze the set of local turbulent separation feature points on the blade, removing transient noise points and retaining those that appeared periodically or quasi-periodically. Time tracking is a crucial step in improving the temporal stability of feature points, distinguishing between persistent turbulent separation and transient disturbances. The tracking process is based on the spatiotemporal consistency of feature points, using a Kalman filter algorithm to predict the positional changes of feature points in consecutive time frames and establishing cross-frame correspondences. For each feature point, its frequency of occurrence within the observation window (typically 10-30 seconds) was statistically analyzed; points with frequencies below a preset threshold (typically 30-50%) were marked as transient noise points and removed. The retained feature points exhibit significant temporal persistence, reflecting a stable turbulent separation structure on the blade surface, greatly improving the reliability and representativeness of the feature point set, and laying the foundation for accurate identification of wake-sensitive areas.

[0106] In this embodiment of the invention, the detailed implementation steps for calculating the wake interference sensitive area of ​​the downwind wind turbine generator, combining the wake propagation dynamic topology diagram and the set of local turbulence separation feature points on the blade, include:

[0107] The wake influence core area in the wake propagation dynamic topology map is projected onto the blade spatial coordinate system of the downwind wind turbine to generate a wake interference projection map. Spatial projection is a fundamental step in establishing the relationship between wake influence and blade position, converting the global wake distribution into an influence representation in the local coordinate system of the blade. The projection process employs a coordinate transformation method, considering wind turbine rotation and blade pitch angle, mapping the wake influence core area in the global coordinate system to the blade surface coordinate system. The transformation process includes translation, rotation, and scaling adjustments to ensure the accuracy and consistency of the projection results. The projection result is a two-dimensional image representing the coverage and intensity distribution of the wake influence core area on the blade surface, providing a spatial mapping basis for overlap calculation.

[0108] Based on the wake interference projection map, the spatial overlap between the wake influence core region and the set of characteristic points of local turbulence separation on the blade is calculated. The spatial overlap is determined by a weighted sum of the area ratio of the overlapping region and the turbulence intensity. Spatial overlap is a key indicator for quantifying the coupling degree between wake interference and turbulence separation, reflecting their spatial matching. The calculation process first identifies the spatial intersection between the wake interference projection map and the turbulence separation characteristic points, and then evaluates the area ratio and turbulence intensity distribution of the intersection region. The formula for calculating spatial overlap is:

[0109] ;

[0110] in, Spatial overlap, The area of ​​the overlapping region. The total area of ​​the blades. For the first Turbulence intensity at a point For the first The probability of turbulent separation at a point. This is the adjustment coefficient (usually taken as 0.3-0.7).

[0111] Spatial overlap takes into account two key factors: the spatial extent of the overlapping area and the turbulence intensity within the overlapping area. It comprehensively evaluates the spatial coupling effect of wake interference and turbulence separation, providing a quantitative basis for the identification of sensitive areas.

[0112] Spatial overlap is quantified in a hierarchical manner to identify blade regions where the overlap exceeds a preset threshold, marking them as candidate sensitive areas for wake interference. Hierarchical quantization is a crucial step in the initial screening of sensitive areas, mapping continuous overlap to discrete sensitivity levels. The quantization process employs a multi-threshold segmentation method, dividing the blade surface into high, medium, and low-level sensitive areas based on overlap values. The preset overlap threshold is typically set between 0.6 and 0.8, dynamically adjusted according to wind conditions and wake intensity; the more significant the wake influence, the lower the threshold is set to capture a wider range of sensitive areas. The screening result is a set of candidate sensitive areas with overlap exceeding the threshold, distributed at blade locations where wake interference and turbulence separation are highly coupled, providing spatial range for further pressure gradient analysis.

[0113] Based on the pressure distribution of the turbulent boundary layer on the blade surface, the pressure gradient anomaly index within the candidate wake disturbance sensitive region is calculated. The pressure gradient anomaly index is determined by the deviation between the local extrema of the pressure gradient and the global mean. Pressure gradient analysis is a crucial step in verifying the effectiveness of the sensitive region, based on the physical phenomenon that wake disturbances typically lead to anomalies in the pressure gradient on the blade surface. The calculation process first extracts the pressure gradient field within the candidate region, then identifies local extrema points and calculates their deviation from the global mean. The formula for calculating the pressure gradient anomaly index is:

[0114] ;

[0115] in, For the region Pressure gradient anomaly index, For the region The number of sampling points within, For the first Pressure gradient at a point The mean of the global pressure gradient. denoted as the standard deviation of the pressure gradient.

[0116] The anomaly index is standardized to eliminate the influence of dimensions, facilitating comparisons between different regions. A higher index value indicates a more significant deviation in the pressure gradient within the region and a more pronounced effect from wake interference. This pressure gradient-based verification method provides a fluid dynamics perspective for identifying sensitive areas, enhancing the physical reliability of the identification results.

[0117] Candidate wake interference sensitive regions whose pressure gradient anomaly index exceeds a preset anomaly threshold are identified as the final wake interference sensitive regions. Threshold screening is a crucial step in determining the sensitive regions, ensuring the accuracy and reliability of the identification results. The preset anomaly threshold is typically set between 1.5 and 2.5, dynamically adjusted based on blade characteristics and wake features. Higher thresholds are set for more complex blade aerodynamic designs to identify the most significant anomaly regions. The screening process considers the influence of region size, appropriately lowering the threshold requirement for larger candidate regions to reflect the spatial extension characteristics of wake interference. The final wake interference sensitive regions are blade regions that simultaneously satisfy both spatial overlap and pressure gradient anomaly verifications, accurately characterizing the key parts most susceptible to wake interference and providing precise spatial targets for subsequent load prediction.

[0118] In this embodiment of the invention, the detailed implementation steps for constructing the vibration response transfer function under wake disturbance using the unit vibration modal parameters include:

[0119] Spectral decomposition was performed on the vibration modal parameters of the wind turbine generator to extract the dominant frequency and damping ratio of each mode. The dominant frequency and damping ratio characterize the dynamic response of the wind turbine generator under wake interference. Spectral decomposition is a fundamental step in analyzing vibration characteristics, breaking down complex vibration signals into modal components with clear physical meaning. The decomposition process employed an improved Fast Fourier Transform and peak identification technique to extract the main frequency components and corresponding damping characteristics from the vibration time-domain signal. For each identified mode, its damping ratio was calculated using the half-power bandwidth method.

[0120] ;

[0121] in, For the first Damping ratio of each mode, The dominant oscillation frequency, and This is the frequency corresponding to the half-power point.

[0122] The dominant frequency typically includes key components such as tower bending mode (0.2-0.5Hz), blade bending mode (1-3Hz), and transmission system torsional mode (5-10Hz). These modal parameters comprehensively characterize the dynamic characteristics of the wind turbine under wake interference and provide basic data for the construction of the transfer function.

[0123] Based on the dominant oscillation frequency, frequency domain response basis functions for the vibration modes are constructed. These basis functions are determined by the harmonic components of the dominant oscillation frequency and the attenuation factor of the damping ratio. The frequency domain response basis functions are mathematical expressions describing the dynamic characteristics of a single mode, providing standardized basic units for the transfer function. The construction process is based on the frequency response function theory of second-order vibration systems, considering the physical characteristics of resonance amplification and damping attenuation. The frequency domain response basis functions adopt the standard second-order system form:

[0124] ;

[0125] in, For the first The frequency domain response basis functions of each mode. For frequency variables, The dominant frequency refers to the natural frequency at which the structural components of a wind turbine tend to vibrate freely after being subjected to external excitation. For the damping ratio, It is the imaginary unit.

[0126] The frequency domain response basis functions intuitively express the vibration response characteristics under different frequency excitations. The response amplitude is significantly amplified near the resonance point and rapidly decays away from the resonance point. This basis function construction method based on a physical model ensures the physical rationality and mathematical accuracy of the transfer function, providing a standardized expression for subsequent modulation coefficient calculations.

[0127] Using the wake influence core region in the wake propagation dynamic topology diagram, the modulation coefficients of the wake interference on the frequency domain response basis functions are calculated. These modulation coefficients are determined by the spatial distribution of wake turbulence intensity and the spatial coupling degree between the vibration modes. The modulation coefficients are key parameters connecting wake interference and vibration response, quantifying the influence of wake characteristics on each vibration mode. The calculation process first analyzes the turbulence characteristics of the wake influence core region, including turbulence intensity distribution and spectral components, and then evaluates its spatial coupling degree with each vibration mode. The formula for calculating the modulation coefficients is:

[0128] ;

[0129] in, For the first The modulation coefficients of each mode, For position Turbulence intensity at that location For the first The modality at position The mode shape at that location, Let be the power spectral density function of the turbulence intensity, characterizing the distribution of turbulence energy in the frequency domain. It is obtained by taking the square of the modulus after performing a Fourier transform on the turbulence intensity time series; the integration range is... The wake influence area; This represents the energy density of turbulence intensity at zero frequency, corresponding to the average energy level of the turbulence.

[0130] The modulation coefficients take into account three key factors: the spatial distribution of turbulence intensity, the spatial distribution of mode shapes, and the matching degree between the turbulence spectrum and the mode frequency. This comprehensively evaluates the impact of wake interference on each mode and provides key parameters for the accurate construction of the transfer function.

[0131] The vibration response transfer function is generated by convolving the modulation coefficients with the frequency domain response basis functions. Convolution is the core step in constructing the complete transfer function, realizing the transformation from basic response characteristics to the response under actual wake excitation. The operation is based on linear system theory, using the modulation coefficients as weights to perform a weighted combination of the response basis functions of each mode. The mathematical expression of the vibration response transfer function is:

[0132] ;

[0133] in, For the complete vibration response transfer function, For the first The modulation coefficients of each mode, For the first The frequency domain response basis functions of each mode. The total number of modes to be considered.

[0134] transfer function It comprehensively describes the frequency domain mapping relationship between wake interference input and vibration response output, taking into account both the dynamic characteristics of the wind turbine itself and the specific effects of wake interference, and provides a mathematical model for dynamic gain factor identification.

[0135] The vibration response transfer function is validated in the time domain, and its parameters are iteratively optimized using the response error of actual vibration signals. Time-domain validation is a crucial step in ensuring the accuracy of the transfer function, verifying and correcting model performance through measured data. The validation process employs a time-domain excitation-response comparison method, inputting the measured wake interference signal into the transfer function model to generate a predicted response, which is then compared with the measured vibration response to calculate the error index. The optimization strategy uses a genetic algorithm, with minimizing the root mean square error as the objective function, to fine-tune key parameters (dominant frequency, damping ratio, and modulation coefficient) in the transfer function. Iterative optimization typically involves 5-10 rounds until the error is reduced to an acceptable range (usually 5-10%) or the maximum number of iterations is reached. Time-domain validation and optimization ensure that the transfer function not only has a sound theoretical basis but also possesses practical application accuracy, laying the foundation for the reliable identification of dynamic gain factors.

[0136] In this embodiment of the invention, the detailed implementation steps for identifying the dynamic gain factor of wake interference on the downwind wind turbine load through the vibration response transfer function include:

[0137] Based on the vibration response transfer function, the spectral characteristics of the wake interference input signal are calculated. The wake interference input signal is determined by the wake turbulence intensity sequence in the wake propagation dynamic topology diagram. Spectral characteristic analysis is a fundamental step in identifying the wake excitation characteristics, revealing the frequency composition and energy distribution of the wake interference. The calculation process first extracts the time series of wake turbulence intensity from the wake propagation dynamic topology diagram, then transforms it to the frequency domain using a fast Fourier transform to obtain the power spectral density function. Spectral characteristics typically include key indicators such as the dominant frequency component (related to the wind turbine speed and blade passage frequency), frequency band energy distribution, and spectral peak values, comprehensively characterizing the dynamic properties of the wake interference and providing input features for vibration response prediction.

[0138] Using the vibration response transfer function, the wake interference input signal is mapped to the vibration output signal of the downwind wind turbine, generating a vibration output spectrum. Frequency domain mapping is the core step in predicting the vibration response. Based on the frequency domain analysis theory of linear systems, the input is multiplied by the transfer function to obtain the output response. The mapping process uses frequency domain multiplication, and the calculation formula is as follows:

[0139] ;

[0140] in, For the vibration output spectrum, The input spectrum is for wake interference. This is the vibration response transfer function.

[0141] The output spectrum visually displays the response amplitude of each frequency component, reflecting the vibration characteristics of the fan under wake interference, and providing a frequency domain characterization for gain peak extraction.

[0142] Gain peaks are extracted from the vibration output spectrum, where the gain peak characterizes the amplification effect of wake interference on the vibration response. Gain peaks are a key indicator for identifying dynamic gain characteristics, quantifying the degree of amplification of wake interference at a specific frequency. The extraction process employs a peak detection algorithm to identify local maxima in the output spectrum where the amplitude is significantly higher than surrounding frequencies. For each identified peak, its gain coefficient is calculated.

[0143] ;

[0144] in, For the first Gain coefficient for each peak, For the output spectrum in The amplitude at that point, For the input spectrum in The amplitude at that point.

[0145] A gain coefficient greater than 1 indicates an amplification effect at that frequency; the larger the value, the more significant the amplification. Peak analysis typically focuses on frequency regions similar to the key modes of the wind turbine, as these regions are more prone to resonant amplification and have a more significant impact on the load.

[0146] Based on the spatial distribution of wake interference-sensitive areas, a weighted average of the gain peak value across different blade regions is calculated, with the spatial overlap of the wake interference-sensitive areas used as the weight. Spatial weighting is a crucial step in considering the differences between blade regions, reflecting the distribution characteristics of the gain effect on the blade surface. The calculation process first maps the gain peak value to the blade surface coordinate system, then spatially correlates it with the wake interference-sensitive areas, calculating the weighted gain value for each region. The weighted average calculation formula is as follows:

[0147] ;

[0148] in, This is the weighted average gain value. For the first Spatial overlap of sensitive areas This represents the gain coefficient for that region. This represents the total number of sensitive areas.

[0149] The weighted average takes into account the differences in importance of sensitive areas, with areas of high overlap receiving greater weight, thus better representing the main impact of wake interference. This spatial weighting method ensures the relevance and representativeness of the gain assessment, providing a comprehensive evaluation for the final dynamic gain factor.

[0150] The weighted average value is normalized and used as the dynamic gain factor. Normalization is a key step in standardizing the gain expression, facilitating comparisons between different operating conditions and model application. The processing employs a linear mapping to map the original weighted average value to a standard range (typically [1,3]) while maintaining the relative magnitudes. The dynamic gain factor is a dimensionless parameter that intuitively expresses the overall amplification effect of wake disturbance on downwind turbine loads. A value of 1 indicates no amplification, while a value greater than 1 indicates amplification, with larger values ​​indicating more significant amplification. This factor has both a clear physical meaning and a concise mathematical form, providing key parameter support for load prediction correction models.

[0151] In this embodiment of the invention, the detailed implementation steps for generating a real-time load prediction correction model for downwind wind turbine generators based on the wake interference sensitive area and dynamic gain factor include:

[0152] An initial load prediction model was constructed, taking wind speed, blade angle of attack, and turbine speed as inputs and blade root stress as output. This initial model forms the basic framework for load prediction, providing standard prediction capabilities without considering wake interference. The model construction process employed a multivariate nonlinear regression method, establishing a mapping relationship between input parameters and load output based on historical operating data. The model structure utilized machine learning algorithms such as Support Vector Regression (SVR) or Random Forest, which exhibit good nonlinear fitting ability and generalization performance. Cross-validation was employed during model training to avoid overfitting and ensure prediction accuracy under normal operating conditions. The initial model provides the foundational prediction values ​​for subsequent wake interference corrections; its accuracy and stability directly impact the final prediction results.

[0153] Based on the wake disturbance sensitive region, the disturbance component of the wake disturbance on the input parameters of the initial load prediction model is extracted. This disturbance component is determined through the coupling effect of the wake turbulence intensity and the set of local turbulence separation feature points on the blade. Disturbance component extraction is a crucial step in capturing the wake influence, quantifying the modification effect of wake disturbance on the standard input parameters. The extraction process first analyzes the flow field characteristics within the wake disturbance sensitive region, including velocity deviation, turbulence intensity, and angle-of-attack fluctuations, and then assesses the degree of influence of these characteristics on the model input parameters. The formula for calculating the disturbance component is:

[0154] ;

[0155] in, For the first The perturbation components of each input parameter The influence coefficient, Sensitive area For the first The weighting coefficient of the point. For the first Turbulence intensity at a point For the first Turbulent separation characteristic value at a point.

[0156] The disturbance component takes into account the turbulence intensity distribution and turbulence separation characteristics in the sensitive area, comprehensively evaluates the impact of wake disturbance on input parameters, and provides key input adjustment for load prediction correction.

[0157] The output of the initial load prediction model is adjusted using a dynamic gain factor to generate a corrected load prediction value. Gain correction is the core step considering the wake amplification effect, directly adjusting the amplitude level of the load prediction. The correction process employs a multiplicative adjustment strategy, applying the dynamic gain factor as a proportional coefficient to the initial prediction result to realize the load amplification effect under wake interference. The correction formula is:

[0158] ;

[0159] in, This is the corrected load prediction value. For dynamic gain factor, The initial load prediction function is based on a nonparametric modeling method using radial basis functions. This method can capture the nonlinear relationship between input parameters and loads and also has good generalization ability. Input parameters after considering disturbance components ( ), It is the original input parameter vector. It is a parameter disturbance component caused by wake interference.

[0160] Gain correction is simple and intuitive, computationally efficient, and can effectively capture the load amplification effect caused by wake interference, making it a key step in load prediction correction.

[0161] A time delay compensation mechanism is introduced, which determines the propagation time delay of wake interference by using the wake propagation path length and the average velocity of the velocity vector field in the wake propagation dynamic topology diagram. Time delay compensation is a key step considering the dynamic characteristics of wake propagation, solving the problem of the time difference between wake generation and its impact on downstream wind turbines. The compensation process first calculates the wake propagation distance and velocity based on the wake propagation dynamic topology diagram, and then estimates the propagation time delay:

[0162] ;

[0163] in, Due to the time lag in transmission, The length of the wake propagation path. Let be the average velocity of the velocity vector field.

[0164] The time delay calculation takes into account the complexity of the wake propagation path and the spatial variation of velocity, ensuring the accuracy of time compensation. The compensation mechanism adopts a forward prediction strategy, predicting the wake impact that downstream wind turbines may be affected by in the future based on the current status of upstream wind turbines, giving the prediction model "foresight" and greatly improving the timeliness and accuracy of real-time prediction.

[0165] The corrected load prediction values ​​are combined with a time delay compensation mechanism to generate a real-time load prediction correction model. Integration is the final step in forming a complete prediction model, unifying the various correction components into a coordinated prediction framework. The integration process adopts a layered design: the bottom layer is the initial load model, the middle layer is the input disturbance and output gain correction, and the top layer is the time delay compensation, forming a complete correction model structure. The model input includes standard operating parameters and wake characteristic parameters. The processing flow covers parameter correction, initial prediction, gain adjustment, and time compensation, and the final output is a real-time load prediction value that takes into account the impact of wake interference. The integrated model has comprehensive wake interference response capabilities and can accurately predict load changes of downwind wind turbines in complex wake environments, providing key technical support for wind turbine group operation monitoring and scheduling optimization.

[0166] In this embodiment of the invention, the detailed implementation steps for adaptive optimization based on the deviation between the predicted value and the actual load include:

[0167] Real-time data acquisition of actual loads from downwind wind turbines is performed, with the data obtained through strain sensors at the blade roots. Real-time acquisition forms the data foundation for model validation and optimization, providing objective load reference values. The acquisition process is based on a distributed sensor network, with a high-precision strain sensor array installed at the blade roots. Continuous load time series are acquired through signal conditioning and a data acquisition system. The sampling frequency is typically set to 20-50Hz, sufficient to capture key load dynamic characteristics. Data preprocessing includes noise filtering, outlier detection, and time synchronization to ensure the quality and reliability of the acquired data. Real-time load data serves as both a benchmark for evaluating prediction accuracy and a basis for optimizing model parameters, providing necessary feedback for adaptive optimization.

[0168] The deviation sequence between the load prediction values ​​output by the real-time load prediction correction model and the actual load data is calculated. Deviation calculation is a fundamental step in evaluating prediction performance, quantifying the difference between the model output and the actual values. The calculation process employs a point-to-point comparison method, calculating the difference between time-aligned predicted and measured values. The deviation sequence is typically expressed in relative error form, eliminating the influence of absolute load magnitude and facilitating comparisons between different operating conditions.

[0169] ;

[0170] in, for The relative deviation of time, To predict load values, This represents the measured load value.

[0171] The bias sequence comprehensively records the performance of the prediction model at different times, providing detailed data support for subsequent time-frequency analysis and parameter adjustment.

[0172] Time-frequency analysis is performed on the deviation sequence to extract the dominant frequency component and amplitude component. The dominant frequency component characterizes the periodicity of the deviation, while the amplitude component characterizes its intensity. Time-frequency analysis is a crucial step in understanding the deviation characteristics, revealing the intrinsic structure and variation patterns of the prediction error. The analysis process employs short-time Fourier transform or wavelet analysis to decompose the deviation sequence into different frequency components and track their time-varying characteristics. The dominant frequency component is identified through power spectrum analysis, selecting the top 3-5 frequency bands with the highest energy. The amplitude component is obtained through root mean square calculation, reflecting the intensity level of the deviation within each frequency band. The time-frequency characteristics comprehensively characterize the dynamic properties of the prediction deviation, providing precise guidance for targeted adjustment of model parameters.

[0173] Based on the dominant frequency component, the damping parameters of the vibration response transfer function in the real-time load prediction correction model are adjusted. Damping adjustment is a key step in optimizing frequency response characteristics, correcting for the periodic characteristics of prediction deviations. The adjustment strategy is based on the frequency matching principle, finely adjusting the transfer function bands close to the dominant deviation frequency. The damping parameter adjustment formula is:

[0174] ;

[0175] in, and These are the damping ratios before and after adjustment, respectively. For the first The percentage of deviation energy in each frequency band This is the learning rate coefficient (usually between 0.1 and 0.3).

[0176] Damping adjustment is highly targeted and can effectively improve the model's predictive performance in specific frequency ranges, making it a core component of frequency domain optimization. The adjustment process is highly adaptive; the adjustment amplitude is larger for frequency bands with higher deviation energy, ensuring efficient allocation of optimization resources.

[0177] Based on the amplitude components, the weighting coefficients of the dynamic gain factor in the real-time load prediction correction model are adjusted. Weight adjustment is a crucial step in optimizing the prediction amplitude and directly affects the overall level of load prediction. The adjustment process is based on the statistical characteristics of amplitude deviation, proportionally correcting the dynamic gain factor. The weighting coefficient adjustment formula is as follows:

[0178] ;

[0179] in, and These are the dynamic gain factors before and after adjustment, respectively. The average value of the deviation sequence. This is the proportionality coefficient (usually taken as 0.5-1.5).

[0180] The weight adjustment mechanism is simple and efficient, capable of quickly responding to and compensating for system deviations, and is a key element in amplitude optimization. The adjustment logic conforms to the principle of negative feedback: when the predicted value is too high, the gain is reduced; when it is too low, the gain is increased, ensuring that the system gradually approaches a balanced state.

[0181] The adjusted real-time load prediction correction model undergoes stability testing. Optimization is considered complete when the deviation amplitude remains below a preset deviation threshold for multiple consecutive frames. Stability testing is a crucial step in ensuring the effectiveness of optimization, preventing over-adjustment and oscillations. The testing process uses a sliding observation window (typically 30-60 seconds) to monitor the statistical characteristics of the deviation within the window in real time. Multiple standards are used to judge stability, including average deviation, maximum deviation, and standard deviation of deviation. Optimization is confirmed only when all indicators remain below their respective preset thresholds for multiple consecutive frames (typically 10-20 frames). Stability testing ensures the convergence and reliability of the optimization process, prevents blind adjustment and over-optimization of model parameters, and provides a guarantee for the long-term stable operation of the system.

[0182] In this embodiment of the invention, a dynamic risk assessment model for the wind turbine cluster array effect is also constructed, and the operation scheduling strategy is optimized. The detailed implementation steps include:

[0183] A dynamic risk assessment model for the array effect of wind turbine clusters is constructed. This model uses the outputs of a wake propagation dynamic topology map and a real-time load prediction correction model as inputs to evaluate the fatigue damage accumulation rate of downwind wind turbine generators. Risk assessment is a crucial step in ensuring the safe operation of wind turbines, quantifying the long-term impact of wake interference on turbine lifespan. The assessment model is based on fatigue damage theory, converting the additional load caused by wake interference into accumulated fatigue damage. The calculation process first obtains the long-term load sequence from the load prediction correction model, then statistically analyzes the load cyclic distribution using the rainflow counting method, and finally applies Miner's linear cumulative damage theory to calculate the fatigue damage rate.

[0184] ;

[0185] in, This represents the cumulative fatigue damage value. Stress amplitude The number of loops, For the corresponding stress level of fatigue life, and These are the parameters for the material's SN curve.

[0186] The fatigue damage accumulation rate directly reflects the rate at which the lifespan of the wind turbine is consumed under the current wake environment. It is a core indicator for risk assessment and provides a quantitative basis for scheduling optimization.

[0187] Based on the cumulative rate of fatigue damage, high-risk areas of wake interference in the wind turbine cluster are identified, and a spatial distribution map of these risk areas is generated. Risk area identification is a crucial step in risk visualization, visually displaying the locations of the most severe wake interference in the wind farm. The identification process first sets a risk threshold (typically 1.5-2 times the fatigue damage rate under normal operating conditions), then filters out wind turbine locations with damage rates exceeding the threshold, forming a set of high-risk points. The spatial distribution map uses a heatmap format, with color depth representing risk levels, covering the entire wind farm area, and specifically marking the boundaries and centers of high-risk areas. The distribution map not only displays the current risk status but also predicts the risk evolution trend over the next 3-6 hours through time-series analysis, providing forward-looking guidance for operation scheduling. The accurate identification and visual display of risk areas provide spatial targets for the development of targeted optimization measures.

[0188] Using a spatial distribution map of risk areas, the operation and scheduling strategy of the wind turbine cluster is optimized. This strategy includes adjusting the yaw angle and blade pitch angle of upstream wind turbines to reduce the impact of wake interference on downwind wind turbines. Scheduling optimization is a proactive intervention to reduce risk by altering the operating parameters of upstream turbines to mitigate wake effects. The optimization process employs a multi-objective genetic algorithm, with the dual objectives of minimizing fatigue damage rate in high-risk areas and minimizing power generation loss, to search for the optimal scheduling scheme. The scheduling strategy mainly includes two types of measures: yaw angle adjustment, which changes the orientation of upstream turbines to deflect the wake away from downwind turbines; and blade pitch angle adjustment, which increases the blade pitch angle of upstream turbines to reduce the thrust coefficient and decrease wake intensity. These two measures are dynamically combined based on wind conditions and risk levels to form the optimal scheduling strategy. The optimization process considers adjustment range limitations (yaw angle typically ±8°, blade pitch angle increases by 1-3°) and execution frequency constraints (avoiding frequent adjustments) to ensure the feasibility and stability of the strategy.

[0189] The optimized operation and scheduling strategy underwent closed-loop verification. The effectiveness of the strategy was evaluated by comparing the actual load changes of downwind wind turbines before and after optimization. Closed-loop verification is a crucial step in ensuring the effectiveness of the optimization, using actual operational data to test the theoretical optimization results. The verification process employed a comparative experimental design. Under similar wind conditions, wind turbine load data were recorded before and after implementing the scheduling strategy, and the percentage changes in key indicators were calculated. Evaluation indicators included the reduction rate of average load, the reduction rate of peak load, the reduction rate of load fluctuation amplitude, and the reduction rate of fatigue damage accumulation, comprehensively reflecting the actual effect of the scheduling optimization. The verification results passed statistical significance testing, ensuring that the observed changes truly stemmed from scheduling optimization rather than random fluctuations. Closed-loop verification not only evaluated the effectiveness of the current strategy but also provided data support for iterative optimization, forming a virtuous cycle of "optimization-verification-re-optimization," ensuring continuous improvement and long-term effectiveness of wind turbine fleet operation and scheduling.

[0190] The above describes a monitoring and sensing method based on a large wind turbine in an embodiment of this application. The following describes a monitoring and sensing system based on a large wind turbine in an embodiment of this application. This system is used to implement the above-described monitoring and sensing method based on a large wind turbine. Please refer to... Figure 2 One embodiment of a monitoring and sensing system based on a large wind turbine in this application includes:

[0191] The data acquisition module is used to acquire real-time operating status data sequences of each wind turbine in the wind turbine group. The real-time operating status data sequences include the pressure distribution of the turbulent boundary layer on the blade surface, the velocity vector field of the upstream wake, and the vibration mode parameters of the unit.

[0192] The wake topology construction module is used to construct a dynamic topology diagram of wake propagation based on the upstream wake velocity vector field.

[0193] The turbulence separation feature extraction module is used to extract the set of local turbulence separation feature points of the blade based on the pressure distribution of the turbulent boundary layer on the blade surface, and to calculate the wake interference sensitive area of ​​the downwind wind turbine by combining the wake propagation dynamic topology map.

[0194] The vibration response analysis module is used to construct the vibration response transfer function under wake interference using the unit's vibration mode parameters, and to identify the dynamic gain factor of wake interference on the downwind wind turbine load through the vibration response transfer function.

[0195] The load prediction model generation module is used to generate a real-time load prediction correction model for downwind wind turbines based on the wake interference sensitive area and dynamic gain factor.

[0196] The adaptive optimization module is used to output the load prediction value of the downwind wind turbine based on the real-time load prediction correction model, and perform adaptive optimization by combining the deviation between the prediction value and the actual load.

[0197] This invention achieves accurate monitoring and prediction of load changes under wake interference in large wind turbines through wake propagation dynamic topology construction, turbulence separation feature extraction, vibration response transfer function analysis, and real-time load prediction correction. The adaptive optimization feature of this invention can adjust prediction parameters in real time according to differences in the wind farm environment and wake interference characteristics, significantly improving prediction accuracy and reducing additional load damage to the wind turbine caused by wake interference.

[0198] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0199] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0200] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A monitoring and sensing method based on large wind turbines, characterized in that, include: The real-time operating status data sequence of each wind turbine in the wind turbine group is obtained. The real-time operating status data sequence includes the pressure distribution of the turbulent boundary layer on the blade surface, the velocity vector field of the upstream wake, and the vibration mode parameters of the unit. Based on the upstream wake velocity vector field, a dynamic topology diagram of wake propagation is constructed; Based on the pressure distribution of the turbulent boundary layer on the blade surface, the set of local turbulent separation feature points of the blade is extracted, and combined with the wake propagation dynamic topology map, the wake interference sensitive area of ​​the downwind wind turbine is calculated. Using the vibration mode parameters of the unit, a vibration response transfer function under wake interference is constructed, and the dynamic gain factor of wake interference on the load of downwind wind turbine is identified through the vibration response transfer function. Based on the wake interference sensitive area and the dynamic gain factor, a real-time load prediction correction model for downwind wind turbine generators is generated. Based on the real-time load prediction correction model, the load prediction value of the downwind wind turbine is output, and adaptive optimization is performed by combining the deviation between the prediction value and the actual load.

2. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The construction of the wake propagation dynamic topology graph includes: The upstream wake velocity vector field is discretized in three dimensions to generate a wake velocity vector grid. Each grid node of the wake velocity vector grid contains a velocity component and a turbulence intensity component. Based on the wake velocity vector grid, the wake energy transfer coefficient between adjacent grid nodes is calculated. The wake energy transfer coefficient is determined by the gradient direction of the velocity component and the spatial attenuation rate of the turbulence intensity. Using the wake energy transfer coefficient, a directed weighted graph of wake propagation is constructed, wherein the edge weights of the directed weighted graph are characterized by the wake energy transfer coefficient, and the nodes represent the grid nodes in the wake velocity vector grid. Dynamic clustering analysis is performed on the directed weighted graph to identify high-energy concentrated regions in the wake propagation path, and the spatial boundaries of the high-energy concentrated regions are marked as the wake influence core area; The wake influence core area is geometrically mapped to the spatial location of the downwind wind turbine, generating the wake propagation dynamic topology map.

3. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The extracted set of local turbulence separation feature points for the blade includes: Wavelet decomposition is performed on the pressure distribution of the turbulent boundary layer on the blade surface to extract high-frequency pressure fluctuation components, which characterize the local disturbance features of turbulent separation on the blade surface. Based on the high-frequency pressure fluctuation component, the turbulence separation probability of each sampling point on the blade surface is calculated. The turbulence separation probability is determined by the ratio of the local variance to the global variance of the pressure fluctuation component. Identify sampling points whose turbulence separation probability exceeds a preset probability threshold and mark them as candidate turbulence separation feature points; Spatial connectivity analysis is performed on the candidate turbulence separation feature points to eliminate isolated points, and the geometric center of the connected region is taken as the final turbulence separation feature point to form the set of local turbulence separation feature points of the blade. The set of local turbulence separation feature points of the blade is tracked over time to remove transient noise points and retain turbulence separation feature points that appear periodically or quasi-periodically.

4. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The calculation of the wake interference sensitive area of ​​the downwind wind turbine includes: The wake influence core area in the wake propagation dynamic topology map is projected onto the blade spatial coordinate system of the downwind wind turbine to generate a wake interference projection map. Based on the wake interference projection map, calculate the spatial overlap between the wake influence core area and the set of local turbulence separation feature points of the blade; The spatial overlap is graded and quantified to identify blade regions whose spatial overlap exceeds a preset overlap threshold and mark them as candidate wake interference sensitive regions. Based on the pressure distribution of the turbulent boundary layer on the blade surface, the pressure gradient anomaly index within the candidate wake disturbance sensitive region is calculated. Candidate wake interference sensitive regions whose pressure gradient anomaly index exceeds a preset anomaly threshold are determined as the final wake interference sensitive regions.

5. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The construction of the vibration response transfer function under wake disturbance includes: The vibration mode parameters of the unit are subjected to spectral decomposition to extract the dominant frequency and damping ratio of each mode. The dominant frequency and damping ratio characterize the dynamic response characteristics of the wind turbine generator under wake interference. Based on the dominant frequency, a frequency domain response basis function for the vibration mode is constructed. The frequency domain response basis function is determined by the harmonic components of the dominant frequency and the attenuation factor of the damping ratio. Using the wake influence core region in the wake propagation dynamic topology diagram, the modulation coefficient of the wake interference on the frequency domain response basis function is calculated; The vibration response transfer function is generated by convolving the modulation coefficients with the frequency domain response basis function. The vibration response transfer function is verified in the time domain, and the parameters of the transfer function are iteratively optimized using the response error of the actual vibration signal.

6. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The dynamic gain factor for identifying the impact of wake interference on the load of downwind wind turbine generators includes: Based on the vibration response transfer function, the spectral characteristics of the wake interference input signal are calculated. The wake interference input signal is determined by the wake turbulence intensity sequence in the wake propagation dynamic topology diagram. Using the vibration response transfer function, the wake interference input signal is mapped to the vibration output signal of the downwind wind turbine generator to generate a vibration output spectrum; Extract the gain peak value from the vibration output spectrum; Based on the spatial distribution of the wake interference sensitive area, the weighted average value of the gain peak value in different blade regions is calculated; The weighted average value is normalized and used as the dynamic gain factor.

7. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The real-time load prediction correction model for generating downwind wind turbine generators includes: An initial load prediction model is constructed, which takes wind speed, blade angle of attack and unit speed as inputs and blade root stress as output. Based on the wake interference sensitive area, the disturbance component of the wake interference on the input parameters of the initial load prediction model is extracted; Using the dynamic gain factor, the output of the initial load prediction model is adjusted to generate a corrected load prediction value. A time delay compensation mechanism is introduced, which determines the propagation time delay of the wake interference by the wake propagation path length and the average velocity of the velocity vector field in the wake propagation dynamic topology graph. The corrected load prediction value is combined with the time delay compensation mechanism to generate the real-time load prediction correction model.

8. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The adaptive optimization based on the deviation between the predicted value and the actual load includes: Real-time acquisition of actual load data of downwind wind turbine generators, which is obtained by measuring strain sensors at the blade root; Calculate the deviation sequence between the load prediction value output by the real-time load prediction correction model and the actual load data; Time-frequency analysis is performed on the deviation sequence to extract the dominant frequency component and amplitude component of the deviation sequence. The dominant frequency component characterizes the periodicity of the deviation, and the amplitude component characterizes the intensity of the deviation. Based on the dominant frequency component, adjust the damping parameter of the vibration response transfer function in the real-time load prediction correction model; Based on the amplitude components, adjust the weight coefficients of the dynamic gain factor in the real-time load prediction correction model; The stability of the adjusted real-time load prediction correction model is tested. When the deviation amplitude of multiple consecutive frames is lower than the preset deviation threshold, the optimization is confirmed to be complete.

9. The monitoring and sensing method based on a large wind turbine generator according to claim 1, characterized in that, The method further includes: A dynamic risk assessment model for the wind turbine array effect is constructed. The dynamic risk assessment model takes the output of the wake propagation dynamic topology map and the real-time load prediction correction model as input to evaluate the fatigue damage accumulation rate of downwind wind turbine generators. Based on the fatigue damage accumulation rate, high-risk areas of wake interference in the wind turbine group are identified, and a spatial distribution map of the risk areas is generated. Using the spatial distribution map of the risk area, the operation and scheduling strategy of the wind turbine group is optimized. The operation and scheduling strategy includes adjusting the yaw angle and blade pitch angle of the upstream wind turbine. The optimized operation and scheduling strategy was validated in a closed loop by comparing the actual load changes of downwind wind turbines before and after optimization to assess the effectiveness of the strategy.

10. A monitoring and sensing system based on a large wind turbine, used to implement the monitoring and sensing method based on a large wind turbine as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire real-time operating status data sequences of each wind turbine generator in the wind turbine group. The real-time operating status data sequences include the pressure distribution of the turbulent boundary layer on the blade surface, the velocity vector field of the upstream wake, and the vibration mode parameters of the generator. The wake topology construction module is used to construct a dynamic topology diagram of wake propagation based on the upstream wake velocity vector field. The turbulence separation feature extraction module is used to extract the set of local turbulence separation feature points of the blade based on the pressure distribution of the turbulent boundary layer on the blade surface, and to calculate the wake interference sensitive area of ​​the downwind wind turbine in combination with the wake propagation dynamic topology map. The vibration response analysis module is used to construct the vibration response transfer function under wake interference using the vibration mode parameters of the unit, and to identify the dynamic gain factor of wake interference on the load of the downwind wind turbine unit through the vibration response transfer function. The load prediction model generation module is used to generate a real-time load prediction correction model for downwind wind turbine generators based on the wake interference sensitive area and the dynamic gain factor. The adaptive optimization module is used to output the load prediction value of the downwind wind turbine based on the real-time load prediction correction model, and to perform adaptive optimization by combining the deviation between the prediction value and the actual load.

Images