Self-adaptive variable pitch control method and system based on turbulence intensity dynamic compensation

Abstract

The invention provides a self-adaptive variable-pitch control method and system based on turbulence intensity dynamic compensation, and relates to the field of wind power generation. Firstly, a real-time turbulence data set in the operation environment of a wind turbine generator is obtained, and the real-time turbulence data set comprises multi-dimensional turbulence characteristic units which are continuously collected and provided with time sequence marks; performing feature extraction on the real-time turbulence data set to obtain turbulence feature distribution containing spatial distribution features and dynamic change features, then generating a dynamic compensation strategy containing a related compensation parameter set according to the turbulence feature distribution, and adjusting control parameters of a preset variable pitch control model based on the dynamic compensation strategy to obtain the variable pitch control model. And finally, a pitch angle adjusting mechanism of the wind turbine generator is controlled through the corrected parameters, self-adaptive adjustment of the pitch angle is achieved, and therefore the operation stability and the output power stability of the wind turbine generator under the complex turbulence working condition can be improved, the damage risk of parts can be reduced, and the service life of the wind turbine generator can be prolonged.

Description

Technical Field

[0001] The present invention relates to the field of wind power generation, and in particular to an adaptive variable pitch control method and system based on dynamic compensation of turbulence intensity. Background Art

[0002] In the field of wind power generation, the operational efficiency and stability of wind turbines are crucial for both power generation profitability and equipment safety. Pitch control is a key technology for regulating wind turbine output power and ensuring stable operation. Currently, traditional pitch control methods are mostly based on fixed control models and parameters. These parameters are typically set through experimentation or experience under specific operating conditions, making them difficult to adapt to the complex and changing real-world operating environment.

[0003] In actual operation, wind turbines operate in complex and diverse environments with turbulent conditions. Turbulence is characterized by uneven spatial distribution and rapid dynamic changes, with significant variations in turbulence intensity and characteristics at different locations and times. Traditional variable pitch control methods fail to fully account for these turbulent characteristics and are unable to dynamically adjust to real-time turbulence conditions. Consequently, in situations of strong or drastically changing turbulence, wind turbines may be unable to adjust the pitch angle in a timely and accurate manner, impacting output power stability and turbine safety. This can even cause fatigue damage to turbine components and shorten the turbine's service life. Summary of the Invention

[0004] In view of the above-mentioned problems, a first aspect of the present invention provides an adaptive pitch control method based on dynamic compensation of turbulence intensity, the method comprising:

[0005] Acquire a real-time turbulence data set in the operating environment of the wind turbine, wherein the real-time turbulence data set includes continuously collected multi-dimensional turbulence feature units with time series labels;

[0006] Performing feature extraction on the real-time turbulence data set to obtain a turbulence feature distribution including spatial distribution features and dynamic change features;

[0007] generating a dynamic compensation strategy according to the turbulence characteristic distribution, wherein the dynamic compensation strategy includes a set of compensation parameters associated with the spatial distribution characteristics and the dynamic change characteristics;

[0008] Adjusting the control parameters of a preset pitch control model based on the dynamic compensation strategy to obtain corrected pitch control parameters;

[0009] The pitch angle adjustment mechanism of the wind turbine generator set is controlled by utilizing the modified variable pitch control parameter to achieve adaptive adjustment of the pitch angle of the wind turbine generator set.

[0010] On the other hand, the present invention also provides an adaptive variable pitch control system based on dynamic compensation of turbulence intensity, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above-mentioned adaptive variable pitch control method based on dynamic compensation of turbulence intensity.

[0011] Based on the above aspects, the present invention obtains a real-time turbulence data set containing multi-dimensional turbulence feature units that are continuously collected and have time series labels, performs feature extraction on the real-time turbulence data set, and obtains a turbulence feature distribution containing spatial distribution features and dynamic change features. According to the turbulence feature distribution, a dynamic compensation strategy containing a compensation parameter set associated with the spatial distribution features and the dynamic change features is generated, thereby achieving a precise response to turbulence characteristics. Based on the dynamic compensation strategy, the control parameters of the preset variable pitch control model are adjusted to obtain the corrected variable pitch control parameters, so that the control model can adapt to real-time turbulence changes. Finally, the corrected variable pitch control parameters are used to control the pitch angle adjustment mechanism of the wind turbine, thereby achieving adaptive adjustment of the pitch angle, thereby significantly improving the operating stability and output power stability of the wind turbine under different turbulent conditions, reducing the fatigue damage risk of the unit components, and extending the service life of the unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the execution flow of the adaptive variable pitch control method based on dynamic compensation of turbulence intensity provided by an embodiment of the present invention.

[0013] Figure 2 It is a schematic diagram of exemplary hardware and software components of an adaptive variable pitch control system based on dynamic compensation of turbulence intensity provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of an adaptive variable pitch control method based on dynamic compensation of turbulence intensity provided by an embodiment of the present invention. The adaptive variable pitch control method based on dynamic compensation of turbulence intensity is introduced in detail below.

[0015] Step S110: obtaining a real-time turbulence data set in the operating environment of the wind turbine generator set, wherein the real-time turbulence data set comprises continuously collected multi-dimensional turbulence feature units with time series markers.

[0016] In the actual scenario of a wind farm, the operating performance of a wind turbine is easily affected by the surrounding turbulent environment. Therefore, in order to accurately grasp the turbulent characteristics and realize the effective control of the pitch angle of the wind turbine, it is necessary to continuously collect real-time turbulence data. In this embodiment, various types of sensors can be installed at different key positions around the wind turbine. For example, an anemometer and a wind vane are installed at a certain distance in front of the wind rotor to measure the wind speed and direction of the incoming flow; an accelerometer is installed on the top of the nacelle to detect the vibration of the wind turbine caused by turbulence; and pressure sensors are also arranged on the wind rotor blades to obtain the pressure distribution changes on the blade surface. The above pressure changes are closely related to the effects of turbulence.

[0017] The sensor can continuously collect data at a fixed sampling frequency. Each piece of data collected is assigned a time series tag. This is because turbulence is a process that changes dynamically over time. The time tag can clearly reflect the order and time interval of data collection.

[0018] Multidimensional turbulence feature units contain turbulence information. For example, at a specific moment, the wind speed recorded by the anemometer includes not only its magnitude but also its short-term fluctuations, that is, its instantaneous change in wind speed. The wind direction recorded by the wind vane also fluctuates to a certain extent, and its directional deflection angle is also an important characteristic. The vibration signal measured by the accelerometer contains components of different frequencies, reflecting the different forms of vibration caused by turbulence. The blade surface pressure data obtained by the pressure sensor can reflect the impact of turbulence on the blade's aerodynamic performance.

[0019] The various information from these sensors is combined to form a multi-dimensional turbulence feature unit. Over time, these multi-dimensional turbulence feature units are continuously collected to form a real-time turbulence data set that records the various characteristics of turbulence around the wind turbine at different times.

[0020] Step S120: performing feature extraction on the real-time turbulence data set to obtain a turbulence feature distribution including spatial distribution features and dynamic change features.

[0021] Because real-time turbulence data sets are large and complex, they contain a lot of redundant information that has little impact on wind turbine pitch angle adjustment. Therefore, it is necessary to deeply mine and analyze this data to extract features that truly reflect turbulence characteristics and have a significant impact on pitch angle adjustment. This will lead to a turbulence feature distribution that includes both spatial distribution characteristics and dynamic change characteristics.

[0022] Step S121: dividing the real-time turbulence data set into a plurality of turbulence data segments with overlapping time windows according to a time series, each of the turbulence data segments comprising turbulence data samples at a plurality of spatial positions.

[0023] To facilitate systematic analysis of real-time turbulence data sets, they need to be divided into appropriate time zones. First, a reasonable time window length is determined based on the changing characteristics of the turbulence and the response characteristics of the wind turbine. If the turbulence changes rapidly and drastically, the time window length can be set relatively short to capture the rapid changes in turbulence in more detail. Conversely, if the turbulence changes slowly, the time window length can be appropriately extended to reduce the data processing workload.

[0024] During the segmentation process, adjacent time windows are allowed to overlap to a certain extent. This is because turbulence changes continuously, and overlapping time windows ensure that key changes are not missed. Starting from the start time of the real-time turbulence data set, a section of data is intercepted within a set time window to form a turbulence data segment. This turbulence data segment covers turbulence data samples from multiple spatial locations within that time period. Because sensors are installed at different locations around the wind turbine, each turbulence data segment contains rich information from different spatial locations.

[0025] For example, four monitoring points can be set up at different spatial locations around a wind turbine: in front of the rotor, to the side of the rotor, on top of the nacelle, and at the base of the tower. During the time period corresponding to a turbulence data segment, the sensors at these four monitoring points each collect data such as wind speed, wind direction, and vibration. These data together constitute the turbulence data samples for multiple spatial locations within that turbulence data segment. The time window is then slid forward by a pre-set step size, and data is captured again to form a new turbulence data segment. This cycle repeats until the entire real-time turbulence data set is covered.

[0026] Step S122: performing statistical analysis on the turbulence data samples in each of the turbulence data segments in the spatial dimension, calculating the turbulence velocity pulsation amplitude, direction deflection angle and pulsation frequency parameters at each spatial position, and generating spatial distribution characteristics in the spatial dimension.

[0027] For each divided turbulence data segment, it is necessary to deeply analyze the turbulence data samples at each spatial position and extract valuable features from the spatial dimension.

[0028] Step S1221: for each turbulence data sample at each spatial position, extract the time domain waveform of the turbulence velocity signal, convert the time domain waveform into a frequency domain signal through fast Fourier transform, and identify the pulsation frequency parameter corresponding to the dominant frequency component.

[0029] At each spatial location, the turbulent velocity data collected by the sensor exists as a time-domain waveform. This waveform intuitively displays how wind speed changes over time. However, to gain a deeper understanding of the frequency characteristics of turbulent velocity, it must be converted to the frequency domain. Fast Fourier transforms (FFTs) decompose time-domain signals into sine and cosine components of different frequencies, yielding frequency-domain signals.

[0030] In a frequency domain signal, the amplitude of different frequency components reflects the energy contribution of that frequency component in the original signal. By analyzing the frequency domain signal, the frequency component with the largest amplitude is identified. This frequency component is then considered the dominant frequency component, and its corresponding frequency value is the pulsation frequency parameter. The pulsation frequency parameter is crucial for understanding the periodic variations in turbulence. Different pulsation frequencies may correspond to different turbulence generation mechanisms and have different impacts on wind turbines. For example, pulsations of certain frequencies may cause resonance in wind turbine blades, threatening the structural safety of the blades.

[0031] Step S1222: performing amplitude spectrum analysis on the frequency components in the frequency domain signal that are higher than a preset frequency threshold, and calculating the weighted sum of the amplitudes of the frequency components as the turbulent velocity pulsation amplitude.

[0032] After obtaining the frequency domain signal, not all frequency components significantly affect the pulsation amplitude of the turbulent velocity. Therefore, a preset frequency threshold is set, and only frequency components above this threshold are considered. The setting of the preset frequency threshold requires comprehensive consideration of the physical properties of turbulence and the response characteristics of the wind turbine. Generally speaking, lower-frequency components may be related to changes in macroscopic meteorological conditions, while higher-frequency components are more likely to reflect localized and dramatic changes in turbulence and have a more direct impact on wind turbines.

[0033] Frequency components above a preset frequency threshold are analyzed for their amplitude spectrum, which reflects the magnitude of each frequency component. Each frequency component is then assigned a weight based on its importance. This weight can be determined empirically or through statistical analysis of extensive historical data. The amplitude of each frequency component is multiplied by its corresponding weight and summed to produce the turbulent velocity fluctuation amplitude. This turbulent velocity fluctuation amplitude reflects the degree of turbulent velocity fluctuation within the high-frequency range.

[0034] Step S1223: Calculate the deflection angle of the direction angle of the velocity vector relative to the average wind direction according to the components of the turbulence velocity signal in the three-dimensional space to obtain the direction deflection angle.

[0035] Turbulent velocity is a three-dimensional vector with three components in space. The average wind direction can be calculated by statistically analyzing wind direction data over a period of time. For each turbulent velocity vector at a given moment, the velocity vector's azimuth is calculated based on its components in three-dimensional space. This azimuth is then compared with the average wind direction, and the difference between the two is calculated. This difference represents the deviation angle of the velocity vector's azimuth relative to the average wind direction.

[0036] The deflection angle reflects the degree to which turbulence affects the direction of wind speed. A larger deflection angle indicates that turbulence has significantly altered the wind speed direction, significantly impacting the efficiency of the wind turbine's rotor in capturing wind energy. Because rotor design is based on the incoming wind direction, excessive deflection can prevent the rotor from effectively converting wind energy into mechanical energy.

[0037] Step S1224: Arrange the turbulent velocity pulsation amplitude, directional deflection angle and pulsation frequency parameters of each spatial position in the order of spatial coordinates to generate a row vector containing multiple spatial position features, and stack the row vectors of all spatial positions in time order to form a spatial distribution feature of the spatial dimension.

[0038] After calculating the turbulent velocity fluctuation amplitude, directional deflection angle, and pulsation frequency parameters at each spatial location, these parameters are arranged into a row vector according to the spatial coordinate order. For example, the corresponding parameters for the four spatial locations are arranged in the order of the rotor front, rotor side, nacelle top, and tower bottom. This row vector describes the turbulence characteristics of that spatial location during a specific time period.

[0039] Over time, each turbulence data segment corresponding to a time window generates a row vector. Stacking the row vectors corresponding to all time windows in chronological order creates a spatial distribution feature. This spatial distribution feature is presented as a matrix, with rows representing different time windows and columns representing different spatial locations. Each element in the matrix corresponds to a turbulence characteristic parameter at a specific spatial location at the target moment. This spatial distribution feature allows for intuitive monitoring of the distribution of turbulence at different spatial locations and over time.

[0040] Step S123: performing trend analysis processing in the time dimension on each of the turbulence data segments, extracting the change rate of the turbulence velocity pulsation amplitude, the fluctuation amplitude of the direction deflection angle, and the periodic change characteristics of the pulsation frequency parameters in adjacent time windows, and generating dynamic change characteristics in the time dimension.

[0041] In addition to the analysis of the spatial dimension, it is also necessary to deeply study each turbulence data segment from the time dimension to capture the dynamic changing characteristics of turbulence over time.

[0042] Step S1231: Calculate the rate of change of the turbulent velocity fluctuation amplitude at the same spatial position within adjacent time windows.

[0043] Compare the turbulent velocity amplitudes within two adjacent time windows at the same spatial location. First, calculate the difference between the two amplitudes. Then, divide this difference by the time interval between the two time windows. The result is the rate of change of the turbulent velocity amplitude. This rate of change reflects how quickly the turbulent velocity amplitude changes over a short period of time.

[0044] For example, if the turbulent velocity fluctuation amplitude at a certain spatial location in the previous time window is A, and the amplitude in the next time window is B, and the time interval between the two time windows is T, then the rate of change is equal to (B-A) / T. A larger rate of change means that the turbulent velocity fluctuation is changing rapidly, which may have a significant impact on the stability of the wind turbine because the wind turbine control system may not have time to respond to such rapid changes.

[0045] Step S1232: Calculate the fluctuation amplitude of the direction deflection angle at the same spatial position in adjacent time windows.

[0046] At the same spatial location, compare the wind direction deflection angles within two adjacent time windows. Calculate the absolute difference between the two angles, which represents the fluctuation amplitude of the wind direction deflection angle. This fluctuation amplitude reflects the degree of change in wind direction within adjacent time windows.

[0047] If the deflection angle in the previous time window is α, and the angle in the next time window is β, then the fluctuation amplitude is equal to |β-α|. Large fluctuation amplitudes indicate unstable wind direction, requiring the wind turbine rotor to frequently adjust its direction to adapt to changing wind direction, which increases mechanical losses and control difficulty.

[0048] Step S1233: Perform periodogram analysis on the pulsation frequency parameters in each time window, identify periodic components with significant energy concentration, and calculate the frequency offset and energy change rate of the periodic components in adjacent time windows as the periodic change characteristics of the pulsation frequency parameters.

[0049] For each time window, a periodogram analysis is performed on the pulsation frequency parameter. A periodogram is a tool used to analyze signal periodicity, clearly displaying the periodic components and their energy distribution. By analyzing the periodogram, periodic components with significant energy concentrations are identified, representing relatively stable periodic patterns in turbulence.

[0050] Compare the frequencies and energies of these periodic components within adjacent time windows. Calculate the frequency difference between these periodic components within adjacent time windows, also known as the frequency offset. This frequency offset reflects the temporal drift of the periodic frequency. Also, calculate the energy difference between these periodic components within adjacent time windows and divide it by the initial energy value to obtain the energy change rate, which reflects the temporal variation of the periodic energy.

[0051] Step S1234: Serialize the change rate of the turbulent velocity pulsation amplitude, the fluctuation amplitude of the directional deflection angle, and the periodic change characteristics of the pulsation frequency parameters in chronological order to generate a feature vector containing time series correlation, and smooth the feature vector through the sliding window averaging method to suppress the interference of high-frequency noise and obtain the dynamic change characteristics of the time dimension.

[0052] The calculated turbulent velocity fluctuation amplitude change rate, directional deflection angle fluctuation amplitude, and periodic change characteristics of the pulsation frequency parameter are arranged in chronological order to form a feature vector. This feature vector contains information about the changes of these features over time and reflects the time series relationship between them.

[0053] Since the raw data may contain some high-frequency noise interference, this noise will affect the accurate analysis of the dynamic change characteristics of turbulence. Therefore, the sliding window averaging method is used to smooth the feature vector. The sliding window averaging method is a simple and effective filtering method. It selects a window of fixed length, averages the data within the window, and then slides the window along the time axis to calculate the average value within each window in turn. The window width is adaptively adjusted according to the sampling frequency of the turbulence data and the time scale of turbulence changes. If the sampling frequency is high, the window width can be set relatively narrow; if the turbulence changes relatively slowly, the window width can be appropriately widened. Through smoothing, the interference of high-frequency noise can be effectively suppressed, making the feature vector smoother and more stable, and ultimately obtaining the dynamic change characteristics of the time dimension.

[0054] Step S124: performing weighted fusion processing on the spatial distribution characteristics of the spatial dimension and the dynamic change characteristics of the time dimension based on the preset spatiotemporal correlation weight to obtain an initial turbulence characteristic distribution.

[0055] After obtaining the spatial distribution characteristics of the spatial dimension and the dynamic change characteristics of the temporal dimension, they need to be organically integrated to comprehensively consider the spatial distribution and temporal variation characteristics of turbulence. The preset spatiotemporal correlation weights are carefully determined based on the physical characteristics of turbulence and the degree of its impact on wind turbines.

[0056] Different spatial locations and different time points may have different degrees of influence on a wind turbine. For example, in some cases, the spatial locations near the rotor are more critical to the wind turbine, so a higher weight can be assigned to the spatial distribution characteristics of these areas. In other cases, the wind turbine is more affected by critical time points where turbulence suddenly changes, so a higher weight can be assigned to the dynamic change characteristics of these time points.

[0057] The spatial distribution features and dynamic change features are weighted and fused according to preset spatiotemporal correlation weights. Specifically, each element in the spatial distribution features and dynamic change features is multiplied by the corresponding weight, and then they are spliced ​​and combined. During the splicing process, the dimensions and order of the features must be consistent to ensure that the fused features accurately reflect the spatiotemporal characteristics of turbulence. This weighted fusion method generates an initial turbulence feature distribution, which integrates the spatial distribution and temporal variation information of turbulence.

[0058] Step S125: Call the pre-trained turbulence feature optimization model to perform noise suppression and feature enhancement processing on the initial turbulence feature distribution, and highlight the key turbulence features that have a significant impact on pitch angle adjustment by adaptively adjusting the attention weight of the feature channel to generate the final turbulence feature distribution.

[0059] There may still be some noise and features with little impact on pitch angle adjustment in the initial turbulence feature distribution, and it is necessary to call the pre-trained turbulence feature optimization model to further refine it.

[0060] Step S1251: Input the initial turbulence feature distribution into the feature channel attention module of the turbulence feature optimization model, calculate the global average pooling value and the global maximum pooling value of each feature channel, splice the pooling values ​​and input them into the multi-layer perceptron to generate a channel attention weight vector.

[0061] In the feature channel attention module, the initial turbulence feature distribution is first processed. For each feature channel in the initial turbulence feature distribution, its global average pooling value and global maximum pooling value are calculated. The global average pooling value calculates the average value of all elements in the feature channel, which reflects the overall average characteristics of the feature channel; the global maximum pooling value finds the maximum value within the feature channel, which represents the maximum feature response in the feature channel.

[0062] The global average pooling value and the global maximum pooling value of each feature channel are concatenated to form a new vector that combines the average and maximum feature information of each feature channel. This vector is then input into a multilayer perceptron for processing. A multilayer perceptron is a feedforward neural network consisting of multiple hidden layers. In the multilayer perceptron, the input vector undergoes a series of linear transformations and nonlinear activation functions. The linear transformation performs a weighted summation of the input vector using weighted connections between neurons, while the nonlinear activation function introduces nonlinear factors, enabling the model to learn more complex feature relationships. Finally, the multilayer perceptron outputs a channel attention weight vector. Each element in this channel attention weight vector corresponds to the attention weight of a feature channel. The larger the weight value, the more important the feature channel is to subsequent processing.

[0063] Step S1252: Perform weighted processing on each channel of the initial turbulence characteristic distribution through the channel attention weight vector, enhance the characteristic channels related to the turbulence velocity pulsation amplitude and directional deflection angle that have a direct impact on the pitch angle adjustment, suppress the noise-dominated pulsation frequency parameter-related channels, and obtain a weighted characteristic distribution.

[0064] The generated channel attention weight vectors are used to weight each channel in the initial turbulence feature distribution. Feature channels related to turbulent velocity fluctuation amplitude and directional deflection angle, which have a direct impact on pitch angle adjustment, are enhanced during the weighted processing due to their corresponding larger channel attention weights. This means that these feature channels will receive greater attention in subsequent processing, and their characteristic information will be highlighted.

[0065] However, for noise-dominated channels related to pulsation frequency parameters, their corresponding channel attention weights are relatively small and are therefore suppressed during weighted processing. This approach highlights the feature channels important for pitch angle adjustment and suppresses interference from noisy channels, allowing the model to focus more on key features. After weighted processing, a weighted feature distribution is obtained.

[0066] Step S1253: Input the weighted feature distribution into the spatial attention module, generate a spatial attention weight matrix by calculating the feature response significance map of the spatial position, perform local area enhancement processing on the turbulence intensity distribution characteristics in the spatial dimension, highlight the key turbulence features in the wind wheel swept area, and obtain the spatial attention feature distribution.

[0067] The weighted feature distribution is input into the spatial attention module. This module first calculates a saliency map of the feature responses at each spatial location. By analyzing the feature responses at each spatial location in the weighted feature distribution, regions with strong feature responses are identified. These regions are key areas that have a significant impact on the wind turbine. For example, the rotor swept area is a critical region for wind turbines to capture wind energy, and the turbulence characteristics in this area are particularly important for wind turbine performance.

[0068] Based on the feature response saliency map, a spatial attention weight matrix is ​​generated. Each element in the matrix corresponds to the attention weight of a spatial position. The larger the weight value, the more important the feature at that spatial position is.

[0069] The spatial attention weight matrix is ​​used to enhance the localized turbulence intensity distribution characteristics in the spatial dimension. Key turbulence features within the rotor's swept area are enhanced due to their corresponding larger spatial attention weights. Specifically, the feature values ​​of this area are multiplied by a larger weight, amplifying their values ​​and highlighting the key turbulence features in that area. Meanwhile, the features of other less important areas are appropriately suppressed by multiplying them with smaller weights, reducing their impact.

[0070] After the above local area enhancement processing, a spatial attention feature distribution is obtained. This spatial attention feature distribution focuses more on the turbulence characteristics in key areas such as the rotor swept area.

[0071] Step S1254: Input the spatial attention feature distribution into the temporal attention module, generate a temporal attention weight sequence by calculating the feature dependency on the time series, focus on the key time points in the dynamically changing features, suppress the interference of irrelevant time points, and obtain the temporal attention feature distribution.

[0072] The spatial attention feature distribution is fed into the temporal attention module. The core task of this module is to calculate feature dependencies over the time series. Because the dynamics of turbulence are a continuous process over time, there are certain correlations between features at different time points. By analyzing the temporal evolution of the spatial attention feature distribution and utilizing sequence analysis methods such as autocorrelation analysis, we can identify the dependencies between features and determine which time points are most important for pitch angle adjustment.

[0073] For example, at a time point when turbulence suddenly changes dramatically, there may be a strong dependency between the features before and after this time point, and the features at this time point are crucial for timely adjusting the pitch angle to adapt to the turbulence change.

[0074] Based on the calculated feature dependencies, a temporal attention weight sequence is generated. Each element in this sequence corresponds to the attention weight of a time point. The larger the weight, the more important the feature at that time point. When generating the weight sequence, factors such as feature importance and the strength of dependencies between features can be considered.

[0075] A temporal attention weight sequence is used to focus on key time points in dynamically changing features. Features at key time points, due to their corresponding larger temporal attention weights, are retained and enhanced in subsequent processing. Specifically, this is done by multiplying the feature values ​​of these key time points by larger weights to highlight their characteristic information. Features at irrelevant time points are suppressed by multiplying them by smaller weights, reducing their interference with the overall analysis.

[0076] After the above processing, a temporal attention feature distribution is obtained. This temporal attention feature distribution pays more attention to the key time points in the dynamic change characteristics, allowing the model to better capture the dynamic change laws of turbulence.

[0077] Step S1255: Add the temporal attention feature distribution element by element to generate a final turbulence feature distribution, where each dimension of the final turbulence feature distribution corresponds to a key turbulence feature that has been screened and enhanced.

[0078] After obtaining the temporal attention feature distribution, it is then subjected to element-by-element addition. Element-by-element addition refers to adding each element in the temporal attention feature distribution at the same position. This addition operation integrates the filtered and enhanced feature information from different aspects.

[0079] In the preceding processing, the feature channel attention module highlights key feature channels, the spatial attention module enhances features in key spatial regions, and the temporal attention module focuses on features at key time points. After these processes, the temporal attention feature distribution already contains the filtered and enhanced key turbulence features. These key features are further fused through element-by-element addition to generate the final turbulence feature distribution.

[0080] Each dimension of the final turbulence feature distribution corresponds to a key turbulence feature that has been filtered and enhanced. These features are derived from the original real-time turbulence data set through layer-by-layer screening and enhancement, and facilitate wind turbine pitch angle adjustment. For example, some dimensions may correspond to characteristics such as the amplitude and directional deflection angle of turbulent velocity fluctuations in the rotor swept area at key time points.

[0081] Step S130: generating a dynamic compensation strategy according to the turbulence characteristic distribution, wherein the dynamic compensation strategy includes a set of compensation parameters associated with the spatial distribution characteristics and the dynamic change characteristics.

[0082] After obtaining the final turbulence characteristic distribution, it is necessary to generate a dynamic compensation strategy based on the turbulence characteristic distribution to deal with the impact of turbulence on the pitch angle adjustment of the wind turbine.

[0083] Step S131: establishing a mapping relationship model between the turbulence characteristic distribution and the pitch angle adjustment compensation parameters, wherein the mapping relationship model is obtained by training with historical operation data and is used to output a corresponding compensation parameter set according to the input turbulence characteristic distribution.

[0084] In actual wind turbine operation, to accurately generate pitch angle adjustment compensation parameters based on turbulence characteristic distribution, a mapping model between the two is necessary. This model relies on a large amount of historical operating data, which records the wind turbine's operating status and corresponding pitch angle adjustment under different turbulence conditions.

[0085] Step S1311: collecting a set of turbulence data collected during the historical operation of the wind turbine, corresponding pitch angle adjustment data and operating status parameters, and constructing a training data set including input features and output labels.

[0086] First, it is necessary to comprehensively collect relevant data from the historical operation of wind turbines. The turbulence data set is collected by sensors distributed at different locations around the wind turbine. These sensors include anemometers, wind vanes, accelerometers, etc., which can measure physical quantities related to turbulence, such as wind speed, wind direction, and vibration. The pitch angle adjustment data records the actual adjustment of the wind turbine pitch angle under different turbulence conditions, including information such as the adjustment angle and adjustment time. The operating status parameters include wind speed, generator speed, current pitch angle position, etc. The above parameters reflect the real-time status of the wind turbine during operation.

[0087] The collected turbulence data set is subjected to feature extraction to obtain a turbulence feature distribution, which serves as the input features of the training dataset. The corresponding pitch angle adjustment compensation parameters, such as pitch angle pre-compensation, compensation action delay time, and compensation force adjustment coefficient, serve as output labels. By mapping the input features to the output labels, a training dataset is constructed. This dataset records the operation and adjustment of the wind turbine under various turbulence conditions in the past, providing sample material for model training.

[0088] Step S1312: Divide the training data set into a training subset and a validation subset, perform standardization conversion processing on the input features, and standardize the pitch angle pre-compensation amount, compensation action delay time, and compensation force adjustment coefficient in the output label.

[0089] To evaluate model performance and prevent overfitting, the training dataset needs to be divided into a training subset and a validation subset. This division is usually done in a certain ratio, for example, most of the data is used as the training subset for model training, and a small amount of data is used as the validation subset to evaluate the model's performance on unseen data.

[0090] Normalize the input features. Because different turbulence features may have different dimensions and ranges, normalization can bring these features to the same scale, facilitating model training. A common normalization method is z-score normalization, which calculates the mean and standard deviation of each feature and transforms the feature values ​​into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0091] Similarly, the pitch angle pre-compensation amount, compensation action delay time, and compensation force adjustment coefficient in the output label are also standardized. This ensures that the model has the same sensitivity to different output parameters during training, improving the model training effect.

[0092] Step S1313: Construct a multilayer perceptron network including an input layer, multiple hidden layers and an output layer as a mapping relationship model, wherein the number of neurons in the input layer is consistent with the dimension of the turbulence characteristic distribution, and the number of neurons in the output layer is consistent with the number of compensation parameters.

[0093] A multilayer perceptron network is constructed as a mapping model. This network consists of an input layer, multiple hidden layers, and an output layer. The number of neurons in the input layer matches the dimensionality of the turbulence feature distribution, ensuring that the input layer accurately receives the input turbulence feature information. Each neuron corresponds to a turbulence feature, and the input layer's function is to transmit the input turbulence feature to the hidden layers.

[0094] The hidden layer is the core of the multilayer perceptron network and consists of multiple neurons. Neurons in the hidden layer are connected via weights, the size of which determines the strength of information transmission between neurons. The hidden layer performs nonlinear transformations and feature extraction on the input turbulence characteristics, exploring the complex relationship between input features and output compensation parameters. The number of hidden layers and the number of neurons in each hidden layer can be adjusted based on the specific problem and dataset characteristics.

[0095] The output layer has the same number of neurons as the compensation parameters, including three neurons: pitch angle pre-compensation, compensation action delay time, and compensation force adjustment coefficient. The output layer generates the corresponding compensation parameter set based on the output of the hidden layer.

[0096] Step S1314: During the training process of inputting the training subset into the mapping relationship model, the mean square error loss function is used to calculate the difference between the compensation parameters predicted by the mapping relationship model and the actual compensation parameters, and the network weight parameters are updated through the back propagation algorithm.

[0097] The training subset is input into the mapping model for training. During training, the model output is first calculated through forward propagation. The input turbulence characteristics are processed from the input layer through the hidden layer and finally reach the output layer, generating the compensation parameters predicted by the model.

[0098] Next, the mean squared error (MSE) loss function is used to calculate the difference between the model's predicted compensation parameters and the actual compensation parameters. The MSE loss function sums and averages the squared differences between the predicted and actual values, measuring the accuracy of the model's predictions. A smaller loss value indicates that the model's predictions are closer to the actual values.

[0099] Based on the loss value, the network's weight parameters are updated using the backpropagation algorithm. Backpropagation is an optimization algorithm based on gradient descent. Starting from the output layer, it calculates the gradient of the error with respect to the weight of each neuron layer by layer, and then updates the weights based on the gradient. By continuously updating the weights, the model's predictions gradually approach the actual values, and the loss value gradually decreases.

[0100] Step S1315: Evaluate the prediction accuracy of the mapping relationship model on the verification subset. When the prediction error of the pitch angle pre-compensation amount is less than the preset threshold, the prediction deviation of the compensation action delay time is within the allowable range, and the prediction accuracy of the compensation force adjustment coefficient reaches the set standard, stop the training of the mapping relationship model to obtain the final mapping relationship model.

[0101] During the training process, the prediction accuracy of the mapping relationship model is regularly evaluated using the validation subset. The errors and deviations between the pitch angle pre-compensation, compensation action delay time, and compensation force adjustment coefficient predicted by the model on the validation subset and the actual values ​​are calculated.

[0102] When the predicted error of the pitch angle pre-compensation is less than a preset threshold, the predicted deviation of the compensation action delay time is within the allowable range, and the predicted accuracy of the compensation force adjustment coefficient meets the set standard, the model's performance is satisfactory. At this point, model training is terminated, and the final mapping model is obtained. This mapping model accurately outputs the corresponding compensation parameter set based on the input turbulence characteristic distribution.

[0103] Step S132: extracting the mean value of the turbulence velocity pulsation amplitude, the directional deflection angle variance in the spatial distribution characteristics, and the peak value of the change rate and the extreme value of the fluctuation amplitude in the dynamic change characteristics from the turbulence characteristic distribution as input features of the mapping relationship model.

[0104] Key features are extracted from the final turbulence characteristic distribution as input to the mapping relationship model. Within the spatial distribution characteristics, the mean turbulence velocity fluctuation amplitude and the variance of the directional deflection angle are extracted. The mean turbulence velocity fluctuation amplitude is calculated by averaging the turbulence velocity fluctuation amplitude at each spatial location within the spatial distribution characteristics, reflecting the average degree of turbulence velocity fluctuation across the entire spatial range. The variance of the directional deflection angle is an indicator of the spatial fluctuation of wind direction. It is obtained by calculating the variance of the directional deflection angle at each spatial location. A larger variance indicates a more severe fluctuation in wind direction.

[0105] From the dynamic change characteristics, we extract the peak value of the rate of change and the extreme value of the fluctuation amplitude. The peak value of the rate of change refers to the maximum rate of change of the turbulent velocity fluctuation amplitude over time, which reflects the rapidity of the turbulent velocity fluctuation. The extreme value of the fluctuation amplitude indicates the maximum fluctuation amplitude of the direction deflection angle over time, reflecting the maximum instability of the wind direction.

[0106] These extracted features are combined to form a feature vector, which serves as the input feature of the mapping model. These features can well reflect the characteristics of turbulence and the degree of its impact on wind turbines. By inputting them into the mapping model, more accurate compensation parameter prediction results can be obtained.

[0107] Step S133: performing feature transformation processing on the input features through the multi-layer perceptron network of the mapping relationship model to generate an initial compensation parameter set including a pitch angle pre-compensation amount, a compensation action delay time, and a compensation force adjustment coefficient.

[0108] The extracted input features are fed into the multilayer perceptron network of the mapping relationship model. In the input layer, the input features are passed to the neurons in the hidden layer. The neurons in the hidden layer perform a weighted summation of the input features. Each neuron has a set of weights that determine the degree to which the input features influence the neuron's output. After the weighted summation, the neuron processes the result through a nonlinear activation function. This introduces nonlinear factors, enabling the model to learn more complex feature relationships. Common nonlinear activation functions include the ReLU function, which sets values ​​less than 0 to 0 and leaves values ​​greater than 0 unchanged.

[0109] After processing through multiple hidden layers, features are continuously transformed and extracted, and the network gradually learns the mapping relationship between input features and compensation parameters. Finally, based on the output of the hidden layers, neurons in the output layer generate an initial set of compensation parameters, including the pitch angle pre-compensation, compensation action delay time, and compensation force adjustment coefficient. The pitch angle pre-compensation is used to adjust the pitch angle in advance to offset the effects of changes in turbulence intensity; the compensation action delay time is used to adjust the output timing of the control signal to match the temporal characteristics of turbulence intensity changes; and the compensation force adjustment coefficient is used to adjust the response speed and regulation accuracy of the control process.

[0110] Step S134: dynamically correcting the initial compensation parameter set according to the current operating state parameters of the wind turbine to obtain a corrected compensation parameter set that matches the current operating state. The operating state parameters include wind speed, generator speed, and current position of the pitch angle.

[0111] After obtaining the initial compensation parameter set, since the operating state of the wind turbine is changing in real time, the initial compensation parameters need to be dynamically corrected according to the current operating state parameters to ensure that the compensation parameters can better adapt to the actual operating conditions of the wind turbine.

[0112] Step S1341: obtaining the current wind speed measurement value, generator speed sensor output value and pitch angle position feedback signal of the wind turbine generator set as current operating state parameters.

[0113] Current operating parameters are obtained through various sensors installed on the wind turbine. Wind speed sensors measure wind speed in real time, generator speed sensors output generator rotational speed, and pitch angle sensors provide feedback on the current pitch angle. These sensor data are updated in real time and accurately reflect the current operating status of the wind turbine.

[0114] Step S1342: establishing an association rule base between the operating state parameters and compensation parameter correction coefficients, wherein the association rule base stores correction coefficient matrices corresponding to different wind speed intervals, generator speed ranges, and pitch angle position intervals.

[0115] Based on extensive historical data and expert experience, a rule library was established to associate operating state parameters with compensation parameter correction coefficients. This rule library stores correction coefficient matrices corresponding to different wind speed ranges, generator speed ranges, and pitch angle position ranges. For example, when the wind speed is within a certain low range, the generator speed is within a certain range, and the pitch angle is at a specific position, a corresponding correction coefficient matrix can be used to adjust the compensation parameters.

[0116] Step S1343: determining the corresponding wind speed interval, generator speed range and pitch angle position interval according to the current operating state parameters, and extracting the corresponding correction coefficient matrix from the association rule library.

[0117] Based on the acquired current operating state parameters, the wind speed interval, generator speed range, and pitch angle position interval are determined. The correction coefficient matrix corresponding to these intervals and ranges is then retrieved from the association rule library. This correction coefficient matrix contains compensation parameter correction information tailored to the current operating state.

[0118] Step S1344: using the correction coefficient matrix to perform linear transformation processing on the pitch angle pre-compensation amount, compensation action delay time and compensation force adjustment coefficient in the initial compensation parameter set to obtain linearly transformed compensation parameters.

[0119] The extracted correction coefficient matrix is ​​applied to the pitch angle pre-compensation, compensation action delay time, and compensation force adjustment coefficient in the initial compensation parameter set to perform a linear transformation. Linear transformation multiplies the initial compensation parameters by the corresponding coefficients in the correction coefficient matrix to obtain the linearly transformed compensation parameters.

[0120] For example, the pitch angle pre-compensation value is multiplied by the corresponding coefficient in the correction coefficient matrix to obtain the corrected pitch angle pre-compensation value. Similarly, the compensation action delay time and the compensation force adjustment coefficient are subjected to the same linear transformation. This linear transformation takes into account the impact of the current operating state on the compensation parameters, ensuring that the compensation parameters are more consistent with the actual operating conditions of the wind turbine.

[0121] Step S1345: Correcting the compensation parameters after the linear transformation through fuzzy logic reasoning to generate a final corrected compensation parameter set.

[0122] Since the relationship between the operating status of a wind turbine and the compensation parameters may be nonlinear, linear transformation alone may not be enough to fully and accurately correct the compensation parameters. Therefore, fuzzy logic reasoning is needed to further correct the compensation parameters after linear transformation.

[0123] Fuzzy logic reasoning is a fuzzy rule-based reasoning method that can handle uncertain and ambiguous information. In this process, a series of fuzzy rules are established based on empirical knowledge and expert rules relating the operating status of wind turbines to compensation parameters. For example, if the wind speed is within a certain range, the generator speed is within a certain range, and the pitch angle is at a specific position, how should the compensation parameters be further adjusted?

[0124] Based on the current operating state parameters, the fuzzy logic inference engine applies these fuzzy rules to modify the linearly transformed compensation parameters, ultimately generating a final set of modified compensation parameters. This set of modified compensation parameters fully considers the current operating state of the wind turbine and enables better adaptive adjustment of the wind turbine pitch angle.

[0125] Step S135: Perform physical feasibility verification on the modified compensation parameter set so that the pitch angle pre-compensation amount is within the stroke range of the mechanical adjustment mechanism, the compensation action delay time is not less than the system response time, and the compensation force adjustment coefficient meets the equipment safety operation constraint conditions, and finally generate an effective dynamic compensation strategy.

[0126] After obtaining the corrected compensation parameter set, it is necessary to conduct physical feasibility verification to ensure that these parameters are feasible in actual application. First, check whether the pitch angle pre-compensation amount is within the travel range of the mechanical adjustment mechanism. The mechanical adjustment mechanism has its own design limitations, and the adjustment range of the pitch angle is limited. If the pitch angle pre-compensation amount exceeds this range, the mechanical adjustment mechanism will not be able to make the adjustment, which may cause damage to the equipment or malfunction. Therefore, it is necessary to compare the pitch angle pre-compensation amount with the travel range of the mechanical adjustment mechanism. If it exceeds the range, it needs to be adjusted to bring it within the feasible range.

[0127] Next, check whether the compensation action delay is longer than the system response time. System response time refers to the time it takes for a wind turbine to receive a control signal and initiate the corresponding action. If the compensation action delay is shorter than the system response time, the control signal may be issued before the system is ready, resulting in poor control performance. Therefore, ensure that the compensation action delay is long enough to ensure the system has sufficient time to respond to the control signal.

[0128] Finally, check whether the compensation adjustment coefficient meets the equipment's safe operation constraints. This coefficient affects the response speed and adjustment accuracy of the wind turbine's control process. If it is too large, the wind turbine may adjust too drastically, increasing mechanical losses. If it is too small, it may not be able to respond to changes in turbulence in a timely and effective manner. Therefore, it is necessary to check and adjust the compensation adjustment coefficient according to the equipment's safe operation requirements to keep it within a safe range.

[0129] After verifying the physical feasibility of the revised compensation parameter set and adjusting any parameters that did not meet the requirements, an effective dynamic compensation strategy was ultimately generated. This strategy includes a verified and adjusted pitch angle pre-compensation amount, compensation action delay time, and compensation force adjustment coefficient, which can better cope with the impact of turbulence on wind turbines.

[0130] Step S140: adjusting the control parameters of the preset pitch control model based on the dynamic compensation strategy to obtain corrected pitch control parameters.

[0131] After obtaining an effective dynamic compensation strategy, it is necessary to adjust the control parameters of the preset variable pitch control model according to the strategy to improve the adaptability of the wind turbine to turbulence.

[0132] Step S141: parsing the compensation parameter set in the dynamic compensation strategy to extract the pitch angle pre-compensation amount, compensation action delay time, and compensation force adjustment coefficient.

[0133] A detailed analysis of the compensation parameter set within the dynamic compensation strategy is conducted. This set, comprised of multiple parameters, is analyzed to identify three key parameters: pitch angle pre-compensation, compensation action delay time, and compensation intensity adjustment coefficient. These three parameters correspond to different aspects of control adjustment: pitch angle pre-compensation directly adjusts the initial pitch angle position, compensation action delay adjusts the timing of control signal output, and compensation intensity adjustment coefficient adjusts the intensity of the control process.

[0134] Step S142: inputting the pitch angle pre-compensation amount into a preset variable pitch control model as a feedforward control item, wherein the feedforward control item is used to adjust the pitch angle in advance to offset the influence of the change in turbulence intensity.

[0135] The extracted pitch angle pre-compensation is input into the preset variable pitch control model as a feedforward control item. Feedforward control is an active control method. It is different from traditional feedback control. Feedback control makes adjustments only after the system output deviates, while feedforward control makes pre-adjustments before the interference factors affect the system. In the variable pitch control of wind turbines, changes in turbulence intensity are an important interference factor. By inputting the pitch angle pre-compensation into the model as a feedforward control item, the pitch angle can be adjusted in advance before the turbulence intensity changes, thereby offsetting the impact of changes in turbulence intensity on the wind turbine. For example, if it is predicted that the turbulence intensity is about to increase, the pitch angle can be increased in advance through the feedforward control item, so that the wind rotor can better adapt when the turbulence increases, reducing power fluctuations and mechanical stress.

[0136] Step S143: Correcting the time delay parameter in the pitch control model according to the compensation action delay time, so that the output timing of the control signal matches the time characteristic of the turbulence intensity change.

[0137] The compensation action delay time reflects the time relationship between the change in turbulence intensity and the response of the wind turbine. According to the compensation action delay time, the time delay parameter in the preset variable pitch control model is corrected. The time delay parameter in the variable pitch control model determines the output timing of the control signal. By correcting the time delay parameter, the output timing of the control signal is matched with the time characteristics of the change in turbulence intensity. For example, if the turbulence intensity change has a certain periodicity, then by adjusting the time delay parameter, the control signal is output at the appropriate time, which can better cope with the change in turbulence. Specifically, if the turbulence intensity change has a rising phase and a falling phase, by adjusting the time delay parameter, the control signal can be output in time at the early stage of the turbulence intensity rise, and the pitch angle can be adjusted to adapt to the upcoming turbulence change.

[0138] Step S144: dynamically scaling the proportional-integral-derivative control parameters in the pitch control model using the compensation force adjustment coefficient to adjust the response speed and regulation accuracy of the control process.

[0139] Proportional-Integral-Derivative (PID) control is a commonly used control method in variable pitch control models. Its control parameters include the proportional coefficient, the integral coefficient, and the differential coefficient. These parameters determine the response speed and adjustment accuracy of the control process. The proportional coefficient is used to quickly respond to errors, the integral coefficient is used to eliminate steady-state errors, and the differential coefficient is used to predict the changing trend of errors.

[0140] The PID control parameters are dynamically scaled using the compensation force adjustment coefficient. If the compensation force adjustment coefficient is large, it indicates that a faster response speed and higher adjustment accuracy are required. In this case, the value of the PID control parameter can be increased. For example, increasing the proportional coefficient can make the system respond to errors more quickly, increasing the integral coefficient can eliminate steady-state errors more quickly, and increasing the differential coefficient can better predict error changes. If the compensation force adjustment coefficient is small, the value of the PID control parameter can be appropriately reduced to reduce the system's response speed and avoid over-adjustment. Through the above dynamic scaling method, the response speed and adjustment accuracy of the control process are adjusted according to the actual turbulence conditions, so that the wind turbine can better adapt to different turbulence conditions.

[0141] Step S145: The feedforward control item, the corrected time delay parameter and the scaled proportional integral differential control parameter are integrated to generate a corrected variable pitch control parameter, wherein the components of the corrected variable pitch control parameter satisfy the dynamic balance relationship in the aerodynamic control theory.

[0142] The feedforward control term, the corrected time delay parameter, and the scaled PID control parameters are fused. This fusion process ensures that the various components maintain a dynamic equilibrium relationship as defined by aerodynamic control theory. Aerodynamic control theory describes the aerodynamic characteristics and control requirements of wind turbines under different operating conditions. When fusing these parameters, their interactions and influences must be considered so that the corrected pitch control parameters achieve optimal control while ensuring stable operation of the wind turbine.

[0143] For example, the feedforward control term adjusts the pitch angle in advance, the corrected time delay parameter determines the timing of the control signal output, and the scaled PID control parameters determine the intensity and accuracy of the control. These three aspects must work together to ensure the smooth operation of the wind turbine in turbulent environments. These parameters are integrated through a rational fusion method to ultimately generate the corrected variable pitch control parameters. The various components of this variable pitch control parameter set coordinate and work together to achieve precise control of the wind turbine pitch angle.

[0144] During the fusion process, the dimensional consistency of each parameter must be ensured. For example, the pitch angle precompensation of the feedforward control item is measured in angular units, the corrected time delay parameter is measured in time units, and the scaled PID control parameters must match their physical meanings in aerodynamic control theory. Direct addition of units of different dimensions or other unreasonable calculations must not occur. At the same time, the matching of characteristic dimensions must also be considered to ensure that the various parameters can be reasonably combined during fusion to meet the dynamic equilibrium relationship in aerodynamic control theory.

[0145] Step S150: using the modified variable pitch control parameter to control the pitch angle adjustment mechanism of the wind turbine generator set, so as to achieve adaptive adjustment of the pitch angle of the wind turbine generator set.

[0146] After obtaining the corrected pitch control parameters, these parameters are input into the wind turbine's pitch angle adjustment mechanism. The pitch angle adjustment mechanism is a key device for adjusting the pitch angle in a wind turbine, and is mainly composed of mechanical transmission components, a motor, and a controller.

[0147] When the corrected pitch control parameters are input into the pitch angle adjustment mechanism's controller, the controller generates corresponding control signals based on these parameters. For the pitch angle pre-compensation corresponding to the feedforward control item, the controller issues a pre-compensated command, driving the motor and mechanical transmission components to adjust the pitch angle to the pre-compensated angle. This allows the wind rotor to be prepared for changes in turbulence intensity before they occur, reducing the impact of turbulence on the wind turbine.

[0148] The modified time delay parameter affects the timing of control signal output. The controller uses this time delay parameter to precisely time the control signal, aligning pitch angle adjustments with changes in turbulence intensity. For example, if an impending increase in turbulence intensity is detected, the controller will use the time delay parameter to issue a pitch-increasing control signal at the appropriate moment, allowing the rotor to adapt to the turbulence change.

[0149] The scaled PID control parameters determine the specific process of pitch angle adjustment. The proportional coefficient enables the controller to respond quickly to pitch angle errors. If the pitch angle deviates from the target angle, the controller quickly adjusts the motor output based on the proportional coefficient to bring the pitch angle closer to the target value. The integral coefficient eliminates steady-state pitch angle errors. Over time, the integral action continuously adjusts the pitch angle until the error is completely eliminated. The differential coefficient predicts the changing trend of the pitch angle error, allowing for proactive adjustments to avoid over-adjustment of the pitch angle.

[0150] Throughout the adaptive adjustment process, the pitch angle adjustment mechanism continuously monitors the actual pitch angle position and feeds it back to the controller. The controller compares the actual pitch angle with the target pitch angle set by the modified pitch control parameters and continuously adjusts the control signal based on the comparison result, forming a closed-loop control system. This closed-loop control allows the wind turbine's pitch angle to be adaptively adjusted in real time to changes in turbulence, ensuring stable and efficient operation of the wind turbine in varying turbulent environments.

[0151] Figure 2 A schematic diagram illustrating exemplary hardware and software components of an adaptive pitch control system 100 based on dynamic compensation for turbulence intensity, which can implement the concepts of the present application, is provided in some embodiments of the present application. For example, the processor 120 can be used in the adaptive pitch control system 100 based on dynamic compensation for turbulence intensity and perform the functions of the present application.

[0152] The adaptive pitch control system 100 based on dynamic compensation of turbulence intensity can be a general-purpose server or a special-purpose server, both of which can be used to implement the adaptive pitch control method based on dynamic compensation of turbulence intensity of the present application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0153] For example, the adaptive pitch control system 100 based on dynamic compensation of turbulence intensity may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the adaptive pitch control system 100 based on dynamic compensation of turbulence intensity may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The adaptive pitch control system 100 based on dynamic compensation of turbulence intensity also includes an I / O interface 150 between the computer and other input and output devices.

[0154] For ease of explanation, only one processor is described in the adaptive pitch control system 100 based on dynamic compensation of turbulence intensity. However, it should be noted that the adaptive pitch control system 100 based on dynamic compensation of turbulence intensity in the present application may also include multiple processors, so the steps performed by one processor described in the present application may also be performed jointly or individually by multiple processors. For example, if the processor of the adaptive pitch control system 100 based on dynamic compensation of turbulence intensity performs step A and step B, it should be understood that step A and step B may also be performed jointly by two different processors or performed individually in one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor perform steps A and B together.

[0155] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned adaptive variable pitch control method based on dynamic compensation of turbulence intensity is implemented.

[0156] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. An adaptive pitch control method based on dynamic compensation of turbulence intensity, characterized in that: The method comprises: Acquire a real-time turbulence data set in the operating environment of the wind turbine, wherein the real-time turbulence data set includes continuously collected multi-dimensional turbulence feature units with time series labels; Performing feature extraction on the real-time turbulence data set to obtain a turbulence feature distribution including spatial distribution features and dynamic change features; generating a dynamic compensation strategy according to the turbulence characteristic distribution, wherein the dynamic compensation strategy includes a set of compensation parameters associated with the spatial distribution characteristics and the dynamic change characteristics; Adjusting the control parameters of a preset pitch control model based on the dynamic compensation strategy to obtain corrected pitch control parameters; The pitch angle adjustment mechanism of the wind turbine generator set is controlled by utilizing the modified variable pitch control parameter to achieve adaptive adjustment of the pitch angle of the wind turbine generator set.

2. The adaptive pitch control method based on dynamic compensation of turbulence intensity according to claim 1, characterized in that: The feature extraction of the real-time turbulence data set to obtain a turbulence feature distribution including spatial distribution features and dynamic change features includes: Dividing the real-time turbulence data set into a plurality of turbulence data segments with overlapping time windows according to a time series, each of the turbulence data segments comprising turbulence data samples at a plurality of spatial positions; Performing statistical analysis on the turbulence data samples in each turbulence data segment in a spatial dimension, calculating the turbulence velocity pulsation amplitude, direction deflection angle and pulsation frequency parameters at each spatial position, and generating spatial distribution characteristics in a spatial dimension; Performing a time-dimensional trend analysis on each of the turbulence data segments, extracting the change rate of the turbulence velocity pulsation amplitude, the fluctuation amplitude of the direction deflection angle, and the periodic change characteristics of the pulsation frequency parameters within adjacent time windows, and generating dynamic change characteristics in the time dimension; Based on a preset spatiotemporal correlation weight, a weighted fusion process is performed on the spatial distribution characteristics of the spatial dimension and the dynamic change characteristics of the time dimension to obtain an initial turbulence characteristic distribution; Calling a pre-trained turbulence feature optimization model to perform noise suppression and feature enhancement processing on the initial turbulence feature distribution, and highlighting key turbulence features that significantly affect pitch angle adjustment by adaptively adjusting the attention weights of feature channels to generate a final turbulence feature distribution; Among them, the training process of the turbulence feature optimization model includes: using the historical turbulence data set and the corresponding pitch angle adjustment data set as training samples, taking the contribution index of the key features in the turbulence feature distribution and the pitch angle adjustment accuracy index as the joint optimization objectives, and iteratively updating the model parameters through the gradient descent algorithm.

3. The adaptive pitch control method based on dynamic compensation of turbulence intensity according to claim 2, characterized in that: The turbulence data samples in each turbulence data segment are statistically analyzed in the spatial dimension to calculate the turbulence velocity pulsation amplitude, direction deflection angle and pulsation frequency parameters at each spatial position to generate spatial distribution characteristics in the spatial dimension, including: For each turbulence data sample at each spatial position, extract the time domain waveform of the turbulence velocity signal, convert the time domain waveform into a frequency domain signal through fast Fourier transform, and identify the pulsation frequency parameter corresponding to the dominant frequency component; Performing amplitude spectrum analysis on frequency components in the frequency domain signal that are higher than a preset frequency threshold, and calculating a weighted sum of the amplitudes of the frequency components as the turbulent velocity pulsation amplitude; Calculating the deflection angle of the velocity vector relative to the average wind direction based on the components of the turbulent velocity signal in the three-dimensional space to obtain the direction deflection angle; The turbulence velocity pulsation amplitude, directional deflection angle and pulsation frequency parameters of each spatial position are arranged in spatial coordinate order to generate a row vector containing multiple spatial position features. The row vectors of all spatial positions are stacked in time order to form a spatial distribution feature of the spatial dimension. Each element of the spatial distribution feature corresponds to a turbulence characteristic parameter of a spatial position at the target time.

4. The adaptive pitch control method based on dynamic compensation of turbulence intensity according to claim 2, characterized in that: The trend analysis processing of the time dimension is performed on each turbulence data segment, and the periodic change characteristics of the turbulence velocity pulsation amplitude, the fluctuation amplitude of the direction deflection angle, and the pulsation frequency parameters in adjacent time windows are extracted to generate the dynamic change characteristics of the time dimension, including: The ratio of the difference in turbulence velocity fluctuation amplitude at the same spatial position in adjacent time windows to the time interval is calculated to obtain the rate of change of the turbulence velocity fluctuation amplitude. Calculate the absolute difference of the direction deflection angle at the same spatial position in adjacent time windows as the fluctuation amplitude of the direction deflection angle; Perform periodogram analysis on the pulsation frequency parameters in each time window to identify periodic components with significant energy concentration. Calculate the frequency offset and energy change rate of the periodic components in adjacent time windows as the periodic variation characteristics of the pulsation frequency parameters. The rate of change of the turbulence velocity pulsation amplitude, the fluctuation amplitude of the directional deflection angle, and the periodic change characteristics of the pulsation frequency parameters are serialized in chronological order to generate a feature vector containing time series associations. The feature vector is smoothed by a sliding window averaging method to suppress the interference of high-frequency noise and obtain dynamic change characteristics in the time dimension. The width of the sliding window is adaptively adjusted according to the sampling frequency of the turbulence data and the time scale of the turbulence change.

5. The adaptive pitch control method based on dynamic compensation of turbulence intensity according to claim 2, characterized in that: The pre-trained turbulence feature optimization model is called to perform noise suppression and feature enhancement processing on the initial turbulence feature distribution, and the key turbulence features that have a significant impact on pitch angle adjustment are highlighted by adaptively adjusting the attention weight of the feature channel to generate the final turbulence feature distribution, including: Inputting the initial turbulence feature distribution into the feature channel attention module of the turbulence feature optimization model, calculating the global average pooling value and the global maximum pooling value of each feature channel, splicing the pooling values ​​and inputting them into the multi-layer perceptron to generate a channel attention weight vector; weighting the channels of the initial turbulence characteristic distribution using the channel attention weight vector, enhancing the characteristic channels related to the turbulence velocity pulsation amplitude and directional deflection angle that have a direct impact on pitch angle adjustment, and suppressing the channels related to the pulsation frequency parameters dominated by noise, to obtain a weighted characteristic distribution; The weighted feature distribution is input into the spatial attention module, and a spatial attention weight matrix is ​​generated by calculating the feature response saliency map of the spatial position. The turbulence intensity distribution characteristics in the spatial dimension are locally enhanced to highlight the key turbulence characteristics in the wind rotor swept area, thereby obtaining the spatial attention feature distribution; The spatial attention feature distribution is input into the temporal attention module, and the temporal attention weight sequence is generated by calculating the feature dependency on the time series. The key time points in the dynamically changing features are focused on and the interference of irrelevant time points is suppressed to obtain the temporal attention feature distribution; The temporal attention feature distributions are added element by element to generate a final turbulence feature distribution, where each dimension of the final turbulence feature distribution corresponds to a key turbulence feature that has been screened and enhanced.

6. The adaptive pitch control method based on dynamic compensation of turbulence intensity according to claim 1, characterized in that: Generating a dynamic compensation strategy according to the turbulence characteristic distribution includes: Establishing a mapping relationship model between the turbulence characteristic distribution and the pitch angle adjustment compensation parameter, wherein the mapping relationship model is obtained by training historical operation data and is used to output a corresponding compensation parameter set according to the input turbulence characteristic distribution; Extracting the mean value of the turbulence velocity pulsation amplitude, the directional deflection angle variance in the spatial distribution characteristics, and the peak value of the change rate and the extreme value of the fluctuation amplitude in the dynamic change characteristics from the turbulence characteristic distribution as input features of the mapping relationship model; Performing feature transformation processing on the input features through the multi-layer perceptron network of the mapping relationship model to generate an initial compensation parameter set including a pitch angle pre-compensation amount, a compensation action delay time, and a compensation force adjustment coefficient; Dynamically correcting the initial compensation parameter set according to the current operating state parameters of the wind turbine to obtain a corrected compensation parameter set that matches the current operating state, wherein the operating state parameters include wind speed, generator speed, and current position of the pitch angle; The physical feasibility of the modified compensation parameter set is verified so that the pitch angle pre-compensation amount is within the travel range of the mechanical adjustment mechanism, the compensation action delay time is not less than the system response time, and the compensation force adjustment coefficient meets the equipment safety operation constraint conditions, ultimately generating an effective dynamic compensation strategy.

7. The adaptive pitch control method based on dynamic compensation of turbulence intensity according to claim 6, characterized in that: The establishing of the mapping relationship model between the turbulence characteristic distribution and the pitch angle adjustment compensation parameter includes: Collect turbulence data sets collected during the historical operation of wind turbines, corresponding pitch angle adjustment data and operating status parameters, and construct a training dataset containing input features and output labels; Dividing the training data set into a training subset and a validation subset, performing standardization conversion processing on the input features, and standardizing the pitch angle pre-compensation amount, compensation action delay time, and compensation force adjustment coefficient in the output label; Constructing a multilayer perceptron network comprising an input layer, multiple hidden layers, and an output layer as a mapping relationship model, wherein the number of neurons in the input layer is consistent with the dimension of the turbulence characteristic distribution, and the number of neurons in the output layer is consistent with the number of compensation parameters; During the training process of inputting the training subset into the mapping relationship model, a mean square error loss function is used to calculate the difference between the compensation parameters predicted by the mapping relationship model and the actual compensation parameters, and a back propagation algorithm is used to update the network weight parameters; The prediction accuracy of the mapping relationship model is evaluated on the verification subset. When the prediction error of the pitch angle pre-compensation amount is less than a preset threshold, the prediction deviation of the compensation action delay time is within an allowable range, and the prediction accuracy of the compensation force adjustment coefficient reaches a set standard, the training of the mapping relationship model is stopped to obtain the final mapping relationship model.

8. The adaptive pitch control method based on dynamic compensation of turbulence intensity according to claim 6, characterized in that: The dynamically correcting the initial compensation parameter set according to the current operating state parameters of the wind turbine generator set to calculate and obtain a corrected compensation parameter set that matches the current operating state includes: Obtaining a current wind speed measurement value, a generator speed sensor output value, and a pitch angle position feedback signal of the wind turbine generator set as current operating state parameters; Establishing an association rule base between the operating state parameters and compensation parameter correction coefficients, wherein the association rule base stores correction coefficient matrices corresponding to different wind speed intervals, generator speed ranges, and pitch angle position intervals; Determine the corresponding wind speed interval, generator speed range and pitch angle position interval according to the current operating state parameters, and extract the corresponding correction coefficient matrix from the association rule library; Using the correction coefficient matrix, a linear transformation is performed on the pitch angle pre-compensation amount, the compensation action delay time, and the compensation force adjustment coefficient in the initial compensation parameter set to obtain compensation parameters after linear transformation; The compensation parameters after the linear transformation are corrected through fuzzy logic reasoning to generate a final set of corrected compensation parameters.

9. The adaptive pitch control method based on turbulence intensity dynamic compensation according to claim 1, characterized in that: The adjusting the control parameters of the preset pitch control model based on the dynamic compensation strategy to obtain the corrected pitch control parameters includes: Analyzing the compensation parameter set in the dynamic compensation strategy to extract the pitch angle pre-compensation amount, compensation action delay time, and compensation force adjustment coefficient; Inputting the pitch angle pre-compensation amount into a preset variable pitch control model as a feedforward control item, wherein the feedforward control item is used to adjust the pitch angle in advance to offset the influence of the change in turbulence intensity; Correcting a time delay parameter in the pitch control model according to the compensation action delay time so that the output timing of the control signal matches the time characteristic of the turbulence intensity change; Dynamically scaling the proportional-integral-derivative control parameters in the pitch control model using the compensation force adjustment coefficient to adjust the response speed and regulation accuracy of the control process; The feedforward control item, the corrected time delay parameter and the scaled proportional integral differential control parameter are integrated to generate a corrected variable pitch control parameter, wherein each component of the corrected variable pitch control parameter satisfies the dynamic balance relationship in aerodynamic control theory.

10. An adaptive variable pitch control system based on dynamic compensation of turbulence intensity, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the adaptive variable pitch control method based on dynamic compensation of turbulence intensity as described in any one of claims 1 to 9.

Images