Variable speed variable pitch control system for wind turbine based on active disturbance rejection control and method thereof

Abstract

The application relates to the field of intelligent control of wind driven generators, and particularly discloses a variable-speed variable-pitch control system and method of a wind driven generator based on self-disturbance rejection control, which adopts an artificial intelligence control technology, extracts multi-scale neighborhood correlations of dynamic change characteristics of wind speed values, pitch angles and output powers of the wind driven generator at multiple predetermined time points including a current time point through a multi-scale neighborhood feature extraction module, dynamically performs data prediction, and then completes variable-speed variable-pitch control of the wind driven generator, and a Bayesian model is introduced for predictive adjustment in the process, so that the predicted result is more accurate. In this way, the output power and the rotor speed can be kept at their rated values under stable power grid fluctuations, the wind power conversion efficiency is controlled, and the wind turbine is protected from damage.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent control of wind power generators, and more particularly, to a variable speed variable pitch control system of wind power generators based on active disturbance rejection control and a method thereof. BACKGROUND

[0002] With the increasing penetration of wind energy into the power grid, the influence of wind power generation system on the stability of power grid frequency and voltage is also more and more significant. The pitch system is a complex nonlinear system disturbed by multiple uncertainties, and is a key component of the wind power generation system.

[0003] When the wind speed exceeds the rated cut-in wind speed, the pitch angle is changed to control the wind power conversion efficiency, so as to obtain the rated power from the wind, that is, when the wind speed is higher than the rated speed of the variable speed variable pitch wind turbine, the pitch angle is changed to maintain the output power and the rotor speed at its rated value. From the perspective of power grid integration, the power control technology of wind turbine generator system also becomes more and more important.

[0004] However, the existing pitch system can reduce the fluctuation of output power in the control process, but the fluctuation amplitude cannot meet the application requirements under complex working conditions, causing power grid fluctuation when connected to the grid. The reason is that the existing pitch system adjusts the pitch angle based on the wind conditions at the current time point, and the adjustment time is the next time, while the wind conditions at the next time have changed. That is, the traditional pitch system has a weak control lag effect, which causes the waveform generated by the pitch system to exacerbate the power grid fluctuation.

[0005] Therefore, a more optimized variable speed variable pitch control scheme of wind power generator is expected. SUMMARY

[0006] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a variable speed variable pitch control system of wind power generators based on active disturbance rejection control and a method thereof, which adopts artificial intelligence control technology, extracts the multi-scale neighborhood correlation of the dynamic change characteristics of the wind speed value, the pitch angle and the output power of the wind power generator at multiple predetermined time points including the current time point through a multi-scale neighborhood feature extraction module, to dynamically predict the data, and then complete the variable speed variable pitch control of the wind power generator. In this process, a Bayesian model is also introduced for predictive adjustment, so that the predicted result is more accurate. In this way, the output power and the rotor speed can be maintained at its rated value to control the wind power conversion efficiency and protect the wind turbine from damage under the condition of stable power grid fluctuation.

[0007] According to an aspect of the present application, a variable speed variable pitch control system for wind turbine based on active disturbance rejection control is provided, which comprises:

[0008] a data acquisition module configured to acquire wind speed values at a plurality of predetermined time points including a current time point, pitch angles at the plurality of predetermined time points including the current time point, and output powers of the wind turbine at the plurality of predetermined time points including the current time point;

[0009] a wind speed data encoding module configured to arrange the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector, and then pass the wind speed input vector through a first multi-scale neighborhood feature extraction module to obtain a multi-scale wind speed feature vector;

[0010] a pitch angle data encoding module configured to arrange the pitch angles at the plurality of predetermined time points including the current time point into a pitch angle input vector, and then pass the pitch angle input vector through a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector;

[0011] a power data encoding module configured to arrange the output powers of the wind turbine at the plurality of predetermined time points including the current time point into a power input vector, and then pass the power input vector through a third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector;

[0012] a Bayesian fusion module configured to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector using a Bayesian probability model to obtain a posterior probability vector;

[0013] a posterior information correction module configured to correct feature values at each position in the posterior probability vector based on a mean and a variance of a set of feature values at all positions in the posterior probability vector to obtain a corrected posterior probability vector; and

[0014] a control result generation module configured to pass the corrected posterior probability vector through a classifier to obtain a classification result, the classification result being used to indicate whether the pitch angle at the current time point should be increased or decreased.

[0015] In the above variable-speed variable-pitch control system of wind turbine based on active disturbance rejection control, the wind speed data encoding module comprises: a wind speed vectorization unit, configured to arrange the wind speed values at the plurality of predetermined time points including the current time point into the wind speed input vector; a wind speed first scale convolution encoding unit, configured to input the wind speed input vector into a first convolution layer of the first multi-scale neighborhood feature extraction module to obtain a first neighborhood scale wind speed correlation feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; a wind speed second scale convolution encoding unit, configured to input the wind speed input vector into a second convolution layer of the first multi-scale neighborhood feature extraction module to obtain a second neighborhood scale wind speed correlation feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a wind speed multi-scale feature concatenation unit, configured to concatenate the first neighborhood scale wind speed correlation feature vector and the second neighborhood scale wind speed correlation feature vector to obtain the multi-scale wind speed feature vector.

[0016] In the above variable-speed variable-pitch control system of wind turbine based on active disturbance rejection control, the pitch angle data encoding module comprises: a pitch angle vectorization unit, configured to arrange the pitch angles at the plurality of predetermined time points including the current time point into the pitch angle input vector; a pitch angle first scale convolution encoding unit, configured to input the pitch angle input vector into a first convolution layer of the second multi-scale neighborhood feature extraction module to obtain a first neighborhood scale pitch angle correlation feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; a pitch angle second scale convolution encoding unit, configured to input the pitch angle input vector into a second convolution layer of the second multi-scale neighborhood feature extraction module to obtain a second neighborhood scale pitch angle correlation feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a pitch angle multi-scale feature concatenation unit, configured to concatenate the first neighborhood scale pitch angle correlation feature vector and the second neighborhood scale pitch angle correlation feature vector to obtain the multi-scale pitch angle feature vector.

[0017] In the variable-speed variable-pitch control system of the wind power generator based on the active disturbance rejection control, the power data coding module comprises: a power vectorization unit, configured to arrange the output power of the wind power generator at a plurality of predetermined time points including a current time point into the power input vector; a power first scale convolution coding unit, configured to input the power input vector into a first convolution layer of the third multi-scale neighborhood feature extraction module to obtain a first neighborhood scale power correlation feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; a power second scale convolution coding unit, configured to input the power input vector into a second convolution layer of the third multi-scale neighborhood feature extraction module to obtain a second neighborhood scale power correlation feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a power multi-scale feature concatenation unit, configured to concatenate the first neighborhood scale power correlation feature vector and the second neighborhood scale power correlation feature vector to obtain the multi-scale power feature vector.

[0018] In the variable-speed variable-pitch control system of the wind power generator based on the active disturbance rejection control, the Bayesian fusion module is further configured to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector to obtain the posterior probability vector using a Bayesian probability model according to the following formula:

[0019] The formula is as follows:

[0020] qi=pi*ai / bi

[0021] wherein pi is a feature value at each position in the multi-scale pitch angle feature vector, ai and bi are feature values at each position in the multi-scale power feature vector and the multi-scale wind speed feature vector respectively, and qi is a feature value at each position in the posterior probability vector.

[0022] In the variable-speed variable-pitch control system of the wind power generator based on the active disturbance rejection control, the posterior information correction module is further configured to correct the feature values at each position in the posterior probability vector according to the following formula to obtain the corrected posterior probability vector based on the mean and variance of the set of feature values at all positions in the posterior probability vector:

[0023] The formula is as follows:

[0024]

[0025] wherein v i represents the feature value at each position in the posterior probability vector, represents the feature value at each position in the corrected posterior probability vector, μ and σ are the mean and variance of the set of feature values vi mean and variance of ∈V, V represents the posterior probability vector, exp(·) represents the exponential operation of the vector, and the exponential operation of the vector raised to the power represents the natural exponential function value with the value of each position of the vector as the power.

[0026] In the wind turbine variable speed variable pitch control system based on active disturbance rejection control described above, the control result generation module is further configured to process the corrected posterior probability vector using the classifier to obtain the classification result according to the following formula: softmax{(W n ,B n ):…:(W1,B1)|X}, wherein W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the corrected posterior probability vector.

[0027] According to another aspect of the present application, a wind turbine variable speed variable pitch control method based on active disturbance rejection control comprises:

[0028] obtaining wind speed values at a plurality of predetermined time points including a current time point, pitch angles at the plurality of predetermined time points including the current time point, and output powers of the wind turbine at the plurality of predetermined time points including the current time point;

[0029] arranging the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector, and then using a first multi-scale neighborhood feature extraction module to obtain a multi-scale wind speed feature vector;

[0030] arranging the pitch angles at the plurality of predetermined time points including the current time point into a pitch angle input vector, and then using a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector;

[0031] arranging the output powers of the wind turbine at the plurality of predetermined time points including the current time point into a power input vector, and then using a third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector;

[0032] using a Bayesian probability model to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector to obtain a posterior probability vector;

[0033] correcting feature values in each position of the posterior probability vector based on the mean and variance of the feature value set in all positions of the posterior probability vector to obtain a corrected posterior probability vector; and

[0034] passing the corrected posterior probability vector through a classifier to obtain a classification result, the classification result being used to indicate that the pitch angle at the current time point should be increased or should be decreased.

[0035] In the wind turbine variable speed variable pitch control method based on active disturbance rejection control described above, the step of arranging the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector and then passing the wind speed input vector through a first multi-scale neighborhood feature extraction module to obtain a multi-scale wind speed feature vector comprises: arranging the wind speed values at the plurality of predetermined time points including the current time point into the wind speed input vector; inputting the wind speed input vector into a first convolutional layer of the first multi-scale neighborhood feature extraction module to obtain a first neighborhood scale wind speed correlation feature vector, wherein the first convolutional layer has a first one-dimensional convolutional kernel of a first length; inputting the wind speed input vector into a second convolutional layer of the first multi-scale neighborhood feature extraction module to obtain a second neighborhood scale wind speed correlation feature vector, wherein the second convolutional layer has a second one-dimensional convolutional kernel of a second length, the first length being different from the second length; and concatenating the first neighborhood scale wind speed correlation feature vector and the second neighborhood scale wind speed correlation feature vector to obtain the multi-scale wind speed feature vector.

[0036] In the wind turbine variable speed variable pitch control method based on active disturbance rejection control described above, the step of arranging the pitch angle values at the plurality of predetermined time points including the current time point into a pitch angle input vector and then passing the pitch angle input vector through a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector comprises: arranging the pitch angle values at the plurality of predetermined time points including the current time point into the pitch angle input vector; inputting the pitch angle input vector into a first convolutional layer of the second multi-scale neighborhood feature extraction module to obtain a first neighborhood scale pitch angle correlation feature vector, wherein the first convolutional layer has a first one-dimensional convolutional kernel of a first length; inputting the pitch angle input vector into a second convolutional layer of the second multi-scale neighborhood feature extraction module to obtain a second neighborhood scale pitch angle correlation feature vector, wherein the second convolutional layer has a second one-dimensional convolutional kernel of a second length, the first length being different from the second length; and concatenating the first neighborhood scale pitch angle correlation feature vector and the second neighborhood scale pitch angle correlation feature vector to obtain the multi-scale pitch angle feature vector.

[0037] In the variable speed variable pitch control method of the wind power generator based on active disturbance rejection control, after the output power of the wind power generator at the plurality of predetermined time points including the current time point is arranged as a power input vector to obtain a multi-scale power feature vector by a third multi-scale neighborhood feature extraction module, the method comprises: arranging the output power of the wind power generator at the plurality of predetermined time points including the current time point as the power input vector; inputting the power input vector into a first convolution layer of the third multi-scale neighborhood feature extraction module to obtain a first neighborhood scale power correlation feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; inputting the power input vector into a second convolution layer of the third multi-scale neighborhood feature extraction module to obtain a second neighborhood scale power correlation feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and concatenating the first neighborhood scale power correlation feature vector and the second neighborhood scale power correlation feature vector to obtain the multi-scale power feature vector.

[0038] In the variable speed variable pitch control method of the wind power generator based on active disturbance rejection control, the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector are fused using a Bayesian probability model to obtain a posterior probability vector, which comprises: fusing the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector using a Bayesian probability model according to the following formula to obtain the posterior probability vector.

[0039] The formula is:

[0040] qi=pi*ai / bi

[0041] Wherein, pi is the feature value of each position in the multi-scale pitch angle feature vector, ai and bi are respectively the feature values of each position in the multi-scale power feature vector and the multi-scale wind speed feature vector, and qi is the feature value of each position in the posterior probability vector.

[0042] In the variable speed variable pitch control method of the wind power generator based on active disturbance rejection control, based on the mean and variance of the set of feature values of all positions in the posterior probability vector, the feature values of each position in the posterior probability vector are corrected to obtain a corrected posterior probability vector, which comprises: based on the mean and variance of the set of feature values of all positions in the posterior probability vector, the feature values of each position in the posterior probability vector are corrected according to the following formula to obtain the corrected posterior probability vector.

[0043] The formula is:

[0044]

[0045] wherein v i represents eigenvalues of each position in the posterior probability vector, represents eigenvalues of each position in the corrected posterior probability vector, μ and σ are mean and variance of the feature set v i ∈ V, V represents the posterior probability vector, exp(·) represents exponential operation of a vector, exponential operation with a vector as power represents natural exponential function value with value of each position of the vector as power.

[0046] In the wind turbine variable speed variable pitch control method based on active disturbance rejection control described above, the corrected posterior probability vector is processed by a classifier to obtain a classification result, and the classification result is used to indicate that the pitch angle at the current time point should be increased or should be decreased, comprising: using the classifier to process the corrected posterior probability vector in the following formula to obtain the classification result, wherein the formula is: softmax{(W n ,B n ):…:(W1,B1)|X}, wherein W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the corrected posterior probability vector.

[0047] Compared with the prior art, the wind turbine variable speed variable pitch control system and method based on active disturbance rejection control provided by the present application adopt artificial intelligence control technology, extract multi-scale neighborhood correlation of dynamic change characteristics of wind speed values, pitch angles and output powers of wind turbines at multiple predetermined time points including the current time point through a multi-scale neighborhood feature extraction module, dynamically perform data prediction, and complete variable speed variable pitch control of the wind turbine, and a Bayesian model is introduced for predictive adjustment in the process, so that the predicted result is more accurate. In this way, the output power and rotor speed can be kept at their rated values in the case of stable grid fluctuations, so as to control the wind power conversion efficiency and protect the wind turbine from damage. BRIEF DESCRIPTION OF DRAWINGS

[0048] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application, taken in conjunction with the accompanying drawings. The drawings provided in the specification and the embodiments of the present application together serve to provide a further understanding that enables others skilled in the art to make or use the present application. The drawings provided are for illustrative purposes and are not intended to limit the present application. In the drawings, the same reference numerals generally refer to the same components or steps throughout the drawings.

[0049] Figure 1 An application scenario diagram of the wind turbine variable speed variable pitch control system based on active disturbance rejection control according to the embodiments of the present application.

[0050] Figure 2 A block diagram of a variable speed variable pitch control system for a wind turbine based on active disturbance rejection control according to an embodiment of the present application.

[0051] Figure 3 A block diagram of a wind speed data encoding module in a variable speed variable pitch control system for a wind turbine based on active disturbance rejection control according to an embodiment of the present application.

[0052] Figure 4 A flow chart of a variable speed variable pitch control method for a wind turbine based on active disturbance rejection control according to an embodiment of the present application.

[0053] Figure 5 An architectural schematic diagram of a variable speed variable pitch control method for a wind turbine based on active disturbance rejection control according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part but not all of the embodiments of the present application, and the present application can be implemented in many different forms. Therefore, the attached drawings should not be used to limit and define the present application, and the present application should cover all changes falling within the scope of the appended claims and their equivalents.

[0055] Scenario overview

[0056] As mentioned above, with the increasing penetration of wind energy into the power grid, the influence of wind power systems on the stability of the grid frequency and voltage is also becoming more and more significant. The pitch system is a complex nonlinear system disturbed by multiple uncertainties, and is a key component of wind power systems.

[0057] When the wind speed exceeds the rated cut-in wind speed, the pitch angle is changed to control the wind power conversion efficiency, so as to obtain the rated power from the wind, and to protect the wind turbine from damage, that is, when the wind speed is higher than the rated speed of the variable speed variable pitch wind turbine, the pitch angle is changed to maintain the output power and the rotor speed at their rated values. From the perspective of grid integration, the power control technology of wind turbine generators also becomes more and more important.

[0058] However, the existing pitch system in the control process can reduce the fluctuation of the output power, but the fluctuation amplitude cannot meet the application requirements under complex working conditions, causing grid fluctuation when connected to the grid. The reason is that the existing pitch system adjusts the pitch angle based on the wind conditions at the current time point, and the adjustment time is already the next time, while the wind conditions at the next time have changed. That is, the traditional pitch system has a weak control lag effect, which causes the waveform generated by it to exacerbate the grid fluctuation. Therefore, a more optimized variable speed variable pitch control scheme for wind turbines is expected.

[0059] Accordingly, the present application considers that when the pitch system of the wind turbine is regulated to control the wind power conversion efficiency and protect the wind turbine from damage, the existing pitch system is adjusted based on the wind condition at the current time point, so that the adjusted pitch angle cannot keep the output power and the rotor speed at the rated value, thereby exacerbating the grid fluctuation. Therefore, in the technical solution of the present application, it is expected to dynamically predict the data based on the wind speed values at multiple predetermined time points including the current time point, the dynamic change characteristic information of the pitch angle and the output power of the wind turbine, and then complete the variable speed variable pitch control of the wind turbine, and in this process, the Bayesian model is also introduced for predictive adjustment, so that the predicted result is more accurate, so that the output power and the rotor speed can be kept at the rated value to control the wind power conversion efficiency and protect the wind turbine from damage in the case of stable grid fluctuation.

[0060] Specifically, in the technical solution of the present application, first, the wind speed values at multiple predetermined time points including the current time point are collected by a wind speed sensor, and the pitch angles at the multiple predetermined time points including the current time point are collected by an angle sensor, and the output powers of the wind turbine at the multiple predetermined time points including the current time point are collected by a power detector.

[0061] Then, the convolutional neural network with excellent performance in local implicit correlation feature extraction is used for implicit feature extraction of data feature correlation. It should be understood that the convolutional neural network is originally a model applied in the image field, but its local feature extraction idea can also be applied to time series data analysis. For example, a time series convolution structure with a convolution kernel size of 3, for time series data input, the convolution kernel moves in the form of a sliding window along the time dimension, and outputs the weighted sum of the data in each time series segment. Each convolution unit stacks multiple convolution kernels to output multi-dimensional features. Large convolution kernels extract features from large-scale time series neighborhoods, where each value in the neighborhood has a smaller impact, thereby reducing the fluctuations of the input data and reducing the impact of noise points on the output features. However, large-scale convolution kernels reduce the difference in value changes, which can easily lead to over-smoothing problems, causing the output features to lose discriminability. On the contrary, small-scale convolution kernels can better preserve the information in the input data, but are also more susceptible to noise points. Therefore, considering the characteristics of convolution of different scales, different sizes of convolution units are combined to extract features of different time scales. Then, feature splicing is used to complete feature fusion, thereby obtaining multi-scale neighborhood features.

[0062] That is, specifically, in the technical solution of the present application, first, the wind speed values of the plurality of predetermined time points including the current time point are arranged as a wind speed input vector to facilitate subsequent feature extraction. Further, the wind speed input vector is respectively one-dimensional convolutional coded using the convolutional layers of the multi-scale neighborhood feature extraction module with different scales of one-dimensional convolutional kernels, and then the feature vectors corresponding to the two different scales of one-dimensional convolutional kernels are concatenated to obtain the multi-scale wind speed feature vector. In particular, in this way, the multi-scale neighborhood correlation of the dynamic change features of the wind speed values in the time sequence dimension is extracted, so that the output features contain both smoothed features and original input features, avoiding information loss and improving the accuracy of subsequent classification. In other examples of the present application, the multi-scale neighborhood feature extraction module can also include a larger number of one-dimensional convolutional layers using one-dimensional convolutional kernels of different lengths for different scale neighborhood correlation feature extraction, which is not limited by the present application.

[0063] Similarly, for the pitch angle of the plurality of predetermined time points including the current time point and the output power of the wind turbine of the plurality of predetermined time points including the current time point, considering that the pitch angle and the output power of the wind turbine also have dynamic implicit change features in the time dimension, in order to extract the multi-scale neighborhood correlation of such dynamic change features, the pitch angle of the plurality of predetermined time points including the current time point and the output power of the wind turbine of the plurality of predetermined time points including the current time point are respectively input into the multi-scale neighborhood feature extraction module for processing to obtain a multi-scale pitch angle feature vector and a multi-scale power feature vector. In this way, the different scale neighborhood correlation features of the dynamic change features of the pitch angle and the output power of the wind turbine can contain not only smoothed features but also original input feature information, avoiding information loss and improving the accuracy of subsequent classification.

[0064] It should be appreciated that, considering the use of the multi-scale pitch angle feature vector as the prior probability, the purpose of the technical solution of the present application is to update the prior probability to obtain the posterior probability when new evidence, i.e., when there is a new wind speed value change. Then according to the Bayes formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability, therefore, in the technical solution of the present application, the Bayesian probability model is used to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector to obtain a posterior probability vector, wherein the multi-scale pitch angle feature vector is taken as the prior, the multi-scale wind speed feature vector is taken as the event, and the multi-scale power feature vector is taken as the evidence. In this way, the posterior probability vector can be obtained through a classifier to obtain a classification result for representing whether the pitch angle at the current time point should be increased or decreased.

[0065] However, since the Bayesian probability model performs point-by-point calculation on the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector, the obtained posterior probability vector may have a problem of poor expression consistency between feature values, thereby affecting the classification effect.

[0066] Based on this, in the technical solution of the present application, the posterior probability vector is further subjected to length-dimension recursive squeezing-activation optimization, denoted as:

[0067]

[0068] Wherein v i represents the feature value at each position in the posterior probability vector, represents the feature value at each position in the corrected posterior probability vector, μ and σ are the mean and variance of the feature set v i ∈ V, V represents the posterior probability vector, exp(·) represents the exponential operation of the vector, and the exponential operation with a vector as the power represents the natural exponential function value with the value at each position of the vector as the power.

[0069] Here, the length-dimension recursive squeezing-activation optimization activates the length-dimension recursive of the feature distribution based on the statistical characteristics of the feature set, thereby inferring the distribution of each sampling point of the feature in the length dimension, and adopts a predetermined dimension squeezing-activation mechanism composed of ReLU-Sigmoid function to obtain the confidence value of the predetermined dimension attention enhancement, so as to improve the expression consistency of the high-dimensional feature flow pattern of the posterior probability vector in the length dimension, and further improve the classification accuracy.

[0070] Based on this, this application proposes a wind turbine variable speed and pitch control system based on active disturbance rejection control, comprising: a data acquisition module for acquiring wind speed values ​​at multiple predetermined time points, including the current time point, the pitch angle at the multiple predetermined time points, and the output power of the wind turbine at the multiple predetermined time points, all acquired by a wind speed sensor; a wind speed data encoding module for arranging the wind speed values ​​at the multiple predetermined time points, including the current time point, into a wind speed input vector, and then obtaining a multi-scale wind speed feature vector through a first multi-scale neighborhood feature extraction module; and a pitch angle data encoding module for arranging the pitch angles at the multiple predetermined time points, including the current time point, into a pitch angle input vector, and then obtaining a multi-scale pitch angle feature vector through a second multi-scale neighborhood feature extraction module; and power... The data encoding module is used to arrange the output power of the wind turbines at multiple predetermined time points, including the current time point, into a power input vector, and then use a third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector; the Bayesian fusion module is used to use a Bayesian probability model to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector to obtain a posterior probability vector; the posterior information correction module is used to correct the feature values ​​at each position in the posterior probability vector based on the mean and variance of the feature value set at all positions in the posterior probability vector to obtain a corrected posterior probability vector; and the control result generation module is used to pass the corrected posterior probability vector through a classifier to obtain a classification result, the classification result being used to indicate whether the pitch angle at the current time point should increase or decrease.

[0071] Figure 1 The illustration shows an application scenario of a wind turbine variable speed and pitch control system based on active disturbance rejection control according to an embodiment of this application. For example... Figure 1 As shown, in this application scenario, firstly, through a wind speed sensor (e.g., such as...) Figure 1 The W shown in the diagram collects wind speed values ​​at multiple predetermined time points, including the current time point, and is deployed in a variable pitch system (e.g., such as...). Figure 1 The angle sensor in P shown (e.g., such as Figure 1 As shown in A), the pitch angles at multiple predetermined time points, including the current time point, are collected, and measured using a power detector (e.g., such as...). Figure 1 The P shown in the diagram collects data from the wind turbine at multiple predetermined time points, including the current time point (e.g., as shown in the diagram). Figure 1the output power of G) as shown in FIG. 1. Then, the obtained wind speed values, pitch angles and output powers of the wind turbine at the plurality of predetermined time points including the current time point are input into a server (for example, a cloud server S as shown in FIG. 1) in which a wind turbine variable speed variable pitch control algorithm based on active disturbance rejection control is deployed, where the server can process the wind speed values, pitch angles and output powers of the wind turbine at the plurality of predetermined time points including the current time point with the wind turbine variable speed variable pitch control algorithm based on active disturbance rejection control to generate a classification result indicating that the pitch angle at the current time point should be increased or should be decreased. Figure 1

[0072] In this application scenario, the registration information includes but is not limited to "operation state", "wireless IP address", "cluster priority" and "cluster controller" and the like.

[0073] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be specifically introduced below with reference to the accompanying drawings.

[0074] Exemplary system

[0075] Figure 2 FIG. 1 illustrates a block diagram of a wind turbine variable speed variable pitch control system based on active disturbance rejection control according to an embodiment of the present application. As shown in FIG. 1, the wind turbine variable speed variable pitch control system based on active disturbance rejection control includes a wind turbine 1, a wind speed sensor 2, a pitch angle sensor 3, a wind turbine output power sensor 4, a wind turbine variable speed variable pitch control algorithm based on active disturbance rejection control server 5 and a wind turbine variable speed variable pitch control algorithm based on active disturbance rejection control client 6. Figure 2 ​As shown, the variable speed variable pitch control system 200 of the wind turbine based on active disturbance rejection control according to the embodiment of the present application comprises: a data acquisition module 210 configured to acquire wind speed values at a plurality of predetermined time points including a current time point, pitch angles at the plurality of predetermined time points including the current time point, and output powers of the wind turbine at the plurality of predetermined time points including the current time point; a wind speed data coding module 220 configured to arrange the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector, and then pass the wind speed input vector through a first multi-scale neighborhood feature extraction module to obtain a multi-scale wind speed feature vector; a pitch angle data coding module 230 configured to arrange the pitch angles at the plurality of predetermined time points including the current time point into a pitch angle input vector, and then pass the pitch angle input vector through a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector; a power data coding module 240 configured to arrange the output powers of the wind turbine at the plurality of predetermined time points including the current time point into a power input vector, and then pass the power input vector through a third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector; a Bayesian fusion module 250 configured to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector using a Bayesian probability model to obtain a posterior probability vector; a posterior information correction module 260 configured to correct feature values at each position in the posterior probability vector based on a mean and a variance of a feature value set at all positions in the posterior probability vector to obtain a corrected posterior probability vector; and a control result generation module 270 configured to pass the corrected posterior probability vector through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the pitch angle at the current time point should be increased or decreased.

[0076] Specifically, in the embodiment of the present application, the data acquisition module 210 is configured to acquire wind speed values at a plurality of predetermined time points including a current time point, pitch angles at the plurality of predetermined time points including the current time point, and output powers of the wind turbine at the plurality of predetermined time points including the current time point, which are collected by the wind speed sensor. As mentioned above, in consideration of the regulation of the pitch system of the wind turbine to control the wind power conversion efficiency and protect the wind turbine from damage, the existing pitch system is adjusted based on the wind condition at the current time point, so that the adjusted pitch angle cannot keep the output power and the rotor speed at their rated values, and thus the grid fluctuation is aggravated. Therefore, in the technical solution of the present application, it is expected to dynamically perform data prediction based on the dynamic change characteristic information of the wind speed values, the pitch angles, and the output powers of the wind turbine at the plurality of predetermined time points including the current time point, and then complete the variable speed and variable pitch control of the wind turbine, and in this process, the Bayesian model is introduced for predictive adjustment, so that the predicted result is more accurate, and the output power and the rotor speed can be kept at their rated values to control the wind power conversion efficiency and protect the wind turbine from damage in the case of stabilizing the grid fluctuation.

[0077] That is, specifically, in the technical solution of the present application, first, the wind speed values at a plurality of predetermined time points including a current time point are collected by a wind speed sensor, and the pitch angles at the plurality of predetermined time points including the current time point are collected by an angle sensor, and the output powers of the wind turbine at the plurality of predetermined time points including the current time point are collected by a power detector.

[0078] Specifically, in the embodiments of the present application, the wind speed data encoding module 220 is configured to arrange the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector, and then perform one-dimensional convolution coding on the wind speed input vector by using a convolution layer having different one-dimensional convolution kernels of different scales, and then concatenate the feature vectors corresponding to the two different one-dimensional convolution kernels to obtain the multi-scale wind speed feature vector. In particular, in this way, the multi-scale neighborhood correlation of the dynamic change features of the wind speed values in the time sequence dimension is extracted, so that the output features not only contain smoothed features, but also save the features of the original input, avoiding the loss of information, and thus improving the accuracy of subsequent classification. In other examples of the present application, the multi-scale neighborhood feature extraction module can also include a larger number of one-dimensional convolution layers, which use one-dimensional convolution kernels of different lengths to extract correlation features in different scales of neighborhoods, and the present application is not limited thereto.

[0079] That is, specifically, in the technical solution of the present application, first, the wind speed values at the plurality of predetermined time points including the current time point are arranged into a wind speed input vector for subsequent feature extraction. Further, the multi-scale neighborhood feature extraction module is used to perform one-dimensional convolution coding on the wind speed input vector by using a convolution layer having different one-dimensional convolution kernels of different scales, and then concatenate the feature vectors corresponding to the two different one-dimensional convolution kernels to obtain the multi-scale wind speed feature vector. In particular, in this way, the multi-scale neighborhood correlation of the dynamic change features of the wind speed values in the time sequence dimension is extracted, so that the output features not only contain smoothed features, but also save the features of the original input, avoiding the loss of information, and thus improving the accuracy of subsequent classification. In other examples of the present application, the multi-scale neighborhood feature extraction module can also include a larger number of one-dimensional convolution layers, which use one-dimensional convolution kernels of different lengths to extract correlation features in different scales of neighborhoods, and the present application is not limited thereto.

[0080] More specifically, in the embodiments of the present application, the wind speed data encoding module comprises: first, arranging the wind speed values at a plurality of predetermined time points including the current time point into the wind speed input vector; then, inputting the wind speed input vector into the first convolution layer of the first multi-scale neighborhood feature extraction module to obtain the first neighborhood scale wind speed correlation feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; then, inputting the wind speed input vector into the second convolution layer of the first multi-scale neighborhood feature extraction module to obtain the second neighborhood scale wind speed correlation feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; finally, concatenating the first neighborhood scale wind speed correlation feature vector and the second neighborhood scale wind speed correlation feature vector to obtain the multi-scale wind speed feature vector.

[0081] Figure 3 Fig. 1 illustrates a block diagram of a wind speed data encoding module in a variable speed variable pitch control system for wind power generators based on active disturbance rejection control according to an embodiment of the present application. As shown in Fig. 1, the wind speed data encoding module 220 comprises a wind speed vectorization unit 221, a wind speed first scale convolution encoding unit 222, a wind speed second scale convolution encoding unit 223, and a wind speed multi-scale feature concatenation unit 224. Figure 3 The wind speed vectorization unit 221 is configured to arrange the wind speed values at a plurality of predetermined time points including the current time point into the wind speed input vector. The wind speed first scale convolution encoding unit 222 is configured to input the wind speed input vector into the first convolution layer of the first multi-scale neighborhood feature extraction module to obtain the first neighborhood scale wind speed correlation feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length. The wind speed second scale convolution encoding unit 223 is configured to input the wind speed input vector into the second convolution layer of the first multi-scale neighborhood feature extraction module to obtain the second neighborhood scale wind speed correlation feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length. The wind speed multi-scale feature concatenation unit 224 is configured to concatenate the first neighborhood scale wind speed correlation feature vector and the second neighborhood scale wind speed correlation feature vector to obtain the multi-scale wind speed feature vector.

[0082] Specifically, in the embodiments of the present application, the pitch angle data encoding module 230 and the power data encoding module 240 are configured to arrange the pitch angles at the plurality of predetermined time points including the current time point into a pitch angle input vector, and then input the pitch angle input vector into the second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector; and arrange the output powers of the wind turbine at the plurality of predetermined time points including the current time point into a power input vector, and then input the power input vector into the third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector. That is, in the technical solution of the present application, for the pitch angles at the plurality of predetermined time points including the current time point and the output powers of the wind turbine at the plurality of predetermined time points including the current time point, considering that the pitch angles and the output powers of the wind turbine also have dynamic implicit change characteristics in the time dimension, in order to extract the multi-scale neighborhood correlation of the dynamic change characteristics, the pitch angles at the plurality of predetermined time points including the current time point and the output powers of the wind turbine at the plurality of predetermined time points including the current time point are respectively input into the multi-scale neighborhood feature extraction module for processing to obtain a multi-scale pitch angle feature vector and a multi-scale power feature vector. In this way, the neighborhood correlation features of different scales of the dynamic change characteristics of the output pitch angles and the output powers of the wind turbine can not only contain smoothed features, but also save the original input feature information, so as to avoid the loss of information, and thus the accuracy of subsequent classification can be improved.

[0083] More specifically, in the embodiments of the present application, the pitch angle data encoding module comprises: a pitch angle vectorization unit configured to arrange the pitch angles at the plurality of predetermined time points including the current time point into the pitch angle input vector; a pitch angle first scale convolution encoding unit configured to input the pitch angle input vector into a first convolution layer of the second multi-scale neighborhood feature extraction module to obtain a first neighborhood scale pitch angle correlation feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; a pitch angle second scale convolution encoding unit configured to input the pitch angle input vector into a second convolution layer of the second multi-scale neighborhood feature extraction module to obtain a second neighborhood scale pitch angle correlation feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a pitch angle multi-scale feature concatenation unit configured to concatenate the first neighborhood scale pitch angle correlation feature vector and the second neighborhood scale pitch angle correlation feature vector to obtain the multi-scale pitch angle feature vector.

[0084] More specifically, in the embodiments of the present application, the power data encoding module comprises: a power vectorization unit, configured to arrange the output power of the wind power generator at a plurality of predetermined time points including the current time point into the power input vector; a power first scale convolutional encoding unit, configured to input the power input vector into a first convolutional layer of the third multi-scale neighborhood feature extraction module to obtain a first neighborhood scale power correlation feature vector, wherein the first convolutional layer has a first one-dimensional convolutional kernel with a first length; a power second scale convolutional encoding unit, configured to input the power input vector into a second convolutional layer of the third multi-scale neighborhood feature extraction module to obtain a second neighborhood scale power correlation feature vector, wherein the second convolutional layer has a second one-dimensional convolutional kernel with a second length, and the first length is different from the second length; and a power multi-scale feature concatenation unit, configured to concatenate the first neighborhood scale power correlation feature vector and the second neighborhood scale power correlation feature vector to obtain the multi-scale power feature vector.

[0085] Specifically, in the embodiments of the present application, the Bayesian fusion module 250 is configured to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector using a Bayesian probability model to obtain a posterior probability vector. It should be understood that, considering that the multi-scale pitch angle feature vector is used as a prior probability, the purpose of the technical solution of the present application is to update the prior probability to obtain the posterior probability when new evidence, i.e., when there is a new wind speed value change. Then according to the Bayesian formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability. Therefore, in the technical solution of the present application, the Bayesian probability model is used to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector to obtain the posterior probability vector, wherein the multi-scale pitch angle feature vector is used as the prior, the multi-scale wind speed feature vector is used as the event, and the multi-scale power feature vector is used as the evidence.

[0086] More specifically, in the embodiments of the present application, the Bayesian fusion module is further configured to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector using a Bayesian probability model according to the following formula to obtain the posterior probability vector:

[0087] wherein the formula is:

[0088] qi=pi*ai / bi

[0089] wherein pi is the feature value of each position in the multi-scale pitch angle feature vector, ai and bi are the feature values of each position in the multi-scale power feature vector and the multi-scale wind speed feature vector respectively, and qi is the feature value of each position in the posterior probability vector.

[0090] Specifically, in the embodiments of the present application, the posterior information correction module 260 is configured to correct the feature values of each position in the posterior probability vector based on the mean and variance of the set of feature values of all positions in the posterior probability vector to obtain a corrected posterior probability vector. It should be understood that, since the Bayesian probability model is calculated point by point for the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector, the obtained posterior probability vector may have a problem of poor expression consistency between feature values, thereby affecting the classification effect. Therefore, in the technical solution of the present application, the posterior probability vector is further subjected to length-dimension recursive squeeze-activation optimization. Here, the length-dimension recursive squeeze-activation optimization activates the length-dimension recursive of feature distribution based on the statistical characteristics of the feature set, thereby inferring the distribution of each sampling point of the feature in the length dimension, and adopts a predetermined dimension squeeze-activation mechanism composed of ReLU-Sigmoid function to obtain a confidence value of predetermined dimension attention enhancement, so as to improve the expression consistency of the high-dimensional feature flow pattern of the posterior probability vector in the length dimension, and further improve the classification accuracy.

[0091] More specifically, in the embodiments of the present application, the posterior information correction module is further configured to correct the feature values of each position in the posterior probability vector based on the mean and variance of the set of feature values of all positions in the posterior probability vector to obtain the corrected posterior probability vector according to the following formula:

[0092] wherein the formula is:

[0093]

[0094] wherein v i represents the feature value of each position in the posterior probability vector, represents the feature value of each position in the corrected posterior probability vector, μ and σ are the mean and variance of the feature set v i ∈V, V represents the posterior probability vector, exp(·) represents the exponential operation of the vector, and the exponential operation of the vector to the power represents the natural exponential function value with the value of each position of the vector as the power.

[0095] Specifically, in the embodiments of the present application, the control result generation module 270 is configured to pass the corrected posterior probability vector through a classifier to obtain a classification result, which is used to indicate whether the pitch angle at the current time point should be increased or decreased. In one specific example, the classifier is used to process the corrected posterior probability vector in the following formula to obtain the classification result, where the formula is: softmax{W n ,B n ):…:(W1,B1)|X} where W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the corrected posterior probability vector.

[0096] In summary, the wind turbine variable speed variable pitch control system 200 based on active disturbance rejection control according to the embodiments of the present application is illustrated, which adopts artificial intelligence control technology, extracts the multi-scale neighborhood correlation of the dynamic change characteristics of the wind speed values, the pitch angles and the output power of the wind turbine at multiple predetermined time points including the current time point through the multi-scale neighborhood feature extraction module, dynamically performs data prediction, and then completes the variable speed variable pitch control of the wind turbine. In this process, a Bayesian model is also introduced for predictive adjustment, so that the predicted result is more accurate. In this way, the output power and the rotor speed can be kept at their rated values in the case of stable grid fluctuations, so as to control the wind power conversion efficiency and protect the wind turbine from damage.

[0097] As described above, the wind turbine variable speed variable pitch control system 200 based on active disturbance rejection control according to the embodiments of the present application can be implemented in various terminal devices, such as a server of the wind turbine variable speed variable pitch control algorithm based on active disturbance rejection control. In one example, the wind turbine variable speed variable pitch control system 200 based on active disturbance rejection control according to the embodiments of the present application can be integrated into a terminal device as a software module and / or a hardware module. For example, the wind turbine variable speed variable pitch control system 200 based on active disturbance rejection control can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the wind turbine variable speed variable pitch control system 200 based on active disturbance rejection control can also be one of the many hardware modules of the terminal device.

[0098] Alternatively, in another example, the wind turbine variable speed variable pitch control system 200 based on active disturbance rejection control and the terminal device can also be separate devices, and the wind turbine variable speed variable pitch control system 200 based on active disturbance rejection control can be connected to the terminal device through a wired and / or wireless network, and transmit interactive information in a predetermined data format.

[0099] Exemplary method

[0100] Figure 4 The diagram illustrates a flowchart of a wind turbine variable speed and pitch control method based on active disturbance rejection control. (For example...) Figure 4 As shown, the wind turbine variable speed and pitch control method based on active disturbance rejection control according to an embodiment of this application includes the following steps: S110, acquiring wind speed values ​​at multiple predetermined time points including the current time point, pitch angles at the multiple predetermined time points including the current time point, and output power of the wind turbine at the multiple predetermined time points including the current time point, all collected by a wind speed sensor; S120, arranging the wind speed values ​​at the multiple predetermined time points including the current time point into a wind speed input vector, and then obtaining a multi-scale wind speed feature vector through a first multi-scale neighborhood feature extraction module; S130, arranging the pitch angles at the multiple predetermined time points including the current time point into a pitch angle input vector, and then obtaining a multi-scale pitch angle feature vector through a second multi-scale neighborhood feature extraction module. S140, after arranging the output power of the wind turbines at multiple predetermined time points, including the current time point, into a power input vector, the multi-scale power feature vector is obtained through a third multi-scale neighborhood feature extraction module; S150, a Bayesian probability model is used to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector to obtain a posterior probability vector; S160, based on the mean and variance of the feature value set at all positions in the posterior probability vector, the feature values ​​at each position in the posterior probability vector are corrected to obtain a corrected posterior probability vector; and S170, the corrected posterior probability vector is passed through a classifier to obtain a classification result, the classification result being used to indicate whether the pitch angle at the current time point should increase or decrease.

[0101] Figure 5 The illustration shows a schematic diagram of the architecture of a wind turbine variable speed and pitch control method based on active disturbance rejection control according to an embodiment of this application. Figure 5 As shown, in the network architecture of the wind turbine variable speed and pitch control method based on active disturbance rejection control, firstly, the obtained wind speed values ​​at multiple predetermined time points, including the current time point (e.g., as shown in the diagram), are... Figure 5 The P1 shown is arranged as the wind speed input vector (e.g., as shown in the figure). Figure 5 The V1 shown in the diagram is then processed by the first multi-scale neighborhood feature extraction module (e.g., as shown in the diagram). Figure 5 MS1 as shown in the figure is used to obtain multi-scale wind speed feature vectors (e.g., such as MS1) to obtain multi-scale wind speed feature vectors (e.g., MS1) Figure 5 The VF1 shown in the figure); then, the obtained pitch angles at multiple predetermined time points, including the current time point (e.g., as shown in the figure); Figure 5 The P2 shown is arranged as the pitch angle input vector (e.g., as shown in the figure).Figure 5 The V2 shown in the diagram is then processed by a second multi-scale neighborhood feature extraction module (e.g., such as...). Figure 5 The MS2 shown is used to obtain multi-scale pitch angle feature vectors (e.g., such as MS2) to obtain multi-scale pitch angle feature vectors (e.g., MS2) Figure 5 The VF2 shown in the figure); then, the output power of the wind turbine at multiple predetermined time points, including the current time point, is obtained (e.g., as shown in the figure). Figure 5 The P3 shown is arranged as a power input vector (e.g., as shown in the figure). Figure 5 The V3 shown in the diagram is then processed by a third multi-scale neighborhood feature extraction module (e.g., such as...). Figure 5 The MS3 shown is used to obtain multi-scale power feature vectors (e.g., such as MS3). Figure 5 The VF3 shown in the image); then, a Bayesian probability model (e.g., as shown in the image) is used. Figure 5 The multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector (as shown in the diagram) are fused to obtain a posterior probability vector (e.g., as shown in the diagram). Figure 5 (as shown in VP1); then, based on the mean and variance of the feature value set at all positions in the posterior probability vector, the feature values ​​at each position in the posterior probability vector are corrected to obtain a corrected posterior probability vector (e.g., as shown in VP1); ​ The VP2 shown in the diagram); and finally, the corrected posterior probability vector is passed through a classifier (e.g., such as...). ​ The circle S shown in the figure is used to obtain the classification result, which is used to indicate whether the pitch angle at the current time point should increase or decrease.

[0102] More specifically, in steps S110 and S120, the wind speed values of a plurality of predetermined time points including the current time point, the pitch angles of the plurality of predetermined time points including the current time point, and the output powers of the wind power generator at the plurality of predetermined time points including the current time point are acquired by the wind speed sensor, and after the wind speed values of the plurality of predetermined time points including the current time point are arranged into a wind speed input vector, a first multi-scale neighborhood feature extraction module is used to obtain a multi-scale wind speed feature vector. It can be understood that, in order to protect the wind turbine from being damaged when the pitch system of the wind power generator is regulated to control the wind power conversion efficiency, the existing pitch system is adjusted based on the wind power situation at the current time point, so that the adjusted pitch angle cannot keep the output power and the rotor speed at their rated values, and thus the grid fluctuation is aggravated. Therefore, in the technical solution of the present application, it is expected to dynamically perform data prediction based on the dynamic change characteristic information of the wind speed values, the pitch angles, and the output powers of the wind power generator at the plurality of predetermined time points including the current time point, and then to complete the variable speed variable pitch control of the wind power generator, and in this process, a Bayesian model is introduced to perform predictive adjustment, so that the predicted result is more accurate, and the output power and the rotor speed can be kept at their rated values to control the wind power conversion efficiency and protect the wind turbine from being damaged in the case of stabilizing the grid fluctuation.

[0103] That is, specifically, in the technical solution of the present application, first, the wind speed values of a plurality of predetermined time points including the current time point are acquired by the wind speed sensor, and the pitch angles of the plurality of predetermined time points including the current time point are acquired by the angle sensor, and the output powers of the wind power generator at the plurality of predetermined time points including the current time point are acquired by the power detector.

[0104] Then, it can be understood that the convolutional neural network is originally a model applied in the field of image, but the idea of local feature extraction can also be applied to time series data analysis. For example, a time series convolution structure with a convolution kernel size of 3, for time series data input, the convolution kernel moves in the form of a sliding window along the time dimension, and outputs the weighted sum of the data in each time series segment. Each convolution unit stacks multiple convolution kernels to output multi-dimensional features. Large convolution kernels extract features from large-scale time series neighborhoods, where each value in the neighborhood has a smaller impact, thereby reducing the fluctuations in the input data and reducing the impact of noise points in the input data on the output features. However, large-scale convolution kernels reduce the difference in value changes, which can cause over-smoothing and loss of discriminative ability of the output features. In contrast, small-scale convolution kernels can better preserve the information in the input data, but are more susceptible to noise points. Therefore, considering the characteristics of convolution of different scales, different sizes of convolution units are combined to extract features of different time scales. Then, feature fusion is completed by feature concatenation to obtain multi-scale neighborhood features.

[0105] More specifically, in steps S130 and S140, the pitch angles at the plurality of predetermined time points including the current time point are arranged into a pitch angle input vector and then passed through a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector, and the output powers of the wind power generator at the plurality of predetermined time points including the current time point are arranged into a power input vector and then passed through a third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector. That is, in the technical solution of the present application, similarly, for the pitch angles at the plurality of predetermined time points including the current time point and the output powers of the wind power generator at the plurality of predetermined time points including the current time point, considering that the pitch angles and the output powers of the wind power generator also have dynamic implicit change characteristics in the time dimension, in order to extract the multi-scale neighborhood correlation of such dynamic change characteristics, the pitch angles at the plurality of predetermined time points including the current time point and the output powers of the wind power generator at the plurality of predetermined time points including the current time point are respectively input into the multi-scale neighborhood feature extraction module for processing to obtain a multi-scale pitch angle feature vector and a multi-scale power feature vector. In this way, the neighborhood correlation features of different scales of the dynamic change characteristics of the output pitch angles and the output powers of the wind power generator can not only contain smoothed features, but also preserve the original input feature information, so as to avoid information loss and improve the accuracy of subsequent classification.

[0106] More specifically, in step S150 and step S160, the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector are fused using a Bayesian probability model to obtain a posterior probability vector, and the feature values in each position of the posterior probability vector are corrected based on the mean and variance of the feature value set of all positions in the posterior probability vector to obtain a corrected posterior probability vector. It should be understood that, considering the use of the multi-scale pitch angle feature vector as a prior probability, the purpose of the technical solution of the present application is to update the prior probability to obtain the posterior probability when there is new evidence, i.e., when there is a new wind speed value change. Then according to the Bayes formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability, so in the technical solution of the present application, the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector are fused using a Bayesian probability model to obtain a posterior probability vector, wherein the multi-scale pitch angle feature vector is taken as a prior, the multi-scale wind speed feature vector is taken as an event, and the multi-scale power feature vector is taken as evidence.

[0107] However, since the Bayesian probability model is point-by-point calculation for the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector, the obtained posterior probability vector may have the problem of poor expression consistency between feature values, thereby affecting the classification effect. Therefore, in the technical solution of the present application, the posterior probability vector is further subjected to length-dimension recursive squeezing-activation optimization. Here, the length-dimension recursive squeezing-activation optimization activates the length-dimension recursive of the feature distribution based on the statistical characteristics of the feature set, thereby inferring the distribution of each sampling point of the feature in its length dimension, and adopts a predetermined dimension squeezing-activation mechanism composed of ReLU-Sigmoid function to obtain a confidence value enhanced by a predetermined dimension attention, so as to improve the expression consistency of the high-dimensional feature flow pattern of the posterior probability vector in the length dimension, and further improve the classification accuracy.

[0108] More specifically, in step S170, the corrected posterior probability vector is passed through a classifier to obtain a classification result, which is used to indicate whether the pitch angle at the current time point should be increased or decreased.

[0109] In summary, the variable speed variable pitch control method for wind driven generators based on active disturbance rejection control according to the embodiments of the present application is illustrated, which adopts the control technology of artificial intelligence, extracts the multi-scale neighborhood correlation of the dynamic change characteristics of the wind speed values, the pitch angles and the output power of the wind driven generators at multiple predetermined time points including the current time point through the multi-scale neighborhood feature extraction module, dynamically performs data prediction, and then completes the variable speed variable pitch control of the wind driven generators, and in the process, the Bayesian model is introduced for predictive adjustment, so that the predicted result is more accurate. In this way, the output power and the rotor speed can be kept at their rated values in the case of stable grid fluctuations, so as to control the wind power conversion efficiency and protect the wind turbine from damage.

[0110] The above describes the basic principles of the present application in combination with specific embodiments, but it should be pointed out that the advantages, advantages, effects and the like mentioned in the present application are only examples and not limitations, and these advantages, advantages, effects and the like cannot be considered as the must-have of each embodiment of the present application. In addition, the above specific details are only for the purpose of example and for the purpose of understanding, and are not limited to the above specific details to realize the present application.

[0111] The block diagrams of the devices, apparatuses, equipment, systems involved in the present application are only illustrative examples and are not intended to require or imply the connection, arrangement, configuration shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, systems can be connected, arranged, configured in any manner. Words such as "include", "contain", "have" and the like are open-ended words, mean "include but not limited to", and can be used interchangeably. The words "or" and "and" used herein mean the word "and / or", and can be used interchangeably unless the context clearly indicates otherwise. The word "such as" used herein means the phrase "such as but not limited to", and can be used interchangeably.

[0112] It should also be noted that in the devices, equipment and methods of the present application, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions of the present application.

[0113] The above description of the disclosed aspects is provided so that any person skilled in the art can make or use the present application. Various modifications to these aspects will be apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but is intended to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0114] The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the application to the forms disclosed herein. Although various example aspects and embodiments have been discussed above, those of ordinary skill in the art will appreciate a variety of modifications, alternatives, permutations, additions, and sub-combinations, which fall within the scope of the application.

Claims

1. A variable speed variable pitch control system for a wind turbine based on active disturbance rejection control, characterized in that, The method comprises: a data acquisition module configured to acquire wind speed values at a plurality of predetermined time points including a current time point, pitch angles at the plurality of predetermined time points including the current time point, and output powers of the wind turbine at the plurality of predetermined time points including the current time point; a wind speed data encoding module configured to arrange the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector, and then input the wind speed input vector into a first multi-scale neighborhood feature extraction module to obtain a multi-scale wind speed feature vector; a pitch angle data encoding module configured to arrange the pitch angles at the plurality of predetermined time points including the current time point into a pitch angle input vector, and then input the pitch angle input vector into a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector; a power data encoding module configured to arrange the output powers of the wind turbine at the plurality of predetermined time points including the current time point into a power input vector, and then input the power input vector into a third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector; a Bayesian fusion module configured to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector using a Bayesian probability model to obtain a posterior probability vector; a posterior information correction module configured to correct feature values at each position in the posterior probability vector based on a mean and a variance of a set of feature values at all positions in the posterior probability vector to obtain a corrected posterior probability vector; and a control result generation module configured to input the corrected posterior probability vector into a classifier to obtain a classification result, wherein the classification result is used to indicate whether the pitch angle at the current time point should be increased or decreased. The wind speed data encoding module comprises:

2. The active disturbance rejection control based variable speed variable pitch control system for wind turbine generator of claim 1, wherein, a wind speed vectorization unit configured to arrange the wind speed values at the plurality of predetermined time points including the current time point into the wind speed input vector; a wind speed first-scale convolutional encoding unit configured to input the wind speed input vector into a first convolutional layer of the first multi-scale neighborhood feature extraction module to obtain a first neighborhood-scale wind speed correlation feature vector, wherein the first convolutional layer has a first one-dimensional convolutional kernel with a first length; a wind speed second-scale convolutional encoding unit configured to input the wind speed input vector into a second convolutional layer of the first multi-scale neighborhood feature extraction module to obtain a second neighborhood-scale wind speed correlation feature vector, wherein the second convolutional layer has a second one-dimensional convolutional kernel with a second length, and the first length is different from the second length; and a wind speed multi-scale feature concatenation unit configured to concatenate the first neighborhood-scale wind speed correlation feature vector and the second neighborhood-scale wind speed correlation feature vector to obtain the multi-scale wind speed feature vector. The pitch angle data encoding module comprises:

3. The active disturbance rejection control based variable speed variable pitch control system for wind turbine generator of claim 2, wherein, a pitch angle vectorization unit configured to arrange the pitch angles at the plurality of predetermined time points including the current time point into the pitch angle input vector; ​ a pitch angle first scale convolution coding unit, configured to input the pitch angle input vector into a first convolution layer of the second multi-scale neighborhood feature extraction module to obtain a first neighborhood scale pitch angle associated feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; a pitch angle second scale convolution coding unit, configured to input the pitch angle input vector into a second convolution layer of the second multi-scale neighborhood feature extraction module to obtain a second neighborhood scale pitch angle associated feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a pitch angle multi-scale feature concatenation unit, configured to concatenate the first neighborhood scale pitch angle associated feature vector and the second neighborhood scale pitch angle associated feature vector to obtain the multi-scale pitch angle feature vector.

4. The active disturbance rejection control based variable speed variable pitch control system for wind turbine generator of claim 3, wherein, The power data coding module comprises: a power vectorization unit, configured to arrange the output power of the wind turbine at a plurality of predetermined time points including a current time point into the power input vector; a power first scale convolution coding unit, configured to input the power input vector into a first convolution layer of the third multi-scale neighborhood feature extraction module to obtain a first neighborhood scale power associated feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; a power second scale convolution coding unit, configured to input the power input vector into a second convolution layer of the third multi-scale neighborhood feature extraction module to obtain a second neighborhood scale power associated feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and a power multi-scale feature concatenation unit, configured to concatenate the first neighborhood scale power associated feature vector and the second neighborhood scale power associated feature vector to obtain the multi-scale power feature vector.

5. The active disturbance rejection control based variable speed variable pitch control system for wind turbine generator of claim 4, wherein, The Bayesian fusion module is further configured to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector and the multi-scale power feature vector to obtain the posterior probability vector using a Bayesian probability model according to the following formula: wherein the formula is: qi=pi*ai / bi wherein pi is a feature value at each position in the multi-scale pitch angle feature vector, ai and bi are feature values at each position in the multi-scale power feature vector and the multi-scale wind speed feature vector respectively, and qi is a feature value at each position in the posterior probability vector.

6. The active disturbance rejection control based variable speed variable pitch control system for wind turbine generator of claim 5, wherein, The posterior information correction module is further configured to correct the feature value at each position in the posterior probability vector to obtain the corrected posterior probability vector according to the following formula based on the mean and variance of the set of feature values at all positions in the posterior probability vector: wherein the formula is: where v i represents the eigenvalues of each position in the posterior probability vector, v i ' represents the eigenvalues of each position in the corrected posterior probability vector, μ and σ are the mean and variance of the set of eigenvalues v i ∈ V, V represents the posterior probability vector, exp(·) represents the exponential operation on a vector, and the exponential operation with a vector as the power represents the natural exponential function value with the value of each position of the vector as the power.

7. The active disturbance rejection control based variable speed variable pitch control system for wind turbine generator of claim 6, wherein, The control result generation module is further configured to use the classifier to process the corrected posterior probability vector according to a formula to obtain the classification result, where the formula is: softmax{(W n ,B n ):…:(W1,B1)|X} where W1 to W n are weight matrices, B1 to B n are bias vectors, and X is the corrected posterior probability vector.

8. A variable speed variable pitch control method for a wind turbine based on active disturbance rejection control, characterized in that, comprises: obtaining wind speed values at a plurality of predetermined time points including a current time point, pitch angles at the plurality of predetermined time points including the current time point, and output powers of a wind turbine at the plurality of predetermined time points including the current time point, which are collected by a wind speed sensor; arranging the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector, and inputting the wind speed input vector into a first multi-scale neighborhood feature extraction module to obtain a multi-scale wind speed feature vector; arranging the pitch angles at the plurality of predetermined time points including the current time point into a pitch angle input vector, and inputting the pitch angle input vector into a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector; arranging the output power of the wind turbine at the plurality of predetermined time points including the current time point into a power input vector, and inputting the power input vector into a third multi-scale neighborhood feature extraction module to obtain a multi-scale power feature vector; using a Bayesian probability model to fuse the multi-scale pitch angle feature vector, the multi-scale wind speed feature vector, and the multi-scale power feature vector to obtain a posterior probability vector; based on the mean and variance of the set of feature values at all positions in the posterior probability vector, correcting the feature values at each position in the posterior probability vector to obtain a corrected posterior probability vector; and inputting the corrected posterior probability vector into a classifier to obtain a classification result, the classification result being used to indicate whether the pitch angle at the current time point should be increased or decreased. The arranging the wind speed values at the plurality of predetermined time points including the current time point into a wind speed input vector, and inputting the wind speed input vector into a first multi-scale neighborhood feature extraction module to obtain a multi-scale wind speed feature vector, comprises:

9. The active disturbance rejection control-based variable speed variable pitch control method of a wind power generator according to claim 8, characterized in that, arranging the wind speed values at the plurality of predetermined time points including the current time point into the wind speed input vector; inputting the wind speed input vector into a first convolutional layer of the first multi-scale neighborhood feature extraction module to obtain a first neighborhood scale wind speed associated feature vector, wherein the first convolutional layer has a first one-dimensional convolutional kernel of a first length; inputting the wind speed input vector into a second convolutional layer of the first multi-scale neighborhood feature extraction module to obtain a second neighborhood scale wind speed associated feature vector, wherein the second convolutional layer has a second one-dimensional convolutional kernel of a second length, the first length being different from the second length; and concatenating the first neighborhood scale wind speed associated feature vector and the second neighborhood scale wind speed associated feature vector to obtain the multi-scale wind speed feature vector. The arranging the pitch angles at the plurality of predetermined time points including the current time point into a pitch angle input vector, and inputting the pitch angle input vector into a second multi-scale neighborhood feature extraction module to obtain a multi-scale pitch angle feature vector, comprises:

10. The active disturbance rejection control-based variable speed variable pitch control method of a wind power generator according to claim 9, characterized in that, arranging the pitch angles at the plurality of predetermined time points including the current time point into the pitch angle input vector; inputting the pitch angle input vector into a first convolutional layer of the second multi-scale neighborhood feature extraction module to obtain a first neighborhood scale pitch angle associated feature vector, wherein the first convolutional layer has a first one-dimensional convolutional kernel of a first length; inputting the pitch angle input vector into a second convolutional layer of the second multi-scale neighborhood feature extraction module to obtain a second neighborhood scale pitch angle associated feature vector, wherein the second convolutional layer has a second one-dimensional convolutional kernel of a second length, the first length being different from the second length; and concatenating the first neighborhood scale pitch angle associated feature vector and the second neighborhood scale pitch angle associated feature vector to obtain the multi-scale pitch angle feature vector. concatenating the first neighborhood scale pitch angle related feature vector and the second neighborhood scale pitch angle related feature vector to obtain the multi-scale pitch angle feature vector.

Images