A wind turbine input torque intelligent identification method and system and a storage medium

Abstract

The present application relates to the technical field of wind turbine operation state monitoring, in particular to a wind turbine input torque intelligent identification method, system and storage medium, comprising: S1, constructing a training data set of wind turbine state data; S2, constructing an intelligent identification model based on a radial basis function neural network; S3, inputting the training data set into the intelligent identification model to obtain a trained input torque intelligent identification model; S4, constructing a real-time data set of wind turbine state data; S5, inputting the real-time data set into the trained intelligent identification model to obtain an intelligent identification result of the input torque. The present application provides a wind turbine input torque intelligent identification method, which realizes long-term identification and monitoring of wind turbine input torque by means of an intelligent algorithm, does not need to install and maintain strain gauges, and reduces the cost of long-term monitoring of wind turbine input torque.

Description

Technical Field

[0001] This invention relates to the field of wind turbine operating status monitoring technology, specifically to a method, system, and storage medium for intelligent identification of wind turbine input torque. Background Technology

[0002] Acquiring wind turbine input torque is crucial for monitoring the operating status and assessing the health of wind turbines. Long-term, real-time, and accurate measurement and monitoring of wind turbine input torque supports fault diagnosis, condition assessment, lifespan prediction, and optimization of maintenance plans for the wind turbine drivetrain. Furthermore, torque data can be used to optimize wind turbine control strategies, improving overall wind energy utilization efficiency. In the wind energy industry, torque monitoring is one of the essential conditions for achieving intelligent operation and maintenance of wind turbines.

[0003] For the measurement, monitoring, and acquisition of input torque in wind turbine generators, referring to the IEC61400-13 torque test procedure for wind turbine main shafts and related standards, existing technologies generally involve attaching strain gauges to the input main shaft to collect strain data, transmitting the strain signals via wired or wireless connections, and calculating the corresponding input torque under different strains using theoretical formulas. However, this approach has several shortcomings:

[0004] 1) The wired connection uses a conductive ring to transmit the strain signal to the acquisition device. With the increasing size of wind turbines and the ultra-large diameter of the main shaft, the installation and fitting conditions of the conductive ring are more stringent and greatly affected by the actual working conditions. This method is not easy to achieve long-term stable acquisition of input torque and has problems such as large measurement error and low reliability.

[0005] 2) Although the wireless method can solve the difficulties faced by using conductive rings, it is difficult to achieve long-term stable and reliable power supply. It requires manual battery replacement periodically, which is subject to problems such as signal interference and loss. At the same time, the cost of input torque monitoring is increased due to the regular battery replacement maintenance. In particular, the time window for offshore operation and maintenance is greatly affected by environmental factors, making regular replacement difficult.

[0006] 3) In addition, during the 20-25 year design life of the wind turbine, the accuracy and reliability of long-term input torque acquisition are difficult to guarantee due to the strain gauge installation method. Summary of the Invention

[0007] This invention provides an intelligent identification method for wind turbine input torque, which relies on intelligent algorithms to achieve long-term identification and monitoring of wind turbine input torque, while reducing the cost of long-term monitoring of wind turbine input torque.

[0008] This application provides the following technical solution:

[0009] A method for intelligent identification of input torque of a wind turbine includes the following steps:

[0010] S1. Construct a training dataset for the state data of wind turbine units;

[0011] S2. Construct an intelligent recognition model based on radial basis function neural networks;

[0012] S3. Input the training dataset into the intelligent recognition model to obtain the trained input torque intelligent recognition model;

[0013] S4. Construct a real-time dataset of wind turbine status data;

[0014] S5. Input the real-time dataset into the trained intelligent recognition model to obtain the intelligent recognition result of the input torque.

[0015] Furthermore, S1 includes:

[0016] S11. Collect the status data of the wind turbine generator set, including the main shaft speed n_II. i Radial vibration acceleration a_II of the first-stage gear ring 1i Secondary gear ring radial vibration acceleration a_II 2i Radial vibration acceleration a_II of the third-stage gear ring 3i Radial vibration acceleration a_II of high-speed output shaft 4i and principal axis strain ε;

[0017] S12. Calculate the input torque T_I based on the spindle strain ε. i ;

[0018] S13, Radial vibration acceleration of the first-stage gear ring (a_II) 1i ), Secondary gear ring radial vibration acceleration (a_II) 2i ), radial vibration acceleration of the third-stage gear ring (a_II) 3i ), high-speed output shaft radial vibration acceleration (a_II) 4i The effective value is calculated with an interval of 1 second.

[0019]

[0020] n is the number of measurement points per second, which is equivalent to the sampling rate of the vibration sensor;

[0021] S14, Regarding the spindle speed n_II i Input torque T_I i and the effective value of vibration acceleration rms_I ji Perform data cleaning;

[0022] S15. The input rotational speed (N_I) after data cleaning i ), RMS value of vibration acceleration (RMS_I)ji ) and input torque (T1_I) i Establish mapping relationships:

[0023] T1_I i =f(N_I i ,RMS_I ji )

[0024] Where: RMS_I ji =(RMS_I 1i ,RMS_I 2i ,RMS_I 3i ,RMS_I 4i ...).

[0025] Furthermore, in step S12, for a solid spindle, the input torque T_I is calculated using the following formula. i :

[0026]

[0027] In the formula, D is the principal axis diameter, E is the principal axis material elastic modulus, ε is the principal axis strain, μ is Poisson's ratio, and N is the number of bridge arms.

[0028] For a hollow spindle, the input torque T_I is calculated using the following formula. i :

[0029]

[0030] In the formula, D o Main spindle outer diameter, D i ε is the diameter of the main spindle bore, E is the elastic modulus of the main spindle material, ε is the strain of the main spindle, μ is Poisson's ratio, and N is the number of bridge arms.

[0031] Furthermore, the data cleaning described in S14 includes:

[0032] S141. Perform data cleaning on the spindle speed to remove outliers, noise points, and interference points from the speed signal. The median smoothing method is as follows:

[0033]

[0034] In the formula, m is the window length, Med{n_I i-v ,…,n_I i ,…,n_I i+v} represents the median of the rotational speed sequence corresponding to the monitoring window;

[0035] S142. The effective value rms_I after calculating the vibration acceleration of each channel. jiData cleaning was performed to remove outliers, noise points, and interference points from the effective values ​​of vibration acceleration. Median smoothing was used as follows:

[0036]

[0037] In the formula, m is the window length, Med{rms_I ji-v ,…,rms_I ji ,…,rms_I ji+v} represents the median of the corresponding effective value sequence of vibration acceleration;

[0038] S143. Input torque T_I after spindle strain calculation i Data cleaning is performed to remove outliers, noise points, and interference points from the input torque signal. Median smoothing is used as follows:

[0039]

[0040] In the formula, m is the window length, Med{T_I i-v ,…,T_I i ,…,T_I i+v} represents the median of the corresponding input torque sequence.

[0041] Furthermore, in step S2, the output of the radial basis function neural network is:

[0042]

[0043] Where y j For network output, w ij c represents the neuron weights. i σ is the cluster center, σ is the variance of the basis functions, and h is the number of nodes in the hidden layer.

[0044] Furthermore, the steps of S4 in constructing the real-time dataset include:

[0045] S41, Online monitoring system speed sensor monitors spindle speed n_II i The radial vibration acceleration (a_II) of the first-stage gear ring was collected according to the standard sampling rate. 1i ), Secondary gear ring radial vibration acceleration (a_II) 2i ), radial vibration acceleration of the third-stage gear ring (a_II) 3i ), high-speed output shaft radial vibration acceleration (a_II) 4i );

[0046] S42. The acquired vibration acceleration time-domain signal (a_II) ji The effective value is calculated with an interval of 1 second.

[0047]

[0048] S43, Regarding the spindle speed n_II i The effective value rms_II after calculating vibration acceleration ji Perform data cleaning.

[0049] Furthermore, S43 includes:

[0050] S431, Regarding the spindle speed n_II i Data cleaning is performed to remove outliers, noise, and interference points from the speed signal. The median smoothing method is as follows:

[0051]

[0052] In the formula, m is the window length, Med{n_I i-v ,…,n_I i ,…,n_I i+v} represents the median of the rotational speed sequence corresponding to the monitoring window;

[0053] S432. Effective value rms_II of vibration acceleration calculated for each channel ji Data cleaning was performed to remove outliers, noise points, and interference points from the effective values ​​of vibration acceleration. Median smoothing was used as follows:

[0054]

[0055] In the formula, m is the window length, Med{rms_II ji-v ,…,rms_II ji ,…,rms_II ji+v} represents the median of the corresponding effective value sequence of vibration acceleration.

[0056] The principles and advantages of this invention are as follows:

[0057] Because radial basis functions (RBF) and radial basis function networks have the characteristics of 'local mapping', simple structure, relatively simple training process, and high computational efficiency, and because RBF networks perform well in approximating arbitrary nonlinear functions and have a fast learning speed, radial basis function neural networks are used to model, learn, and train recognition models to obtain implicit nonlinear mapping relationships and realize the recognition of input torque.

[0058] 1. First, the input spindle speed, vibration acceleration at different gear rings, and short-term spindle strain are collected using an online monitoring system (CMS). Then, the collected signals are smoothed, filtered, and outliers are removed before the effective value of the vibration acceleration is calculated.

[0059] 2. Next, an input torque recognition model based on radial basis functions is established, with the input being rotational speed and effective value of vibration acceleration, and the output being the corresponding torque. The network is then solved and trained.

[0060] 3. For newly monitored vibration acceleration and spindle speed, after smoothing and noise reduction, calculate the effective value of vibration acceleration. Substitute the calculated effective values ​​of vibration acceleration at different measuring points and the input speed into the recognition model to perform real-time recognition of input torque.

[0061] The trained input torque recognition network relies on intelligent algorithms to achieve long-term identification and monitoring of wind turbine input torque. The input parameters only include the main shaft speed and vibration acceleration at different measuring points of the gearbox, without using input data of main shaft stress. Therefore, it is not necessary to install strain gauges to collect main shaft strain, which solves the technical difficulties in long-term input torque monitoring and acquisition of wind turbines. The acquisition of state data can be completed based on the online monitoring system, without the need for long-term maintenance of strain gauges, thus reducing the cost of long-term monitoring.

[0062] On the other hand, the present invention provides an intelligent identification system for wind turbine input torque, including a data acquisition module, a data cleaning module, and an intelligent identification module. The data acquisition module is used to collect the status data of the wind turbine, including the main shaft speed n_II. i Radial vibration acceleration a_II of the first-stage gear ring 1i Secondary gear ring radial vibration acceleration a_II 2i Radial vibration acceleration a_II of the third-stage gear ring 3i Radial vibration acceleration a_II of high-speed output shaft 4i and principal axis strain ε;

[0063] The data processing module is used to process the collected wind turbine status data, including an input torque calibration submodule and a data cleaning submodule. The input torque calibration submodule is used to calculate the input torque T_I based on the main shaft strain ε. i The acceleration processing submodule is used for calibrating the intelligent recognition model and for processing the spindle speed n_II. i Radial vibration acceleration a_II of the first-stage gear ring 1i Secondary gear ring radial vibration acceleration a_II 2i Radial vibration acceleration a_II of the third-stage gear ring 3i Radial vibration acceleration a_II of high-speed output shaft 4i Perform data cleaning to obtain the input dataset;

[0064] The intelligent recognition module stores a trained intelligent recognition model for the input torque of the wind turbine, and then processes the cleaned input speed (N_I) data. i), RMS value of vibration acceleration (RMS_I) ji ) and input torque (T1_I) i Establish a mapping relationship and generate intelligent recognition results of input torque based on the input dataset.

[0065] On the other hand, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent identification method for wind turbine input torque as described above. Attached Figure Description

[0066] Figure 1 This is a flowchart of a method for intelligent identification of input torque of a wind turbine generator according to the present invention;

[0067] Figure 2 The original spindle speed data for Example 1;

[0068] Figure 3 The original data of the effective value of vibration acceleration in Example 1;

[0069] Figure 4 The original data for the spindle input torque in Example 1;

[0070] Figure 5 The spindle speed data after data cleaning in Example 1;

[0071] Figure 6 The data is the effective value of vibration acceleration after data cleaning in Example 1;

[0072] Figure 7 The spindle input torque data after data cleaning in Example 1;

[0073] Figure 8 This is a comparison between the input torque identification value and the measured value in Example 1. Detailed Implementation

[0074] This invention uses wind turbine status data collected by an online monitoring system and designs an intelligent identification model using radial basis function networks to achieve long-term identification and monitoring of wind turbine input torque. This effectively solves the technical difficulties in long-term input torque monitoring and collection for wind turbines, while reducing the cost of long-term monitoring of wind turbine input torque.

[0075] The following detailed description illustrates the specific implementation method:

[0076] Example 1

[0077] A method for intelligent identification of input torque of wind turbine generators, such as Figure 1 As shown, it includes the following steps:

[0078] S1. Construct a training dataset for the state data of wind turbine units, including the following sub-steps:

[0079] S11. Collect training status data of the wind turbine generator set. Collect status data of the wind turbine generator set through the online monitoring system, including the main shaft speed n_II. i Radial vibration acceleration a_II of the first-stage gear ring 1i Secondary gear ring radial vibration acceleration a_II 2i Radial vibration acceleration a_II of the third-stage gear ring 3i Radial vibration acceleration a_II of high-speed output shaft 4i And principal axis strain ε.

[0080] Meanwhile, in accordance with the IEC61400-13 wind turbine main shaft torque test procedure and related standards, the online monitoring system attaches strain gauges at a 45° angle to the main shaft axis to measure the strain at the main shaft.

[0081] Figure 2 The raw data for the spindle speed is displayed. Figure 3 The raw data of radial vibration acceleration for each gear ring stage are displayed.

[0082] S12. Calculate the input torque T_I based on the spindle strain ε. i Used for the calibration of intelligent recognition models;

[0083] For solid spindles, calculate using the following formula:

[0084]

[0085] In the formula, D is the principal axis diameter, E is the principal axis material elastic modulus, ε is the principal axis strain, μ is Poisson's ratio, and N is the number of bridge arms.

[0086] For hollow spindles, calculate using the following formula:

[0087]

[0088] In the formula, D o Main spindle outer diameter, D i ε is the diameter of the main spindle bore, E is the elastic modulus of the main spindle material, ε is the strain of the main spindle, μ is Poisson's ratio, and N is the number of bridge arms.

[0089] The raw data of the spindle input torque are as follows: Figure 4 As shown.

[0090] S13, Radial vibration acceleration of the first-stage gear ring (a_II) 1i ), Secondary gear ring radial vibration acceleration (a_II) 2i ), radial vibration acceleration of the third-stage gear ring (a_II) 3i), high-speed output shaft radial vibration acceleration (a_II) 4i The effective value is calculated with an interval of 1 second.

[0091]

[0092] n is the number of measurement points per second, which is equivalent to the sampling rate of the vibration sensor;

[0093] S14, Regarding the spindle speed n_II i Input torque T_I i and the effective value of vibration acceleration rms_I ji Data cleaning includes:

[0094] S141. Perform data cleaning on the spindle speed to remove outliers, noise points, and interference points from the speed signal. The median smoothing method is as follows:

[0095]

[0096] In the formula, m is the window length, Med{n_I i-v ,…,n_I i ,…,n_I i+v} represents the median of the rotational speed sequence corresponding to the monitoring window;

[0097] S142. The effective value rms_I after calculating the vibration acceleration of each channel. ji Data cleaning was performed to remove outliers, noise points, and interference points from the effective values ​​of vibration acceleration. Median smoothing was used as follows:

[0098]

[0099] In the formula, m is the window length, Med{rms_I ji-v ,…,rms_I ji ,…,rms_I ji+v} represents the median of the corresponding effective value sequence of vibration acceleration;

[0100] S143. Input torque T_I after spindle strain calculation i Data cleaning is performed to remove outliers, noise points, and interference points from the input torque signal. Median smoothing is used as follows:

[0101]

[0102] In the formula, m is the window length, Med{T_I i-v ,…,T_I i ,…,T_I i+v} represents the median of the corresponding input torque sequence.

[0103] S15. The input rotational speed (N_I) after data cleaning i ), RMS value of vibration acceleration (RMS_I) ji ) and input torque (T1_I) i Establish mapping relationships:

[0104] T1_I i =f(N_I i ,RMS_I ji )

[0105] Where: RMS_I ji =(RMS_I 1i ,RMS_I 2i ,RMS_I 3i ,RMS_I 4i ...).

[0106] Figure 5 The spindle speed data after the data cleaning step is displayed. Figure 6 The data displayed shows the radial vibration acceleration data of each gear ring after the data cleaning step. Figure 7 The input torque data is displayed after the data cleaning step.

[0107] S2. Construct an intelligent recognition model based on radial basis function neural networks;

[0108] Radial Basis Function Networks (RBFNs) are a special type of feedforward neural network primarily used for regression and classification tasks. Unlike traditional Multilayer Perceptrons (MLPs), RBFNs use radial basis functions as activation functions for their hidden layers.

[0109] In this embodiment, the output of the radial basis function neural network is:

[0110]

[0111] Where y j For network output, w ij c represents the neuron weights. i σ is the cluster center, σ is the variance of the basis functions, and h is the number of nodes in the hidden layer.

[0112] S3. Input the training dataset into the intelligent recognition model to obtain the trained input torque intelligent recognition model;

[0113] Input the training dataset into the intelligent recognition model for solving and training, including:

[0114] Clustering algorithms are used to determine the location of the center points;

[0115] The variance is determined based on the average distance between the center points;

[0116] Calculate the hidden layer output;

[0117] Use the least squares method or other optimization algorithms to determine the weight vector and bias term.

[0118] The recognition results generated by the trained intelligent input torque recognition model are evaluated against the actual measured values, and the results are as follows: Figure 8 As shown in the figure, the identified value is very close to the measured value, indicating that the intelligent recognition method has high accuracy.

[0119] S4. Construct a real-time dataset of wind turbine status data, including:

[0120] S41, Online monitoring system speed sensor monitors spindle speed n_II i The radial vibration acceleration (a_II) of the first-stage gear ring was collected according to the standard sampling rate. 1i ), Secondary gear ring radial vibration acceleration (a_II) 2i ), radial vibration acceleration of the third-stage gear ring (a_II) 3i ), high-speed output shaft radial vibration acceleration (a_II) 4i );

[0121] S42. The acquired vibration acceleration time-domain signal (a_II) ji The effective value is calculated with an interval of 1 second.

[0122]

[0123] S43, Regarding the spindle speed n_II i The effective value rms_II after calculating vibration acceleration ji Data cleaning includes:

[0124] S431, Regarding the spindle speed n_II i Data cleaning is performed to remove outliers, noise, and interference points from the speed signal. The median smoothing method is as follows:

[0125]

[0126] In the formula, m is the window length, Med{n_I i-ν ,…,n_I i ,…,n_I i+v} represents the median of the rotational speed sequence corresponding to the monitoring window;

[0127] S432. Effective value rms_II of vibration acceleration calculated for each channel jiData cleaning was performed to remove outliers, noise points, and interference points from the effective values ​​of vibration acceleration. Median smoothing was used as follows:

[0128]

[0129] In the formula, m is the window length, Med{rms_II ji-v ,…,rms_II ji ,…,rms_II ji+v} represents the median of the corresponding effective value sequence of vibration acceleration.

[0130] S5. Input the real-time dataset into the trained intelligent recognition model to obtain the intelligent recognition result of the input torque.

[0131] Since strain gauges no longer need to be installed in actual use, and long-term maintenance of strain gauges is not required, the cost of long-term monitoring of wind power torque is reduced.

[0132] Example 2

[0133] This embodiment is a wind turbine input torque intelligent identification system, including a data acquisition module, a data cleaning module, and an intelligent identification module. The data acquisition module is used to collect the status data of the wind turbine, including the main shaft speed n_II. i Radial vibration acceleration a_II of the first-stage gear ring 1i Secondary gear ring radial vibration acceleration a_II 2i Radial vibration acceleration a_II of the third-stage gear ring 3i Radial vibration acceleration a_II of high-speed output shaft 4i and principal axis strain ε;

[0134] The data processing module is used to process the collected wind turbine status data, including an input torque calibration submodule and a data cleaning submodule. The input torque calibration submodule is used to calculate the input torque T_I based on the main shaft strain ε. i The acceleration processing submodule is used for calibrating the intelligent recognition model and for processing the spindle speed n_II. i Radial vibration acceleration a_II of the first-stage gear ring 1i Secondary gear ring radial vibration acceleration a_II 2i Radial vibration acceleration a_II of the third-stage gear ring 3i Radial vibration acceleration a_II of high-speed output shaft 4i Perform data cleaning to obtain the input dataset;

[0135] The intelligent recognition module stores a trained intelligent recognition model for the input torque of the wind turbine, and then processes the cleaned input speed (N_I) data. i ), RMS value of vibration acceleration (RMS_I)ji ) and input torque (T1_I) i Establish a mapping relationship and generate intelligent recognition results of input torque based on the input dataset.

[0136] Example 3

[0137] This embodiment is a computer-readable storage medium storing a computer program that, when executed by a processor, implements the intelligent identification method for wind turbine input torque as described above. In some possible implementations, various aspects of the present invention can also be implemented as a program product comprising program code. When the program product is run on a terminal device, the program code causes the terminal device to perform the steps of the various exemplary embodiments described in the foregoing embodiments section of this specification.

[0138] The above are merely embodiments of the present invention, and the invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for intelligent identification of input torque of a wind turbine generator set, characterized in that, Includes the following steps: S1, constructing a training dataset for the state data of wind turbine generators; S1 includes: S11. Collect the status data of the wind turbine generator set, including the main shaft speed. Radial vibration acceleration of the first-stage gear ring Radial vibration acceleration of the second-stage gear ring Radial vibration acceleration of the third-stage gear ring Radial vibration acceleration of high-speed output shaft and principal axis strain ; S12, Based on principal shaft strain Calculate input torque Used for calibrating intelligent recognition models; S13, Radial vibration acceleration of the first-stage gear ring Radial vibration acceleration of the second-stage gear ring Radial vibration acceleration of the third-stage gear ring Radial vibration acceleration of high-speed output shaft The effective value is calculated with an interval of 1 second. n The number of measurement points per second is equal to the sampling rate of the vibration sensor. S14, Adjust spindle speed Input torque and the effective value of vibration acceleration Perform data cleaning; S15. Input speed after data cleaning Effective value of vibration acceleration With input torque Establish mapping relationship: In the formula: ; S2. Construct an intelligent recognition model based on radial basis function neural networks; S3. Input the training dataset into the intelligent recognition model to obtain the trained input torque intelligent recognition model; S4. Construct a real-time dataset of wind turbine status data; S5. Input the real-time dataset into the trained intelligent recognition model to obtain the intelligent recognition result of the input torque.

2. The intelligent identification method for input torque of a wind turbine generator according to claim 1, characterized in that: In step S12, for a solid spindle, the input torque is calculated using the following formula. : In the formula, D Main spindle diameter E The elastic modulus of the main shaft material. Principal axis strain Poisson's ratio, N This represents the number of bridge arms; For a hollow spindle, the input torque is calculated using the following formula. : The outer diameter of the main shaft The diameter of the spindle inner bore. E The elastic modulus of the main shaft material. Principal axis strain Poisson's ratio, N This represents the number of bridge arms.

3. The intelligent identification method for input torque of a wind turbine generator according to claim 2, characterized in that: The data cleaning described in S14 includes: S141. Perform data cleaning on the spindle speed to remove outliers, noise points, and interference points from the speed signal. The median smoothing method is as follows: In the formula m For window length, This represents the median of the rotational speed sequence corresponding to the monitoring window. S142. Effective values ​​of vibration acceleration calculated for each channel Data cleaning was performed to remove outliers, noise points, and interference points from the effective values ​​of vibration acceleration. Median smoothing was used as follows: In the formula m For window length, This represents the median of the corresponding effective value sequence of vibration acceleration. S143, Input torque after spindle strain calculation Data cleaning is performed to remove outliers, noise points, and interference points from the input torque signal. Median smoothing is used as follows: In the formula m For window length, This is the median of the corresponding input torque sequence.

4. The intelligent identification method for input torque of a wind turbine generator according to claim 3, characterized in that: In step S2, the output of the radial basis function neural network is: in For network output, For neuron weights, As cluster center, Let be the variance of the basis functions. h This represents the number of nodes in the hidden layer.

5. The intelligent identification method for input torque of a wind turbine generator according to claim 4, characterized in that: The steps for constructing a real-time dataset using S4 include: S41, Online monitoring system speed sensor monitors spindle speed Radial vibration acceleration of the first-stage gear ring was collected according to the standard sampling rate. Radial vibration acceleration of the second-stage gear ring Radial vibration acceleration of the third-stage gear ring Radial vibration acceleration of high-speed output shaft ; S42. The acquired vibration acceleration time-domain signal The effective value is calculated with an interval of 1 second. S43, Adjust spindle speed and the effective value of vibration acceleration Perform data cleaning.

6. The intelligent identification method for input torque of a wind turbine generator according to claim 5, characterized in that: S43 includes: S431, Adjust spindle speed Data cleaning is performed to remove outliers, noise, and interference points from the speed signal. The median smoothing method is as follows: In the formula m For window length, This represents the median of the rotational speed sequence corresponding to the monitoring window. S432. Effective values ​​of vibration acceleration calculated for each channel Data cleaning was performed to remove outliers, noise points, and interference points from the effective values ​​of vibration acceleration. Median smoothing was used as follows: In the formula m For window length, This represents the median of the corresponding effective value sequence of vibration acceleration.

7. A wind turbine input torque intelligent identification system, using the wind turbine input torque intelligent identification method as described in any one of claims 1-6, characterized in that: It includes a data acquisition module, a data cleaning module, and an intelligent recognition module. The data acquisition module is used to collect the status data of the wind turbine, including the main shaft speed. Radial vibration acceleration of the first-stage gear ring Radial vibration acceleration of the second-stage gear ring Radial vibration acceleration of the third-stage gear ring Radial vibration acceleration of high-speed output shaft and principal axis strain ; The data processing module is used to process the collected wind turbine status data, including an input torque calibration submodule and a data cleaning submodule. The input torque calibration submodule is used to process the data based on the main shaft strain. Calculate input torque The acceleration processing submodule is used for calibrating the intelligent recognition model and for controlling the spindle speed. Radial vibration acceleration of the first-stage gear ring Radial vibration acceleration of the second-stage gear ring Radial vibration acceleration of the third-stage gear ring Radial vibration acceleration of high-speed output shaft Perform data cleaning to obtain the input dataset; The intelligent recognition module stores a trained intelligent recognition model for the input torque of the wind turbine, and cleans the input speed data. Effective value of vibration acceleration With input torque Establish a mapping relationship and generate intelligent recognition results of input torque based on the input dataset.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent identification method for wind turbine input torque as described in any one of claims 1-6.

Images