Fault discrimination method and system for wind turbine generator

Abstract

The invention relates to a fault judgment method and system for a wind turbine generator. The method comprises the steps that core electrical parameters and auxiliary environment parameters in the wind turbine generator are collected to serve as an initial data set; preprocessing data in the initial data set by adopting a data abnormal value judgment and data alignment method to obtain a target data set; performing noise reduction on the target data set by adopting an improved minimum mean square algorithm to obtain noise-reduced data; a time-frequency domain feature extraction algorithm is adopted to extract time-frequency domain features of the noise reduction data, and aligned time-frequency features are obtained; a dynamic weight generator based on a long short-term memory network carries out fault discrimination on the time-frequency characteristics; and performing fault judgment processing based on the overall feature data of the wind turbine generator to obtain a predicted fault type corresponding to a power grid where the target wind turbine generator is located. According to the method, the influence of factors such as data acquisition, noise reduction processing and time domain feature fusion is fully considered, the reliability degree of the result is determined, and the effect of accurately reflecting the fault state of the wind turbine generator is achieved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a fault diagnosis method and system for wind turbine generators. Background Technology

[0002] The wind power industry is a renewable energy industry that utilizes natural wind energy and converts the kinetic energy of wind into electrical energy through wind turbines. It has advantages such as being clean, renewable, low-carbon, and environmentally friendly, and is of great significance for optimizing the energy structure, reducing carbon emissions, promoting energy transformation, and sustainable development.

[0003] Currently, the identification methods for wind turbine units are usually based on raw data such as vibration, current or temperature collected by sensors. The collected data is then passed through a pre-set threshold. If the data exceeds the threshold, a fault is determined. The classifier is pre-set with a threshold range for faults, and the fault category is determined based on the current data.

[0004] However, during actual operation, wind turbines may experience frequency drift due to mechanical vibration or electromagnetic interference. If the frequency of mechanical vibration is not addressed, the noise reduction effect will be affected, leading to a decrease in time-domain smoothness. Furthermore, current time-frequency domain processing does not further integrate the obtained data, which can easily result in false alarms and make it difficult to obtain accurate fault status of the wind turbine. Summary of the Invention

[0005] Based on the aforementioned problems in the existing technology, the purpose of this invention is to provide a fault diagnosis method and system for wind turbine units.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A fault diagnosis method for wind turbine generators includes: S1, collect the core electrical parameters and auxiliary environmental parameters of the wind turbine as the initial dataset; S2, the data in the initial dataset is preprocessed using methods for outlier detection and data alignment to obtain the target dataset; S3, an improved least mean square algorithm is used to denoise the target dataset to obtain denoised data; S4, a time-frequency domain feature extraction algorithm is used to extract time-frequency domain features from the denoised data to obtain aligned time-frequency features; S5, based on a dynamic weight generator of a long short-term memory network, enhances the fault features of time-frequency features; S6. Based on the overall characteristic data of the wind turbine, fault discrimination processing is performed to obtain the predicted fault type corresponding to the power grid where the target wind turbine is located.

[0007] A further improvement of this invention is that, in S1, the collection of core electrical parameters and auxiliary environmental parameters of the wind turbine as the initial dataset includes: The core electrical parameters include: comprehensive and accurate collection of core electrical parameters and auxiliary environmental parameters by sensors, providing rich and reliable raw data for fault identification of wind turbine units, thereby more accurately capturing the electrical characteristics and environmental interference information of power grid faults, and using the collected data information as the initial dataset.

[0008] A further improvement of the present invention is that, in S2, the data in the initial dataset is preprocessed using a method of data outlier detection and data alignment to obtain the target dataset, and outlier identification is performed on the data in the initial dataset using the 3σ principle based on statistical principles. The initial dataset after outlier identification is aligned using a multi-source data spatiotemporal alignment method, and the integrated data matrix is ​​normalized using Min-Max standardization.

[0009] A further improvement of this invention lies in employing an improved least mean square algorithm to denoise the target dataset, resulting in denoised data. This denoising involves: eliminating environmental noise in the target dataset by setting an environmental noise value; and using an improved variable step size least mean square algorithm adaptive filter, with the following filtering formula:

[0010] in, This represents the filtering error at time k, which is the difference between the original signal and the filter output signal; For preprocessed A single-parameter time-domain sequence in the data, The output of the filter at time k.

[0011] A further improvement of this invention lies in employing a time-frequency domain feature extraction algorithm to extract time-frequency domain features from the denoised data, resulting in aligned time-frequency features. This includes using an improved Huber regression model to extract time-domain features from a window of the denoised data. The formula for the Huber loss function is as follows:

[0012] in, Here is the Huber loss function with parameter δ. The residual is the difference between the model's predicted value and the actual value, and δ is the threshold parameter of the Huber loss function.

[0013] A further improvement of this invention lies in the fact that the fault feature enhancement of time-frequency features based on the dynamic weight generator of the Long Short-Term Memory (LSTM) network includes: inputting time-frequency domain feature information into the model; the LSTM layer learning the dynamic change trend of time-frequency features in the time series through memory units and gating mechanisms, and mining the importance evolution law of time-frequency features at different stages in the fault development process; the fully connected layer mapping the features output by the LSTM layer to weights α, β, and γ, which are used for feature fusion and represent the contribution of time-domain features, frequency-domain features, and cross features, respectively; the weight constraint formula is as follows:

[0014] Where α is the positive time-domain feature weight, β is the frequency-domain feature weight, and γ is the cross-feature weight.

[0015] A fault diagnosis system for wind turbine generators includes: The data acquisition unit collects the core electrical parameters and auxiliary environmental parameters of the wind turbine as the initial dataset; The data preprocessing unit uses outlier detection and data alignment methods to preprocess the data in the initial dataset to obtain the target dataset; The data denoising unit uses an improved least mean square algorithm to denoise the target dataset and obtain denoised data. The time-frequency domain feature extraction unit uses a time-frequency domain feature extraction algorithm to extract time-frequency domain features from the denoised data, and obtains aligned time-frequency features. The fault feature enhancement unit enhances the time-frequency features based on the dynamic weight generator of the long short-term memory network. The fault identification unit performs fault identification processing based on the overall characteristic data of the wind turbine to obtain the predicted fault type corresponding to the power grid where the target wind turbine is located.

[0016] A further improvement of this invention is that, in the data acquisition unit, the core electrical parameters and auxiliary environmental parameters of the wind turbine generator are collected as the initial dataset, including: The core electrical parameters include: comprehensive and accurate collection of core electrical parameters and auxiliary environmental parameters by sensors, providing rich and reliable raw data for fault identification of wind turbine units, thereby more accurately capturing the electrical characteristics and environmental interference information of power grid faults, and using the collected data information as the initial dataset.

[0017] A network-side server includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the fault detection method for wind turbines.

[0018] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned fault diagnosis method for wind turbine generators.

[0019] Compared with the prior art, the present invention has at least the following beneficial technical effects: This invention provides a fault identification method and system for wind turbines. The method involves collecting core electrical parameters and auxiliary environmental parameters from the wind turbine as an initial dataset; preprocessing the data in the initial dataset using outlier detection and data alignment methods to obtain a target dataset; denoising the target dataset using an improved least mean square algorithm to obtain denoised data; extracting time-frequency domain features from the denoised data using a time-frequency domain feature extraction algorithm to obtain aligned time-frequency features; performing fault identification based on a dynamic weight generator using a Long Short-Term Memory (LSTM) network; and performing fault identification processing based on the overall feature data of the wind turbine to obtain the predicted fault type corresponding to the power grid where the target wind turbine is located. By fully considering the influence of factors such as data acquisition, denoising processing, and time-domain feature fusion, the reliability of the results is clarified, achieving the effect of accurately reflecting the fault state of the wind turbine. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a fault diagnosis method for wind turbines according to the present invention.

[0022] Figure 2 This is a structural block diagram of a fault diagnosis system for wind turbines according to the present invention.

[0023] Figure 3 This is a schematic diagram of the network-side server provided by the present invention. Detailed Implementation

[0024] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0025] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0026] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0027] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0028] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0029] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] Example 1 This invention provides a fault identification method for wind turbine generators. The method involves collecting core electrical parameters and auxiliary environmental parameters of the wind turbine generator as an initial dataset; preprocessing the data in the initial dataset using outlier detection and data alignment methods to obtain a target dataset; denoising the target dataset using an improved least mean square algorithm to obtain denoised data; extracting time-frequency domain features from the denoised data using a time-frequency domain feature extraction algorithm to obtain aligned time-frequency features; enhancing the time-frequency features with fault characteristics using a dynamic weight generator based on a Long Short-Term Memory (LSTM) network; and performing fault identification processing based on the overall feature data of the wind turbine generator to obtain the predicted fault type corresponding to the power grid where the target wind turbine generator is located. By collecting core electrical parameters and auxiliary environmental parameters from wind turbines and identifying outliers, the accuracy of the output data was improved. Noise reduction of the target dataset enhanced the smoothness of the time-domain waveform. The use of a time-frequency domain feature extraction algorithm to extract time-frequency domain features from the denoised data improved the accuracy of the joint characterization of five types of faults, including voltage sag, harmonic distortion, and transient faults. By fusing time-frequency domain features, the fault state can be described more comprehensively and accurately, enhancing the ability to express fault characteristics. By fully considering the impact of data acquisition, noise reduction, and time-domain feature fusion, the reliability of the results was clarified, achieving the effect of accurately reflecting the fault state of the wind turbine.

[0031] The following is a detailed description of the implementation details of the fault diagnosis method for wind turbines according to the present invention. The following content is only for ease of understanding and is not necessary for implementing this solution. See Figure 1 S1, collect the core electrical parameters and auxiliary environmental parameters of the wind turbine as the initial dataset.

[0032] Specifically, by comprehensively and accurately collecting core electrical parameters and auxiliary environmental parameters through sensors, rich and reliable raw data are provided for the fault identification of wind turbine units, thereby more accurately capturing the electrical characteristics and environmental interference information of power grid faults, and using the collected data information as the initial dataset.

[0033] The core electrical parameters include: three-phase line voltage, used to monitor voltage stability and identify faults such as voltage sags and harmonic pollution; three-phase line current, used to reflect load fluctuations, equipment failures, and short circuits; grid frequency, which is an indicator of active power balance; active power, used to display factors affecting grid frequency stability; and reactive power, used to display grid voltage stability.

[0034] Auxiliary environmental parameters include: wind turbine speed, used to correct for the interference of mechanical vibration on the acquisition of electrical parameters; environmental noise, used to distinguish between real fault characteristics and noise interference; and temperature, used to measure the temperature of the environment around the wind turbine and the temperature of the wind turbine equipment.

[0035] The following sensors were used during the data acquisition phase: Voltage sensor: accuracy class 0.2 (error < ±0.2% of rated value), bandwidth 0-5kHz (covering harmonics and transient signals); Current sensor: accuracy class 0.2, bandwidth 0-10kHz (captures high-frequency transient current). Frequency sensor: Measurement error < ±0.01Hz, sampling response time <10ms; Power sensor: Active / reactive power measurement accuracy of 0.5 class, supports harmonic power calculation (up to the 31st harmonic). Auxiliary parameter sensors: rotational speed measurement error <±0.1r / min, noise measurement frequency band coverage 2-20kHz, temperature measurement error <±1℃.

[0036] By collecting core electrical parameters and auxiliary environmental parameters of wind turbine units, the electrical characteristics of power grid faults and environmental interference information can be captured.

[0037] S2, the data in the initial dataset is preprocessed using methods for outlier detection and data alignment to obtain the target dataset; Step S21: Outlier identification is performed on the data in the initial dataset using the 3σ principle based on statistical principles.

[0038] The 3σ rule states that if data follows a normal distribution, data points outside the mean ± 3 standard deviations are considered outliers. The specific formula is as follows: Lower Limit = μ-3σ Upper Limit = μ + 3σ Wherein, Lower Limit represents the lower limit of the data, that is, the minimum boundary of normal data; Upper Limit represents the upper limit of the data, that is, the maximum boundary of normal data; μ is the sample mean, representing the average level of the data; σ is the sample standard deviation, used to measure the dispersion of the data, the larger the value, the more dispersed the data; 3 is the multiplier coefficient, which takes a fixed value in the 3σ principle and is used to define the range of outliers.

[0039] Step S22: Use multi-source data spatiotemporal alignment to align the initial dataset after outlier identification.

[0040] Specifically, multi-sensor data, after outlier processing, are aligned to the same timeline based on timestamps, with a time granularity of 1ms to ensure consistency across different sensor types (such as wind speed, vibration, and temperature sensors) in the time dimension. During alignment, missing timestamps are filled using cubic spline interpolation to avoid signal distortion caused by simple linear interpolation. The aligned data is then integrated into a structured data matrix. , (8 represents the number of samples, and 8 represents the parameter dimensions, covering key parameters such as wind speed, rotational speed, voltage, current, vibration amplitude, temperature, humidity, and power).

[0041] Step S23: Normalize the integrated data matrix using Min-Max standardization (deviation standardization).

[0042] Min-Max standardization: Standardizes data and compresses the values ​​to the [0,1] interval. The formula is:

[0043] in, The data has been normalized; after preprocessing the data in the initial dataset, the values ​​range from [0,1]. Represents the original signal matrix One of the original data in the data, Original signal matrix The minimum value of all data in the data; Original signal matrix The maximum value of all data in the data; The target dataset is obtained by identifying outliers in the initial dataset, then processing the initial dataset to remove outliers through data alignment, and finally processing it through normalization.

[0044] S3, an improved least mean square algorithm is used to denoise the target dataset to obtain denoised data; Specifically: by setting the environmental noise value, the environmental noise in the target data set is eliminated to avoid noise interference (the environmental noise is mostly mechanical vibration noise of 5-25Hz, generated by gearbox and blade rotation).

[0045] An improved variable-step-size minimum mean square (LMS) adaptive filter is used, and the filtering formula is as follows:

[0046] in, This represents the filtering error at time k, which is the difference between the original signal and the filter output signal; For preprocessed A single-parameter time-domain sequence in the data, The output of the filter at time k; The filtered values ​​are updated using a step-size formula. The step size determines the adjustment of the weight vector in each iteration. The step-size update formula is as follows:

[0047] in, This represents the step size at time k. The maximum step size is set. The minimum step size is set. The parameters used to control the step decay rate, This refers to the filtering error in the above filtering error formula; The weight update formula is as follows:

[0048] in, This is the weight vector updated at time k+1; This is the weight vector at the k-th time step; The step size at the k-th time step; Let be the filtering error at time k; This is the reference noise window vector at time k; By performing noise reduction processing on the target dataset, the smoothness of the time-domain waveform is improved, and noise-reduced data is obtained.

[0049] S4, a time-frequency domain feature extraction algorithm is used to extract time-frequency domain features from the denoised data to obtain aligned time-frequency features; Specifically, an improved Huber regression model (a robust regression method) is used to extract temporal features from a window of denoised data: The formula for the Huber loss function is as follows:

[0050] in, Here is the Huber loss function with parameter δ. The residual is the difference between the model's predicted value and the actual value. δ is the threshold parameter of the Huber loss function, δ=1.5. The feature indicators are: for voltage and current sequences, extract the rate of change of amplitude and peak value at the beginning and end of the window, forming four time-domain features, which together form a time-domain feature vector. .

[0051] Specifically, in wind turbine grid fault diagnosis, voltage and current sequences contain rich fault information. Amplitude change rate ( This reflects the rate of change of voltage and current within a time window. Rapid amplitude changes indicate the occurrence of a fault; for example, during a short-circuit fault, the rate of change of current amplitude increases sharply. Peak value ( The peak values ​​represent the maximum voltage and current intensity within the specified window. Abnormal peak values ​​may indicate faults such as overvoltage or overcurrent. By extracting these four time-domain features from the beginning and end of the window, the dynamic changes in voltage and current can be captured from a temporal perspective, ultimately forming a time-domain feature vector. .

[0052] Using Morlet (wavelet transform, a signal analysis method based on complex-valued Morlet wavelet functions), from... Extracting frequency domain features: The wavelet transform formula is as follows:

[0053] in, The result is the wavelet transform; a is the scale parameter; b is the time offset parameter; x(t) is the original signal. Morlet wavelet function The complex conjugate, To perform scaling and translation operations on time t, we use a to scale the time scale and b to translate the time position, where dt is the derivative of the integral variable t.

[0054] There are four typical faults in the characteristic frequency band, as follows: 0-5Hz (frequency fluctuation): (a corresponds to 0-5Hz); In power systems, frequency fluctuations usually occur in lower frequency bands, and the 0-5Hz band is mainly used to capture abnormal changes in the power grid frequency; This indicates the calculation of signals within this frequency band. The average of absolute values ​​quantifies the degree of frequency fluctuation. 45-55Hz (fundamental frequency anomaly): (a corresponds to 45-55Hz); In power systems, the 45-55Hz frequency band is mainly used to detect abnormal deviations or instabilities in the fundamental frequency. 150-250Hz (harmonic distortion): (a corresponds to 150-250Hz); Harmonics in power systems are usually integer multiples of the fundamental frequency. The 150-250Hz frequency band contains the main frequency range of the 3rd and 5th harmonics and is often used to detect harmonic distortion.

[0055] 1kHz (transient fault): When a transient fault occurs, high-frequency signal components are generated. The frequency band above 1kHz is used to capture the high-frequency characteristics generated by the transient fault.

[0056] Based on the requirements of fault characteristic analysis, four key frequency bands (such as the fundamental frequency band, harmonic frequency band, etc.) are selected, and frequency domain feature vectors are output. .

[0057] Finally, the aligned time-frequency characteristics are obtained as follows: .

[0058] S5 uses a dynamic weight generator based on a Long Short-Term Memory (LSTM) network to enhance fault features in time-frequency features.

[0059] Specifically, time-domain features and frequency domain features As input, time-frequency domain feature information is fed into the model; the LSTM layer learns the dynamic change trend of time-frequency features in the time series through memory units and gating mechanisms, and explores the importance evolution law of time-frequency features at different stages in the fault development process; the fully connected layer maps the features output by the LSTM layer to weights α, β, γ, which are used for feature fusion and represent the contribution of time-domain features, frequency-domain features and cross features, respectively.

[0060] The weight constraint formula is as follows:

[0061] Where α is the positive time-domain feature weight, β is the frequency-domain feature weight, and γ is the cross-feature weight.

[0062] The above weight constraint formula ensures that the sum of the relative importance of each feature in the fusion process is 1, thus avoiding the imbalance of weights that could lead to some features being overemphasized or ignored.

[0063] The feature fusion formula is as follows:

[0064] in, This is element-wise multiplication; This means that the time-domain features are weighted according to the calculated time-domain weights, highlighting the role of key information in the time-domain features under the current fault state; Weighting of frequency domain features highlights the role of key information in the frequency domain features under the current fault state; By using element-wise multiplication, we can uncover the synergistic relationship between time-domain features and frequency-domain features; The above fusion formula enables the fused features to describe the fault state more comprehensively and accurately, enhancing the ability to express fault features.

[0065] S6. Based on the overall characteristic data of the wind turbine, fault discrimination processing is performed to obtain the predicted fault type corresponding to the power grid where the target wind turbine is located.

[0066] Specifically, the standardized fusion features after attention-weighted fusion As the core input fault detection module, it transforms the original features into dimensions that are suitable for the network input through feature dimension mapping operations. Combined with batch normalization (BN) technology, it alleviates the gradient vanishing problem and improves the model training stability and convergence speed.

[0067] In the feature extraction stage, features will be fused. The dimension is reduced from (M×8) to (224×224), and a pre-trained ResNet-34 backbone network is used. Its residual connection structure can deeply mine the hierarchical expression of fault features and capture subtle changes in voltage and current signals at different scales. To solve the problem of distribution offset between the source domain (historical fault data) and the target domain due to differences in operating conditions, a domain adaptation layer containing a gradient reversal layer (GRL) is introduced.

[0068] A three-stage course learning fine-tuning strategy is adopted: the first stage fixes the backbone network parameters and trains only the classifier to quickly establish basic classification capabilities; the second stage unfreezes some backbone network layers and trains them together with a smaller learning rate to gradually adapt to the target domain data; the third stage fine-tunes all parameters globally and uses an adaptive learning rate optimizer to further converge.

[0069] Finally, a classifier constructed through a fully connected layer outputs the predicted probability distribution of five fault types, including voltage sag, frequency fluctuation, harmonic distortion, voltage imbalance, and power oscillation; thus obtaining the predicted fault type corresponding to the power grid where the target wind turbine is located.

[0070] This invention provides a fault identification method for wind turbine generators. The method involves collecting core electrical parameters and auxiliary environmental parameters of the wind turbine generator as an initial dataset; preprocessing the data in the initial dataset using outlier detection and data alignment methods to obtain a target dataset; denoising the target dataset using an improved least mean square algorithm to obtain denoised data; extracting time-frequency domain features from the denoised data using a time-frequency domain feature extraction algorithm to obtain aligned time-frequency features; enhancing the time-frequency features with fault characteristics using a dynamic weight generator based on a Long Short-Term Memory (LSTM) network; and performing fault identification processing based on the overall feature data of the wind turbine generator to obtain the predicted fault type corresponding to the power grid where the target wind turbine generator is located. By collecting core electrical parameters and auxiliary environmental parameters from wind turbines and identifying outliers, the accuracy of the output data was improved. Noise reduction of the target dataset enhanced the smoothness of the time-domain waveform. The use of a time-frequency domain feature extraction algorithm to extract time-frequency domain features from the denoised data improved the accuracy of the joint characterization of five types of faults, including voltage sag, harmonic distortion, and transient faults. By fusing time-frequency domain features, the fault state can be described more comprehensively and accurately, enhancing the ability to express fault characteristics. By fully considering the impact of data acquisition, noise reduction, and time-domain feature fusion, the reliability of the results was clarified, achieving the effect of accurately reflecting the fault state of the wind turbine.

[0071] Example 2 like Figure 2 As shown, the present invention provides a fault diagnosis system for wind turbine generators, comprising: The data acquisition unit collects the core electrical parameters and auxiliary environmental parameters of the wind turbine as the initial dataset; The data preprocessing unit uses outlier detection and data alignment methods to preprocess the data in the initial dataset to obtain the target dataset; The data denoising unit uses an improved least mean square algorithm to denoise the target dataset and obtain denoised data. The time-frequency domain feature extraction unit uses a time-frequency domain feature extraction algorithm to extract time-frequency domain features from the denoised data, and obtains aligned time-frequency features. The fault feature enhancement unit enhances the time-frequency features based on the dynamic weight generator of the long short-term memory network. The fault identification unit performs fault identification processing based on the overall characteristic data of the wind turbine to obtain the predicted fault type corresponding to the power grid where the target wind turbine is located.

[0072] In the data acquisition unit of this embodiment, the core electrical parameters and auxiliary environmental parameters of the wind turbine generator are collected as the initial dataset, including: The core electrical parameters include: comprehensive and accurate collection of core electrical parameters and auxiliary environmental parameters by sensors, providing rich and reliable raw data for fault identification of wind turbine units, thereby more accurately capturing the electrical characteristics and environmental interference information of power grid faults, and using the collected data information as the initial dataset.

[0073] Example 3 like Figure 3 As shown, the present invention provides a network-side server including at least one processor 302; and a memory 301 communicatively connected to at least one processor 302; wherein the memory 301 stores instructions that can be executed by at least one processor 302, and the instructions are executed by at least one processor 302 to enable at least one processor 302 to perform the above-described data processing method.

[0074] The memory 301 and processor 302 are connected via a bus, which may include any number of interconnecting buses and bridges. The bus connects various circuits of one or more processors 302 and memory 301 together. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 302 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 302.

[0075] Processor 302 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 301 can be used to store data used by processor 302 during operation.

[0076] Example 4 This invention provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements a fault diagnosis method for wind turbine generators.

[0077] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] This application is described with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0082] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A fault diagnosis method for wind turbine generators, characterized in that, include: S1, collect the core electrical parameters and auxiliary environmental parameters of the wind turbine as the initial dataset; S2, the data in the initial dataset is preprocessed using methods for outlier detection and data alignment to obtain the target dataset; S3, an improved least mean square algorithm is used to denoise the target dataset to obtain denoised data; S4, a time-frequency domain feature extraction algorithm is used to extract time-frequency domain features from the denoised data to obtain aligned time-frequency features; S5, based on a dynamic weight generator of a long short-term memory network, enhances the fault features of time-frequency features; S6. Based on the overall characteristic data of the wind turbine, fault discrimination processing is performed to obtain the predicted fault type corresponding to the power grid where the target wind turbine is located.

2. The fault diagnosis method for wind turbine units according to claim 1, characterized in that, In S1, the core electrical parameters and auxiliary environmental parameters of the wind turbine generator are collected as the initial dataset, including: The core electrical parameters include: comprehensive and accurate collection of core electrical parameters and auxiliary environmental parameters by sensors, providing rich and reliable raw data for fault identification of wind turbine units, thereby more accurately capturing the electrical characteristics and environmental interference information of power grid faults, and using the collected data information as the initial dataset.

3. The fault diagnosis method for wind turbine units according to claim 2, characterized in that, In S2, the data in the initial dataset is preprocessed using the methods of outlier detection and data alignment to obtain the target dataset, which includes outlier identification of the data in the initial dataset using the 3σ principle based on statistical principles. The initial dataset after outlier identification is aligned using a multi-source data spatiotemporal alignment method, and the integrated data matrix is ​​normalized using Min-Max standardization.

4. The fault diagnosis method for wind turbine units according to claim 1, characterized in that, In S3, an improved least mean square algorithm is used to denoise the target dataset. The denoised data includes: eliminating environmental noise in the target dataset by setting an environmental noise value; and using an improved variable step size least mean square algorithm adaptive filter, with the following filtering formula: in, This represents the filtering error at time k, which is the difference between the original signal and the filter output signal; For preprocessed A single-parameter time-domain sequence in the data, The output of the filter at time k.

5. The fault diagnosis method for wind turbine units according to claim 1, characterized in that, In S4, a time-frequency domain feature extraction algorithm is used to extract time-frequency domain features from the denoised data. The aligned time-frequency features include: using an improved Huber regression model to extract time-domain features from the window of the denoised data. The formula for the Huber loss function is as follows: in, Here is the Huber loss function with parameter δ. The residual is the difference between the model's predicted value and the actual value, and δ is the threshold parameter of the Huber loss function.

6. The fault diagnosis method for wind turbine units according to claim 1, characterized in that, In S5, the dynamic weight generator based on the Long Short-Term Memory (LSTM) network enhances the fault features of time-frequency characteristics by: inputting time-frequency domain feature information into the model; the LSTM layer, through memory units and gating mechanisms, learns the dynamic changing trend of time-frequency features in the time series, and explores the evolution of the importance of time-frequency features at different stages in the fault development process; the fully connected layer maps the features output by the LSTM layer to weights α, β, and γ, which are used for feature fusion and represent the contribution of time-domain features, frequency-domain features, and cross-features, respectively; the weight constraint formula is as follows: Where α is the positive time-domain feature weight, β is the frequency-domain feature weight, and γ is the cross-feature weight.

7. A fault diagnosis system for wind turbine generators, characterized in that, include: The data acquisition unit collects the core electrical parameters and auxiliary environmental parameters of the wind turbine as the initial dataset; The data preprocessing unit uses outlier detection and data alignment methods to preprocess the data in the initial dataset to obtain the target dataset; The data denoising unit uses an improved least mean square algorithm to denoise the target dataset and obtain denoised data. The time-frequency domain feature extraction unit uses a time-frequency domain feature extraction algorithm to extract time-frequency domain features from the denoised data, and obtains aligned time-frequency features. The fault feature enhancement unit enhances the time-frequency features based on the dynamic weight generator of the long short-term memory network. The fault identification unit performs fault identification processing based on the overall characteristic data of the wind turbine to obtain the predicted fault type corresponding to the power grid where the target wind turbine is located.

8. A fault diagnosis system for wind turbine generators according to claim 7, characterized in that, The data acquisition unit collects core electrical parameters and auxiliary environmental parameters from the wind turbine as the initial dataset, including: The core electrical parameters include: comprehensive and accurate collection of core electrical parameters and auxiliary environmental parameters by sensors, providing rich and reliable raw data for fault identification of wind turbine units, thereby more accurately capturing the electrical characteristics and environmental interference information of power grid faults, and using the collected data information as the initial dataset.

9. A network-side server, characterized in that, include: At least one processor; The system includes a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the fault diagnosis method for wind turbines as described in any one of claims 1 to 6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the fault diagnosis method for wind turbines as described in any one of claims 1 to 6.

Images