Fault diagnosis method of wind driven generator based on big data

Abstract

The invention discloses a fault diagnosis method for a wind driven generator based on big data, and relates to the technical field of intelligent diagnosis and maintenance of wind power generation equipment, and the method comprises the steps: collecting the fault diagnosis data of the wind driven generator through a sensor, transmitting the fault diagnosis data to a cloud big data platform for preprocessing, the method comprises the steps of preprocessing fault diagnosis data, performing feature extraction on the preprocessed fault diagnosis data to obtain time domain features and frequency domain features, deeply fusing the time domain features and the frequency domain features into fault feature vectors by adopting a dynamic weight distribution self-attention mechanism, and constructing a fault diagnosis model by adopting a hierarchical feature activated deep neural network DNN in combination with a long and short term memory network LSTM. The fault feature vector is subjected to layer-by-layer nonlinear mapping, the diagnosis result is output, the parameters of the fault diagnosis model are updated according to real-time data, comprehensive perception of the fault of the wind driven generator is achieved, the accuracy of fault diagnosis is improved, and meanwhile the intelligent level and the automation level are enhanced by adopting the deep neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent diagnosis and maintenance of wind power generation equipment, and particularly to a fault diagnosis method for wind turbines based on big data. Background Art

[0002] With the continuous growth of the global demand for renewable energy, wind power generation, as a clean and sustainable energy form, has been widely used. As the core equipment of the wind power generation system, the stability and reliability of the operation state of wind turbines are directly related to the economic benefits of wind farms and the stability of energy supply. Its components are easily affected by multiple factors such as fatigue damage, vibration shock, and environmental erosion, resulting in failures. Traditional fault diagnosis methods are difficult to cope with its complexity. Vibration analysis, acoustic emission monitoring, and temperature monitoring technologies based on signal processing are widely used in the fault diagnosis of wind turbines. These methods usually rely on the extraction and analysis of single-signal features and are difficult to comprehensively reveal the fault characteristics of wind turbines.

[0003] With the development of big data processing technology, fault diagnosis methods based on big data have gradually become a hot topic. By analyzing the multi-dimensional data collected by sensors, the potential laws of the operation state of wind turbines can be mined, providing higher accuracy for fault diagnosis. The operation data of wind turbines usually contains a large amount of noise and redundant information. How to perform efficient preprocessing on these data to improve data quality is an important challenge. Existing feature extraction methods mostly focus on single time-domain and frequency-domain features and lack the ability to deeply fuse time-domain and frequency-domain information, resulting in the diagnostic model being unable to fully mine the correlation and potential information between multi-dimensional features. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a fault diagnosis method for wind turbines based on big data to solve the problem that existing feature extraction methods mostly focus on single time-domain and frequency-domain features and lack the ability to deeply fuse time-domain and frequency-domain information, resulting in the diagnostic model being unable to fully mine the correlation and potential information between multi-dimensional features.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a fault diagnosis method for wind turbines based on big data, which includes:

[0008] Collecting fault diagnosis data of wind turbines through sensors;

[0009] Transmitting the fault diagnosis data to a cloud big data platform for preprocessing;

[0010] Extract features from the preprocessed fault diagnosis data to obtain time domain features and frequency domain features;

[0011] A dynamic weight allocation self-attention mechanism is used to deeply fuse time domain features and frequency domain features into a fault feature vector;

[0012] A deep neural network (DNN) with hierarchical feature activation combined with a long short-term memory (LSTM) network is used to build a fault diagnosis model, which performs layer-by-layer nonlinear mapping on the fault feature vector and finally outputs the diagnosis result.

[0013] Update the parameters of the fault diagnosis model based on real-time data.

[0014] As a preferred solution of the wind turbine fault diagnosis method based on big data of the present invention, wherein: the fault diagnosis data of the wind turbine is collected by the sensor, and the specific steps are:

[0015] Arrange vibration sensors, acoustic sensors, temperature sensors and speed sensors on the blades, gearbox, main shaft and generator of the wind turbine;

[0016] The sampling frequency is set according to user needs to collect fault diagnosis data such as vibration frequency data, acoustic frequency data, temperature data and rotation speed data.

[0017] As a preferred solution of the wind turbine fault diagnosis method based on big data of the present invention, the fault diagnosis data is transmitted to the cloud big data platform for preprocessing, and the specific steps are as follows:

[0018] Use MQTT protocol to transmit fault diagnosis data to the big data platform;

[0019] The minimum-maximum normalization method is used to normalize the fault diagnosis data to obtain dimensionless fault diagnosis data;

[0020] The filtering algorithm is used to filter the noise of dimensionless fault diagnosis data to obtain pure fault diagnosis data;

[0021] The outlier detection algorithm based on Mahalanobis distance is used to remove outliers from the pure fault diagnosis data to obtain fault diagnosis data without outliers.

[0022] Dynamic time warping is used to time align the fault diagnosis data with outliers removed to obtain high-precision aligned fault diagnosis data, and the high-precision aligned fault diagnosis data is used as preprocessed fault diagnosis data.

[0023] As a preferred solution of the fault diagnosis method of the wind turbine based on big data according to the present invention, wherein: the step of extracting features from the preprocessed fault diagnosis data to obtain time-domain features and frequency-domain features is as follows:

[0024] Calculate the mean, standard deviation and maximum value of the preprocessed fault diagnosis data as the time-domain feature f1;

[0025] Perform a fast Fourier transform FFT on the preprocessed fault diagnosis data, and extract the main frequency component and spectral energy as the frequency-domain feature f2.

[0026] As a preferred solution of the fault diagnosis method of the wind turbine based on big data according to the present invention, wherein: the step of using a dynamic weight allocation self-attention mechanism to deeply fuse the time-domain feature and the frequency-domain feature into a fault feature vector is as follows:

[0027] Calculate the weights of the time-domain feature and the frequency-domain feature through the self-attention mechanism, and the expression is:

[0028] α1 = softmax(W1·f1);

[0029] α2 = softmax(W2·f2);

[0030] Among them, α1 represents the weight of the time-domain feature f1, α2 represents the weight of the frequency-domain feature f2, softmax represents the probability distribution function, W1 represents the weight matrix of the time-domain feature f1, which is obtained by learning from historical data, and W2 represents the weight matrix of the frequency-domain feature f2, which is obtained by learning from historical data;

[0031] Based on the time-domain feature and the frequency-domain feature, a weighted fusion method is used for fusion to obtain a fault feature vector, and the expression is:

[0032] F = α1·f1 + α2·f2;

[0033] Among them, F represents the fault feature vector.

[0034] As a preferred solution of the fault diagnosis method of the wind turbine based on big data according to the present invention, wherein: the step of using a deep neural network DNN combined with a long short-term memory network LSTM with hierarchical feature activation to construct a fault diagnosis model and perform a layer-by-layer non-linear mapping on the fault feature vector is as follows:

[0035] Input the fault feature vector, and through the non-linear mapping of the first layer of the network, the expression is:

[0036]

[0037] Among them, s1 represents the output of the first layer of the neural network, It represents the Swish activation function, U1 represents the weight matrix of the first-layer neural network, which is generated through a Gaussian distribution, and b1 represents the bias vector of the first-layer neural network;

[0038] The output s1 of the first-layer neural network is input into the second layer for further non-linear mapping. The expression is:

[0039]

[0040] Among them, s2 represents the output of the second-layer neural network, U2 represents the weight matrix of the second-layer neural network, and b2 represents the bias vector of the second-layer neural network;

[0041] The output s2 after being mapped by the second-layer neural network is input into the long short-term memory network LSTM to capture the dynamic correlation of features in the time dimension. The expression is:

[0042] h t ,c t = LSTM(s2, h t-1 ,c t-1 );

[0043] Among them, t represents the time index, h t represents the hidden state at time t, c t represents the memory cell state at time t, LSTM represents the long short-term memory network LSTM, h t-1 represents the hidden state at the previous moment, c t-1 represents the memory cell state at the previous moment;

[0044] Finally, the long short-term memory network LSTM processes the output s2 of the second-layer neural network at the last moment T, and outputs the hidden state h T at the last time as the comprehensive feature representation, where T represents the last moment.

[0045] As a preferred solution of the fault diagnosis method for a wind turbine based on big data according to the present invention, wherein: the specific steps for the final output of the diagnosis result are:

[0046] The expression of the final output layer is:

[0047] P = Softmax(U3·h T + b3);

[0048] Among them, Softmax represents the activation function Softmax, U3 represents the weight matrix of the output layer, b3 represents the bias term of the output layer, and P represents the predicted probability of the fault category;

[0049] The predicted probability P of the fault category = {p1, p2, …, pa}, which is the predicted probability for each fault category, where a represents the number of fault categories.

[0050] As a preferred solution of the fault diagnosis method for a wind turbine based on big data according to the present invention, wherein: the step of updating the parameters of the fault diagnosis model according to real-time data is as follows:

[0051] Define a loss function for updating the parameters of the fault diagnosis model, and the expression is:

[0052]

[0053] where L t represents the real-time loss at time point t, j represents the index of the number of fault categories, represents the true label of the j-th fault at time point t, represents the occurrence probability of the j-th fault at time point t, and log represents the logarithmic function;

[0054] The weight matrix U d and the bias vector b d The update formulas are:

[0055]

[0056] where d represents the index of the number of hidden layers, U d represents the weight matrix of the d-th hidden layer, b d represents the bias vector of the d-th hidden layer, U ′ d represents the updated weight matrix of the d-th hidden layer, b ′ d represents the updated bias vector of the d-th hidden layer, η represents the learning rate, represents the gradient of the loss function with respect to the weight matrix U d and represents the gradient of the loss function with respect to the bias vector b d and

[0057] In a second aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and wherein: when the computer program is executed by the processor, any step of the fault diagnosis method for a wind turbine based on big data as described in the first aspect of the present invention is implemented.

[0058] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and wherein: when the computer program is executed by the processor, any step of the fault diagnosis method for a wind turbine based on big data as described in the first aspect of the present invention is implemented.

[0059] The beneficial effects of the present invention are as follows: By combining comprehensive acquisition of multi-source data, high-quality preprocessing, multi-dimensional feature fusion, intelligent diagnostic algorithms, and dynamic learning mechanisms, the accuracy, real-time performance, and robustness of wind turbine fault diagnosis are significantly improved, solving problems such as single data dependence, insufficient feature extraction, and model staticization commonly existing in traditional methods. It is applicable to intelligent diagnosis of wind turbines under complex working conditions, providing an innovative technical solution for the safe and efficient operation of wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0061] Figure 1 It is a flowchart of the fault diagnosis method for a wind turbine based on big data in Embodiment 1.

[0062] Figure 2 It is a schematic diagram for calculating the diagnostic result in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will give a detailed description of the specific embodiments of the present invention with reference to the drawings of the specification.

[0064] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0065] Secondly, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0066] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a fault diagnosis method for a wind turbine based on big data, including the following steps:

[0067] S1. Collect fault diagnosis data of the wind turbine through sensors;

[0068] Vibration sensors, acoustic sensors, temperature sensors and rotational speed sensors are arranged on the blades, gearboxes, main shafts and generators of wind turbines;

[0069] High-precision industrial-grade sensors are selected, including accelerometers to collect vibration signals, MEMS microphones to collect acoustic signals, thermocouples to collect temperature signals and optoelectronic tachometers to collect rotational speed data. The reason for choosing these sensors is that they have high sensitivity and reliability and can meet the data collection requirements in the complex operating environment of wind turbines;

[0070] By arranging vibration sensors, acoustic sensors, temperature sensors and rotational speed sensors on the blades, gearboxes, main shafts and generators of wind turbines, multi-modal data of equipment operation are comprehensively collected;

[0071] This arrangement method can cover the key components of wind turbines and ensure that complete operating state information can be obtained when the equipment fails;

[0072] Comprehensive perception of the operating state of wind turbines is realized, ensuring the diversity and integrity of data, and providing high-quality and all-round data support for subsequent fault diagnosis;

[0073] Through multi-point sensor arrangement and multi-modal data collection, behaviors such as mechanical vibration, acoustic anomalies, temperature changes and rotational speed fluctuations of wind turbines can be captured, comprehensively covering potential fault characteristics during the operation of wind turbines;

[0074] By arranging vibration sensors, acoustic sensors, temperature sensors and rotational speed sensors on key components such as the blades, gearboxes, main shafts and generators of wind turbines, vibration data, acoustic data, temperature data and rotational speed data are collected;

[0075] These data cover multi-modal information during the operation of wind turbines, can comprehensively reflect the operating state and potential fault characteristics of the equipment. At the same time, setting appropriate sampling frequencies ensures the accuracy of different data types in terms of time and frequency;

[0076] Set the sampling frequency according to user requirements to collect fault diagnosis data of vibration frequency data, acoustic frequency data, temperature data and rotational speed data;

[0077] Multi-modal sensor arrangement covers the operating states of key components of wind turbines;

[0078] At the same time, the sampling frequency can be flexibly set according to user requirements, which can adapt to the monitoring requirements under different working conditions;

[0079] This step can not only obtain multi-modal data of wind turbines in real time, but also meet the data collection requirements under different working conditions by flexibly adjusting the sampling frequency;

[0080] Through the arrangement of multi-type sensors, the efficient acquisition of multi-modal operation data of wind turbines is realized, comprehensively covering the key status information during equipment operation, laying the foundation for subsequent data analysis, and ultimately improving the accuracy and comprehensiveness of fault diagnosis.

[0081] S2. Transmit the fault diagnosis data to the cloud big data platform for preprocessing;

[0082] Use the MQTT protocol to transmit the fault diagnosis data to the big data platform;

[0083] The cloud big data platform realizes the efficient transmission and preprocessing of data, improving the data quality;

[0084] Use the MQTT protocol to transmit the collected multi-modal data to the cloud big data platform, and perform standardization processing on the original data through multi-step preprocessing including normalization, filtering, outlier detection, and dynamic time warping;

[0085] Transmit data to the cloud through the MQTT communication protocol to ensure the efficiency and real-time nature of data transmission

[0086] Use the min-max normalization method to normalize the fault diagnosis data to obtain dimensionless fault diagnosis data;

[0087] Use a filtering algorithm to filter the noise of the dimensionless fault diagnosis data to obtain pure fault diagnosis data;

[0088] Use an outlier detection algorithm based on Mahalanobis distance to remove outliers from the pure fault diagnosis data to obtain fault diagnosis data with outliers removed;

[0089] Use dynamic time warping to align the time of the fault diagnosis data with outliers removed to obtain fault diagnosis data with high-precision alignment, and use the fault diagnosis data with high-precision alignment as the preprocessed fault diagnosis data;

[0090] Through a series of preprocessing operations, the quality and consistency of the data are improved, making the data more suitable for feature extraction and training of the diagnostic model;

[0091] Normalization eliminates the influence of different sensor data scales and the dimensional differences of different data types, facilitating subsequent feature extraction and model training. The filtering algorithm removes the interference of environmental noise on the data and the random noise interference in the data, improving the accuracy of the data. Mahalanobis distance eliminates outliers, enhancing the reliability of the data, effectively removing outliers and reducing the impact of incorrect data on the model. Dynamic time warping solves the problem of time synchronization between sensor data, ensuring the time synchronization of multi-modal data and solving the possible time offset problem during the acquisition of different sensor data;

[0092] Through the above operations, high-precision, dimensionless, pure and time-aligned fault diagnosis data are obtained, eliminating the possible interference factors in the data, improving the data quality, and laying a solid foundation for subsequent feature extraction and model analysis;

[0093] This step solves the problems of inconsistent dimensions, noise interference, outliers and time misalignment existing in the original data, making the data more suitable for subsequent feature extraction and modeling requirements;

[0094] Through the unified preprocessing of the cloud big data platform, the efficient cleaning and standardization of multi-modal data are realized, eliminating data noise and outliers, improving the quality and reliability of the data, and providing accurate basic data for subsequent feature extraction and model construction.

[0095] S3. Extract features from the preprocessed fault diagnosis data to obtain time-domain features and frequency-domain features;

[0096] Calculate the mean, standard deviation and maximum value of the preprocessed fault diagnosis data as the time-domain feature f1;

[0097] Perform a fast Fourier transform (FFT) on the preprocessed fault diagnosis data to extract the main frequency component and spectral energy as the frequency-domain feature f2.

[0098] The combination of time-domain features and frequency-domain features can comprehensively describe the overall characteristics of the signal, thus reflecting the operating states of different components of the wind turbine;

[0099] Use the fast Fourier transform (FFT) to transform the processed fault diagnosis data into the frequency domain to obtain the spectral data X(θ) in complex form, where θ represents the frequency component and X represents the frequency-domain representation vector of the signal;

[0100] Calculate the amplitude spectrum |X(θ)| of the spectrum, and the expression is:

[0101]

[0102] where Re(X) represents the real part of the Fourier transform result and Im(X) represents the imaginary part of the Fourier transform result;

[0103] The main frequency component is the frequency component with the largest amplitude in the amplitude spectrum. Traverse the spectrum amplitude to find the frequency position with the largest amplitude, which is the main frequency component;

[0104] The spectrum energy q is the sum of the powers of the frequency-domain signals, used to measure the energy distribution of the signal in the frequency domain. The expression is:

[0105]

[0106] where represents the square of the spectrum amplitude, that is, the power of the frequency component, represents the Nyquist frequency, that is, the highest effective frequency of the signal;

[0107] Extract time-domain features by calculating the mean, standard deviation, and maximum value of the data, and use the fast Fourier transform FFT to extract the frequency-domain features of the main frequency component and spectrum energy to comprehensively capture the statistical and spectrum characteristics of the data;

[0108] The time-domain features reflect the overall trend and statistical characteristics of the sensor signal, while the frequency-domain features reveal the frequency change pattern and energy distribution of the signal. The combination of the two can comprehensively describe the characteristics of the fault signal;

[0109] It improves the feature expression ability of fault diagnosis, provides richer and more comprehensive input features for the diagnosis model, and enhances the accuracy of diagnosis;

[0110] Combining the extraction of time-domain and frequency-domain features can describe the operating state of the wind turbine from multiple dimensions;

[0111] Through the extraction of time-domain and frequency-domain features, the comprehensive mining of the deep characteristics of the data is realized, which not only improves the discrimination ability of the features, but also enhances the characterization ability of complex fault patterns, providing rich information for subsequent feature fusion and fault classification.

[0112] S4. Adopt a dynamic weight assignment self-attention mechanism to deeply fuse the time-domain features and frequency-domain features into a fault feature vector;

[0113] Calculate the weights of the time-domain features and frequency-domain features through the self-attention mechanism. The expression is:

[0114] α1 = softmax(W1·f1);

[0115] α2 = softmax(W2·f2);

[0116] Among them, α1 represents the weight of the time-domain feature f1, α2 represents the weight of the frequency-domain feature f2, softmax represents the probability distribution function, W1 represents the weight matrix of the time-domain feature f1, which is obtained by learning from historical data, and W2 represents the weight matrix of the frequency-domain feature f2, which is obtained by learning from historical data;

[0117] The introduction of the self-attention mechanism enables the fusion of time-domain features and frequency-domain features not to be a simple linear superposition, but to dynamically adjust the weights according to the importance of the features;

[0118] By dynamically adjusting the feature weights, the flexibility and expression ability of feature fusion are significantly improved, providing higher accuracy and robustness for fault diagnosis under complex working conditions;

[0119] This weight calculation method based on softmax can adaptively adjust the contribution degree of features in different operating states, avoiding the problem of diagnostic failure caused by fixed feature weights in traditional methods;

[0120] Based on the time-domain feature and the frequency-domain feature, a weighted fusion method is used for fusion to obtain a fault feature vector, and the expression is:

[0121] F = α1·f1 + α2·f2;

[0122] Among them, F represents the fault feature vector;

[0123] The fault feature vector is the main input of the wind turbine fault diagnosis model, which condenses the key information of multi-modal data;

[0124] By generating the fault feature vector, the high integration of multi-modal features is realized, the model input is simplified, and the diagnostic efficiency is improved;

[0125] Through feature extraction, the information concentration of multi-modal data is realized, providing accurate feature data for subsequent feature fusion and model construction;

[0126] Feature fusion unifies the expression forms of time-domain features and frequency-domain features, effectively enhancing the discrimination ability of features, and providing a basis for subsequent generation of fault feature vectors;

[0127] Through feature fusion, the unified expression of time-domain and frequency-domain information is realized, improving the representativeness of features and the input quality of the diagnostic model;

[0128] Through fusion, the fault feature vector combines the advantages of time-domain features and frequency-domain features, not only completely retains the key information of multi-modal data, but also simplifies the model input structure, improving the generalization ability of the model;

[0129] The fault feature vector F is generated by weighted fusion, which integrates the time domain characteristics and frequency domain characteristics of multi-modal data to characterize the operating status of the wind turbine in a unified way;

[0130] The step of generating feature vectors fully considers the weight of each feature, thus avoiding interference between features during the fusion process;

[0131] Through feature fusion, a unified fault feature vector is generated. This feature vector reduces the redundancy between features while maintaining the integrity of time domain features and frequency domain characteristics, improves the overall expression ability of the feature set, and provides better input for subsequent model training and inference;

[0132] By generating fault feature vectors, the compact expression of multimodal features is achieved, the input dimension of the model is simplified, the effectiveness of feature input is improved, and the foundation is laid for improving fault diagnosis performance.

[0133] S5. A deep neural network (DNN) with hierarchical feature activation combined with a long short-term memory (LSTM) network is used to build a fault diagnosis model, which performs layer-by-layer nonlinear mapping on the fault feature vector and finally outputs the diagnosis result.

[0134] Input the fault feature vector and pass it through the nonlinear mapping of the first layer of the network. The expression is:

[0135]

[0136] Among them, s1 represents the output of the first layer of neural network, represents the Swish activation function, U1 represents the weight matrix of the first layer of the neural network, which is generated by Gaussian distribution, and b1 represents the bias vector of the first layer of the neural network;

[0137] The output s1 of the first layer of neural network is input into the second layer to continue nonlinear mapping. The expression is:

[0138]

[0139] Among them, s2 represents the output of the second layer of neural network, U2 represents the weight matrix of the second layer of neural network, and b2 represents the bias vector of the second layer of neural network;

[0140] The output s2 after mapping by the second layer of neural network is input into the long short-term memory network LSTM to capture the dynamic correlation of features in the time dimension. The expression is:

[0141] h t ,c t =LSTM(s2,h t-1 ,c t-1 );

[0142] Among them, t represents the time index, and h t represents the hidden state at time t, and c t represents the memory cell state at time t. LSTM represents the long short-term memory network LSTM, and h t-1 represents the hidden state at the previous moment, and c t-1 represents the memory cell state at the previous moment;

[0143] Finally, the LSTM processes the output s2 of the second-layer neural network at the last moment T and outputs the hidden state h T at the last time as the comprehensive feature representation, where T represents the last moment;

[0144] By capturing the dynamic correlation of fault features in the time dimension, LSTM effectively makes up for the deficiencies of traditional neural networks in processing time series data;

[0145] The deep learning model combines multi-modal features to achieve efficient diagnosis and classification of wind turbine faults;

[0146] The combination of DNN and LSTM not only realizes the deep learning of the feature space but also captures the dynamic changes in the time dimension, providing accurate model support for the fault diagnosis of complex dynamic systems

[0147] Through the deep learning algorithm, deep modeling of the fault feature vector is achieved, improving the accuracy and robustness of fault diagnosis;

[0148] The deep neural network can mine the implicit rules in the fault feature vector and accurately classify the fault types of wind turbines;

[0149] The classification modeling of the fault feature vector is realized through the deep neural network DNN. The first layer and the second layer of the neural network perform non-linear mappings respectively. The use of the Swish activation function improves the expression ability of the network. The weight matrix is initialized by the Gaussian distribution, which can accelerate the convergence process of the model;

[0150] Through multi-layer non-linear mappings, the deep neural network can learn the complex relationships between features, extract high-order features, and thus achieve accurate classification of fault patterns;

[0151] Through the multi-layer non-linear mappings of the deep neural network, complex pattern learning of high-dimensional feature vectors is realized, the recognition ability of fault types is enhanced, the processing ability of the model for complex non-linear relationships is improved, and finally the accuracy and robustness of fault diagnosis are improved;

[0152] The expression of the final output layer is:

[0153] P = Softmax(U3·h T + b3);

[0154] Wherein, Softmax represents the activation function Softmax, U3 represents the weight matrix of the output layer, b3 represents the bias term of the output layer, and P represents the predicted probability of the fault category;

[0155] Through the Softmax function of the output layer, the output of the last layer of the deep neural network is mapped into a probability distribution, thereby generating predicted probabilities for each fault category;

[0156] The Softmax function normalizes the probability distribution of the output, ensures that the sum of all category probabilities is 1, and at the same time provides clear probability values for each fault type, facilitating user understanding and decision-making;

[0157] The predicted probability P of the fault category = {p1, p2, …, p a}, which is the predicted probability of each fault category, where a represents the number of fault categories;

[0158] Through the probabilistic output form, accurate diagnosis and classification of wind turbine faults are achieved, providing a reliable decision-making basis for users;

[0159] The diagnosis result is presented in the form of probability, facilitating the identification and judgment of the specific fault type of the wind turbine;

[0160] The predicted probability P of the fault category represents the possibility of each fault category, and can intuitively display the credibility of the diagnosis result;

[0161] Through the output of the Softmax function, the predicted probability of each fault type is intuitively provided for users, facilitating users' intuitive understanding of the diagnosis result and subsequent decision-making;

[0162] At the same time, this step enhances the interpretability of the model prediction through the probability output method.

[0163] S6. Update the parameters of the fault diagnosis model according to the real-time data;

[0164] Define a loss function for updating the parameters of the fault diagnosis model, and the expression is:

[0165]

[0166] Wherein, L t represents the real-time loss at time point t, j represents the index of the number of fault categories, represents the true label of the jth fault at time point t, represents the occurrence probability of the jth fault at time point t, and log represents the logarithmic function;

[0167] Weight matrix U d and bias vector b d have the following update formulas:

[0168]

[0169] where d represents the index of the number of hidden layers, U d represents the weight matrix of the d-th hidden layer, b d represents the bias vector of the d-th hidden layer, U ′ d represents the updated weight matrix of the d-th hidden layer, b ′ d represents the updated bias vector of the d-th hidden layer, η represents the learning rate, represents the gradient of the loss function with respect to the weight matrix U d and represents the gradient of the loss function with respect to the bias vector b d and

[0170] By defining a real-time loss function and using the gradient descent method to update the model parameters, the present invention realizes the online update ability of the fault diagnosis model;

[0171] This dynamic learning mechanism enables the diagnostic model to adapt to the changes in the operating environment of the wind turbine and maintain the stability of the diagnostic performance at all times;

[0172] Compared with the static model, the real-time update mechanism significantly improves the dynamic adaptability of the diagnostic system and provides technical support for the long-term stable operation of the wind turbine.

[0173] This embodiment also provides a computer device applicable to the case of the fault diagnosis method of the wind turbine based on big data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the fault diagnosis method of the wind turbine based on big data as proposed in the above embodiment.

[0174] The computer device may be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, carrier networks, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads provided on the housing of the computer device, or can also be an external keyboard, touchpad, or mouse, etc.

[0175] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for fault diagnosis of a wind turbine based on big data as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0176] In summary, through the combination of comprehensive acquisition of multi-source data, high-quality preprocessing, multi-dimensional feature fusion, intelligent diagnostic algorithms, and dynamic learning mechanisms, the present invention significantly improves the accuracy, real-time performance, and robustness of wind turbine fault diagnosis, solves the problems of single data dependence, insufficient feature extraction, and model staticization commonly existing in traditional methods, is applicable to intelligent diagnosis of wind turbines under complex working conditions, and provides an innovative technical solution for the safe and efficient operation of wind farms.

[0177] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A fault diagnosis method for a wind turbine based on big data, characterized in that: include, Collect fault diagnosis data of wind turbines through sensors; Transmit fault diagnosis data to the cloud big data platform for preprocessing; Extract features from the preprocessed fault diagnosis data to obtain time domain features and frequency domain features; A dynamic weight allocation self-attention mechanism is used to deeply fuse time domain features and frequency domain features into a fault feature vector; A deep neural network (DNN) with hierarchical feature activation combined with a long short-term memory (LSTM) network is used to build a fault diagnosis model, which performs layer-by-layer nonlinear mapping on the fault feature vector and finally outputs the diagnosis result. Update the parameters of the fault diagnosis model based on real-time data.

2. The wind turbine fault diagnosis method based on big data according to claim 1, characterized in that: The specific steps of collecting the fault diagnosis data of the wind turbine generator by the sensor are as follows: Arrange vibration sensors, acoustic sensors, temperature sensors and speed sensors on the blades, gearbox, main shaft and generator of the wind turbine; The sampling frequency is set according to user needs to collect fault diagnosis data such as vibration frequency data, acoustic frequency data, temperature data and rotation speed data.

3. The fault diagnosis method for wind turbines based on big data according to claim 2, characterized in that: The specific steps of transmitting the fault diagnosis data to the cloud big data platform for preprocessing are as follows: Use MQTT protocol to transmit fault diagnosis data to the big data platform; The minimum-maximum normalization method is used to normalize the fault diagnosis data to obtain dimensionless fault diagnosis data; The filtering algorithm is used to filter the noise of dimensionless fault diagnosis data to obtain pure fault diagnosis data; The outlier detection algorithm based on Mahalanobis distance is used to remove outliers from the pure fault diagnosis data to obtain fault diagnosis data without outliers. Dynamic time warping is used to time align the fault diagnosis data with outliers removed to obtain high-precision aligned fault diagnosis data, and the high-precision aligned fault diagnosis data is used as preprocessed fault diagnosis data.

4. The wind turbine fault diagnosis method based on big data as claimed in claim 3, characterized in that: The feature extraction of the preprocessed fault diagnosis data to obtain time domain features and frequency domain features is specifically performed as follows: Calculate the mean, standard deviation and maximum value of the preprocessed fault diagnosis data as the time domain feature f1; Perform fast Fourier transform (FFT) on the preprocessed fault diagnosis data to extract the main frequency component and spectrum energy as the frequency domain feature f2.

5. The fault diagnosis method for wind turbines based on big data according to claim 4, characterized in that: The dynamic weight allocation self-attention mechanism is used to deeply fuse the time domain features and frequency domain features into a fault feature vector. The specific steps are as follows: The weights of time domain features and frequency domain features are calculated through the self-attention mechanism, and the expression is: α1=softmax(W1·f1); α2=softmax(W2·f2); Among them, α1 represents the weight of the time domain feature f1, α2 represents the weight of the frequency domain feature f2, softmax represents the probability distribution function, W1 represents the weight matrix of the time domain feature f1, which is learned through historical data, and W2 represents the weight matrix of the frequency domain feature f2, which is learned through historical data; Based on the time domain features and frequency domain features, the weighted fusion method is used to fuse and obtain the fault feature vector, which is expressed as: F = α1·f1+α2·f2; Where F represents the fault feature vector.

6. The wind turbine fault diagnosis method based on big data according to claim 5, characterized in that: The fault diagnosis model is constructed by using a deep neural network DNN with hierarchical feature activation combined with a long short-term memory network LSTM, and the fault feature vector is nonlinearly mapped layer by layer. The specific steps are as follows: Input the fault feature vector and pass it through the nonlinear mapping of the first layer of the network. The expression is: Among them, s1 represents the output of the first layer of neural network, represents the Swish activation function, U1 represents the weight matrix of the first layer of the neural network, which is generated by Gaussian distribution, and b1 represents the bias vector of the first layer of the neural network; The output s1 of the first layer of neural network is input into the second layer to continue nonlinear mapping. The expression is: Among them, s2 represents the output of the second layer of neural network, U2 represents the weight matrix of the second layer of neural network, and b2 represents the bias vector of the second layer of neural network; The output s2 after mapping by the second layer of neural network is input into the long short-term memory network LSTM to capture the dynamic correlation of features in the time dimension. The expression is: h t ,c t =LSTM(s2,h t-1 ,c t-1 ); Among them, t represents the time index, h t represents the hidden state at time t, c t represents the state of the memory unit at time t, LSTM represents the long short-term memory network LSTM, and h t-1 represents the hidden state at the previous moment, c t-1 Indicates the state of the memory unit at the previous moment; Finally, the long short-term memory network LSTM processes the output s2 of the second layer neural network at the last time T and outputs the hidden state h at the last time T , as a comprehensive feature representation, where T represents the last moment.

7. The wind turbine fault diagnosis method based on big data according to claim 6, characterized in that: The final output of the diagnosis result comprises the following specific steps: The final output layer expression is: P=Softmax(U3·h T +b3); Among them, Softmax represents the activation function Softmax, U3 represents the weight matrix of the output layer, b3 represents the bias term of the output layer, and P represents the predicted probability of the fault category; The predicted probability of the fault category P = {p1, p2, ..., p a }, is the predicted probability of each fault category, where a represents the number of fault categories.

8. The fault diagnosis method for wind turbines based on big data according to claim 7, characterized in that: The specific steps of updating the parameters of the fault diagnosis model according to the real-time data are as follows: Define a loss function to update the parameters of the fault diagnosis model, the expression is: Among them, L t represents the real-time loss at time point t, j represents the index of the number of fault categories, represents the true label of the j-th fault at time point t, represents the probability of occurrence of the jth fault at time point t, and log represents the logarithmic function; Weight matrix U d and the bias vector b d The update formula is: Among them, d represents the index of the hidden layer, U d represents the weight matrix of the dth hidden layer, b d represents the bias vector of the dth hidden layer, U ′ d represents the updated weight matrix of the dth hidden layer, b ′ d represents the updated bias vector of the dth hidden layer, η represents the learning rate, Represents the loss function on the weight matrix U d The gradient of Represents the loss function for the bias vector b d gradient.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the fault diagnosis method for a wind turbine based on big data according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the wind turbine fault diagnosis method based on big data described in any one of claims 1 to 8 are implemented.

Images