A wind turbine fault intelligent detection system and method

Abstract

The present invention discloses an intelligent detection system and method for wind turbine faults, which relates to the field of deep learning technology, including collecting multidimensional data and performing preprocessing; extracting multidimensional features from the preprocessed multidimensional data, and fusing the multidimensional features through graph neural networks and short-time Fourier transform to generate time-frequency domain signals; extracting key features from the time-frequency domain signals, and detecting the fault characteristics of the wind turbine operating status through a support vector machine model; constructing a fusion neural network model based on the fault characteristics, generating a fault prediction, outputting a diagnosis report, and proposing maintenance suggestions; the present invention combines a convolutional neural network and a long short-term memory network through a fusion neural network model to extract spatial features and time series data, thereby realizing multi-dimensional characterization of wind turbine fault characteristics.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a wind turbine fault intelligent detection system and method. Background Art

[0002] In recent years, with the rapid development of wind power technology, the operational reliability and maintenance costs of wind turbines have become the focus of industry attention. Traditional fault detection methods mainly rely on regular manual inspections and data analysis of single sensors. These methods are not only time-consuming and labor-intensive, but also unable to accurately capture the fault characteristics of wind turbines in real time. With the improvement of sensor technology and data processing capabilities, multi-sensor data fusion technology has been widely used. It can comprehensively analyze multi-dimensional data such as temperature, vibration, and current, providing a more comprehensive perspective for fault detection. However, there are still major challenges in the effective fusion of multi-source data and the extraction of time-frequency domain features in existing technologies.

[0003] Existing wind turbine fault detection technology has the following main shortcomings: First, the data from a single sensor is easily affected by environmental noise and changes in operating conditions, making it difficult to ensure detection accuracy. Second, traditional data processing and feature extraction methods often ignore the correlation between different data sources, resulting in insufficient information utilization. Finally, most existing fault detection models lack real-time and intelligence, and are unable to quickly respond to and predict potential faults. Summary of the Invention

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

[0005] Therefore, the present invention provides a wind turbine fault intelligent detection method to solve the problem of accurate prediction and real-time diagnosis of faults.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an intelligent detection method for wind turbine faults, which includes collecting multidimensional data and performing preprocessing; extracting multidimensional features from the preprocessed multidimensional data, and fusing the multidimensional features through graph neural networks and short-time Fourier transform to generate time-frequency domain signals; extracting key features from the time-frequency domain signals, and detecting the fault characteristics of the wind turbine operating status through a support vector machine model; based on the fault characteristics, constructing a fusion neural network model, generating a fault prediction, and outputting a diagnostic report and making maintenance recommendations.

[0008] As a preferred solution of the intelligent detection method for wind turbine faults described in the present invention, the multi-dimensional data includes temperature data, vibration data and current data.

[0009] As a preferred solution of the intelligent detection method for wind turbine faults described in the present invention, the preprocessing includes denoising, filtering and outlier removal.

[0010] As a preferred solution of the wind turbine fault intelligent detection method described in the present invention, the multi-dimensional features include:

[0011] Extract average temperature, temperature change rate, and temperature peak value from temperature data as temperature features;

[0012] Extract vibration amplitude, vibration frequency, and vibration energy from vibration data as vibration features;

[0013] The average current, current fluctuation and current peak are extracted from the current data as current features.

[0014] As a preferred solution of the wind turbine fault intelligent detection method described in the present invention, wherein: the multi-dimensional features are fused through graph neural network and short-time Fourier transform to generate time-frequency domain signals, the specific steps are as follows:

[0015] The temperature characteristics, vibration characteristics and current characteristics are used as nodes of the graph;

[0016] The node features in the graph are fused through the graph neural network to generate a comprehensive feature set, which is expressed as:

[0017]

[0018] Among them, hv (k) is the integrated feature set after fusion, h is the feature vector, k is the kth iteration in the graph neural network, v is the node, σ is the activation function, N(v) is the set function, is the feature vector of the neighbor node u in the k-1th iteration of the graph neural network, where u is the neighbor node of node v;

[0019] Based on the fused comprehensive feature set, short-time Fourier transform is performed to generate time-frequency domain signals, which are expressed as:

[0020]

[0021] Among them, Z(t,ω) is the time-frequency domain signal, t is the time, ω is the angular frequency, A(θ) is the time domain signal, θ is the integral variable, w(θ-t) is the window function, e -jωθ is a complex exponential function, j is an imaginary unit, is the integral symbol, ∞ is positive infinity, and -∞ is negative infinity.

[0022] As a preferred solution of the wind turbine fault intelligent detection method described in the present invention, wherein: the key features are extracted and the fault characteristics of the wind turbine operating state are detected by a support vector machine model, specifically the following steps:

[0023] Extract spectral features, time domain features and time-frequency features from time-frequency domain signals;

[0024] Through multi-scale analysis, the time-varying characteristics of different frequency components in the time-frequency domain signals are captured, and the spikes and abnormal changes in the time-frequency domain signals are identified;

[0025] The dimensionality reduction of spectrum features, time domain features, time-frequency features and time-varying features is performed through principal component analysis;

[0026] Use local linear embedding to find the neighbors of data points in the spectral features, time domain features, time-frequency features, and time-varying features after dimension reduction, and determine the local neighborhood;

[0027] In the local neighborhood of each data point, the linear reconstruction weight is calculated to minimize the embedding reconstruction error;

[0028] Find low-dimensional embedding coordinates that minimize the embedding-reconstruction error and capture nonlinear structural features;

[0029] The spectrum features, time domain features, time-frequency features and time-varying features after principal component analysis and local linear embedding are input into multiple classifiers, and the classification results are fused by voting to generate a comprehensive feature vector;

[0030] Use the pre-trained support vector machine model to train and classify the comprehensive feature vector to identify the fault type and severity of the wind turbine;

[0031] Cross-validate the support vector machine model to evaluate performance;

[0032] Continuously optimize feature extraction and classification based on verification results;

[0033] The optimized support vector machine model is applied to real-time monitoring to detect fault characteristics of wind turbine operating status.

[0034] As a preferred solution of the wind turbine fault intelligent detection method described in the present invention, the steps of constructing a fusion neural network model, generating fault prediction, outputting a diagnosis report, and proposing maintenance suggestions are as follows:

[0035] Define the fusion neural network model input as fault characteristics and output as fault type and prediction results;

[0036] Extract spatial features from fault features through convolutional layers;

[0037] Use the adaptive pooling layer to reduce the dimensionality of the spatial features extracted by the convolution layer and dynamically adjust the pooling parameters according to the input fault characteristics;

[0038] Capture time series data in fault characteristics through the long short-term memory network layer to obtain long-term dependencies and dynamic features in time series data;

[0039] The spatial features and dynamic features are integrated to generate the comprehensive fault feature vector, which is expressed as:

[0040] P t =tanh(CNN(x t )+LSTM(x t-1 )+b);

[0041] Among them, P t is the fault comprehensive feature vector, tanh is the hyperbolic tangent activation function, CNN(x t ) is the output of the convolutional layer, x t is the input feature at time t, LSTM(x t-1 ) is the output of the long short-term memory network layer, x t-1 is the input feature at time t-1, b is the hyperbolic tangent activation function;

[0042] The comprehensive fault feature vector is mapped to a multi-dimensional space through a fully connected layer to generate a fault prediction;

[0043] The output layer outputs the fault type and prediction results based on the fault prediction;

[0044] Generate a fault diagnosis report with fault type, fault location, fault severity, timestamp and data chart based on the fault type and prediction results;

[0045] Through fault diagnosis reports, maintenance suggestions are automatically generated and maintenance measures are taken.

[0046] In the second aspect, the present invention provides an intelligent detection system for wind turbine faults, including an acquisition module, a fusion module, a detection module and a prediction module; the acquisition module is used to acquire multidimensional data and perform preprocessing; the fusion module is used to extract multidimensional features from the preprocessed multidimensional data, and fuse the multidimensional features through graph neural networks and short-time Fourier transform to generate time-frequency domain signals; the detection module is used to extract key features of time-frequency domain signals based on the time-frequency domain signals, and detect the fault characteristics of the wind turbine operating status through a support vector machine model; the prediction module is used to construct a fusion neural network model based on the fault characteristics, generate a fault prediction, output a diagnostic report, and make maintenance recommendations.

[0047] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the intelligent detection method for wind turbine faults as described in the first aspect of the present invention is implemented.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the intelligent detection method for wind turbine faults as described in the first aspect of the present invention is implemented.

[0049] The beneficial effects of the present invention are as follows: the present invention extracts spatial features and time series data through a fusion neural network model, combines convolutional neural networks and long short-term memory networks, realizes multi-dimensional characterization of wind turbine fault characteristics, and performs dimensionality reduction in the pooling layer to improve computational efficiency. After the features are fused, a comprehensive fault feature vector is generated, which is mapped to multi-dimensional space through a fully connected layer to generate accurate fault predictions. Finally, the output layer generates diagnostic reports and maintenance recommendations based on the fault predictions, which not only improves the accuracy and real-time performance of fault detection, but also significantly improves the overall reliability and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 This is a flow chart of a wind turbine fault intelligent detection method in Example 1;

[0052] Figure 2 This is a module diagram of an intelligent wind turbine fault detection system in Example 1. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0056] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for intelligent detection of wind turbine faults, comprising the following steps:

[0057] S1. Collect multidimensional data and perform preprocessing.

[0058] Furthermore, a variety of sensors are installed at key locations of the wind turbine, including temperature sensors, vibration sensors, and current sensors;

[0059] These sensors collect temperature data, vibration data and data of wind turbines in real time during operation, and these data are transmitted to the data processing center through wireless communication networks;

[0060] Temperature data, which monitors temperature changes in key wind turbine components, is crucial for detecting problems such as overheating and temperature anomalies;

[0061] Vibration data, which monitors the vibrations generated by wind turbines during operation. Vibration data can help detect mechanical failures, wear, and imbalances;

[0062] Current data, which monitors current changes in wind turbines, is important for detecting problems such as electrical faults, short circuits, and current overloads.

[0063] S1.2, pre-processing, specifically,

[0064] Apply a low-pass filter or a band-pass filter to eliminate high-frequency noise;

[0065] Use sliding average or median filtering to smooth the data and reduce the impact of random noise;

[0066] Perform Fourier transform on the time domain signal to convert it into frequency domain signal;

[0067] Apply filters in the frequency domain to remove unwanted frequency components, and then restore the time domain signal through inverse Fourier transform;

[0068] Calculate the mean and standard deviation of the data and detect outliers using the 3σ principle;

[0069] Detect outliers using algorithms such as the Isolation Forest algorithm or the Local Outlier Factor;

[0070] Remove the detected outliers and use interpolation methods (such as linear interpolation and spline interpolation) to fill the positions of the outliers;

[0071] Scale the data to the range [0, 1];

[0072] Transform the data into a standard normal distribution with a mean of 0 and a standard deviation of 1;

[0073] It should be noted that by installing a variety of sensors, temperature, vibration and current data of wind turbines are collected in real time and transmitted to the data processing center. Temperature data helps detect overheating and temperature anomalies, vibration data is used to identify mechanical failures and wear, and current data monitors electrical failures. The preprocessing steps include denoising, smoothing, outlier detection and interpolation, and the data is scaled and standardized. In this way, the accuracy and reliability of the input data are ensured through high-quality data processing.

[0074] S2. Extract multidimensional features from the preprocessed multidimensional data, fuse the multidimensional features through graph neural network and short-time Fourier transform to generate time-frequency domain signals.

[0075] Furthermore, the average temperature, temperature change rate, and temperature peak value are extracted from the temperature data as temperature features;

[0076] The average temperature reflects the overall temperature level of the wind turbine over a period of time. It can help detect whether there is overheating and thus prevent temperature-related failures.

[0077] The temperature change rate describes how quickly the temperature changes over time. By monitoring the temperature change rate, abnormal temperature increases or decreases can be detected in a timely manner, providing early warning of possible failures.

[0078] Temperature peak records the highest temperature during the monitoring period. Temperature peak can provide information about the operating status of the equipment under extreme conditions and help identify potential failures due to overheating.

[0079] Extract vibration amplitude, vibration frequency, and vibration energy from vibration data as vibration features;

[0080] The vibration amplitude indicates the intensity of the vibration. A larger vibration amplitude usually indicates looseness of the mechanical structure or abnormal wear of the components. It is an important indicator of mechanical failure.

[0081] Vibration frequency reflects the periodicity or frequency characteristics of vibration. Vibration frequency analysis helps identify specific mechanical components or failure modes, such as bearing damage, gear wear, etc.

[0082] Vibration energy refers to the energy level of the vibration signal. By monitoring vibration energy, the health status of the equipment under different operating conditions can be identified, and abnormal operating conditions can be detected.

[0083] Extracting average current, current fluctuation, and current peak from current data as current features;

[0084] Average current indicates the average current consumed by the device during operation. Abnormal average current values ​​may indicate a malfunction of the electrical components, such as a short circuit or overload.

[0085] Current fluctuation describes the change of current over time. Significant current fluctuation may indicate electrical instability or interference and is an important indicator for fault warning.

[0086] The current peak value records the maximum current value during the monitoring period. The current peak value helps to detect the response of the equipment under load changes and identify possible electrical faults;

[0087] The temperature feature, vibration feature, and current feature are taken as nodes of the graph. For example, the temperature feature is one node, the vibration feature is another node, and the current feature is the third node.

[0088] In graph neural networks, edges between nodes represent the relationship between sensor data. If the data from two sensors are highly correlated in time (e.g., the change patterns of temperature and vibration data are similar), an edge is added between them.

[0089] The node features in the graph are fused through the graph neural network to generate a comprehensive feature set, which is expressed as:

[0090]

[0091] Among them, hv (k) is the fused comprehensive feature set, representing the feature vector of node v in the kth iteration, such as a feature of the temperature sensor. h is the feature vector, which is used to represent the state and attributes of the node. In the graph neural network, the feature vector contains a set of numerical values ​​about the node, which can describe some important information of the node. For example, for the temperature sensor node, the feature vector may include the average, maximum, and minimum values ​​of the temperature. k is the kth iteration in the graph neural network, v is the node, a data point in the graph, such as the feature node of the temperature sensor, σ is the activation function, such as ReLU, Sigmoid, or Tanh, which is used to introduce nonlinearity to the model. |N(v)| is the inverse of the number of neighbors of node v, which is used to calculate the average value. N(v) is a set function, which represents the set of all neighbor nodes of node v. This set contains all nodes that are directly connected to node v. For example, if node v represents a temperature sensor, then N(v) may contain vibration sensors and current sensors that are directly connected to the temperature sensor. is the feature vector of the neighbor node u in the graph neural network in the k-1th iteration. u is the neighbor node of node v. It is a specific neighbor node in N(v). u is an element in the set N(v), that is, a specific neighbor node. In the aggregation operation, we will traverse all neighbor nodes u in N(v) and use their feature vectors for calculation;

[0092] Through this formula, the feature of node v is the average of the features of its neighboring nodes and is processed by the activation function. This method can effectively aggregate the data features from different sensors to obtain the fused feature representation;

[0093] Based on the fused comprehensive feature set, short-time Fourier transform is performed to generate time-frequency domain signals, which are expressed as:

[0094]

[0095] Among them, Z(t,ω) is the time-frequency domain signal, which represents the signal value at time t and frequency ω. This signal is the result of short-time Fourier transform processing, which combines time and frequency information. t is the time, which represents the center moment of the analysis window, that is, the specific time point we are interested in. ω is the angular frequency, which represents the frequency component of the signal. A(θ) is the time domain signal, which represents the original signal at time θ. θ is the integral variable, which represents the time of the original signal. During the integration process, it traverses the time range of the entire signal. w(θ-t) is the window function, which focuses on a small section of the signal near time t. Common windows include rectangular windows, Hanning windows, etc. -jωθ is a complex exponential function, representing the frequency component of Fourier transform, j is the imaginary unit, ω is the angular frequency, is the integral symbol, which represents the integral operation of time θ, ranging from negative infinity to positive infinity, ∞ is positive infinity, and -∞ is negative infinity;

[0096] This formula achieves efficient processing of wind turbine signals and accurate fault detection through time-frequency analysis, local characteristic analysis, enhanced signal recognition, and real-time monitoring, thereby improving operational reliability and maintenance efficiency.

[0097] It should be noted that by extracting features from temperature, vibration and current data and fusing them using graph neural networks and short-time Fourier transforms, time-frequency domain signals are generated, and different types of features are integrated together to reflect the overall operating status of the wind turbine. The graph neural network fuses multi-sensor data to extract comprehensive features, while the short-time Fourier transform provides time-frequency analysis of the signal. This comprehensive method improves the accuracy and real-time performance of fault detection, can effectively identify potential faults and provide accurate fault diagnosis, and ultimately improves the operating reliability and maintenance efficiency of wind turbines.

[0098] S3. Extract key features from time-frequency domain signals and detect fault characteristics of wind turbine operating status through support vector machine model.

[0099] Furthermore, spectrum features, time domain features and time-frequency features are extracted from time-frequency domain signals;

[0100] Spectral features extract the energy distribution within a specific frequency range from time-frequency domain signals, such as power spectral density (PSD);

[0101] Time domain features are to calculate the statistical characteristics of time-frequency domain signals over time, such as mean, variance, peak, skewness, kurtosis, etc.

[0102] Time-frequency features are to extract features such as instantaneous frequency, instantaneous amplitude, and time-frequency energy distribution;

[0103] The power spectral density (PSD) in the spectrum feature is calculated by calculating the spectrum energy distribution in each time period and identifying the energy peaks at specific frequencies. These peaks may correspond to specific fault modes.

[0104] The instantaneous frequency and instantaneous amplitude in the time-frequency characteristics are extracted using methods such as Hilbert-Huang transform (HHT). These features can reflect the sudden change and non-stationary characteristics of the signal.

[0105] Through multi-scale analysis, the time-varying characteristics of different frequency components in the time-frequency domain signals are captured, and the spikes and abnormal changes in the time-frequency domain signals are identified;

[0106] The spectral features, time domain features, time-frequency features, and time-varying features are reduced in dimension through principal component analysis to extract the main components of the features (including power spectral density in spectral features, mean and variance in time domain features, instantaneous frequency and instantaneous amplitude in time-frequency features, and spikes and abnormal changes in time-varying features).

[0107] Use local linear embedding to find the neighbors of data points in the spectral features, time domain features, time-frequency features, and time-varying features after dimension reduction, and determine the local neighborhood;

[0108] In the local neighborhood of each data point, the linear reconstruction weight is calculated to minimize the embedding reconstruction error; the data point can be represented by a linear combination of its neighbors to minimize the reconstruction error. The expression is:

[0109]

[0110] Among them, ∈ represents the reconstruction error, which represents the sum of the reconstruction errors of all data points. The reconstruction error measures the difference between the original eigenvector and the reconstructed eigenvector. i is the index variable of the data point, which is used to indicate the number of the data point. There are Q data points in total. The index i traverses each data point from 1 to Q. i is the original feature vector of the i-th data point, which contains important information describing the data point. is the weighted average eigenvector of the neighbors of data point i, which is calculated by linear reconstruction weights and represents the reconstructed eigenvector obtained by weighting the eigenvectors of the neighbors in the local neighborhood;

[0111] Find low-dimensional embedding coordinates that minimize the embedding reconstruction error, capture nonlinear structural features, and improve feature representativeness and discrimination;

[0112] Keeping the reconstruction weight unchanged, find the low-dimensional embedding coordinates that minimize the embedding error in the low-dimensional space, and obtain the new coordinates of each data point in the low-dimensional space;

[0113] By solving the eigenvalue problem, we find the low-dimensional embedding coordinates that minimize the embedding error and obtain a low-dimensional representation, that is, the new coordinates of each data point in the low-dimensional space. This step ensures that the features after dimensionality reduction still retain their important geometric and structural information in the new space.

[0114] The spectral features, time domain features, time-frequency features, and time-varying features after principal component analysis and local linear embedding are input into multiple classifiers (such as decision trees, random forests, extreme gradient boosting, etc.), and the classification results are fused by voting to generate a comprehensive feature vector;

[0115] Normalize the comprehensive feature vector to ensure that the feature values ​​are within the same scale range, which facilitates subsequent feature analysis and fault detection;

[0116] Use the pre-trained support vector machine model to train and classify the comprehensive feature vector to identify the fault type and severity of the wind turbine;

[0117] Cross-validate the support vector machine model and evaluate its performance (such as accuracy, recall, and F1 score);

[0118] Continuously optimize feature extraction and classification based on verification results to improve the accuracy of fault detection and prediction;

[0119] The optimized support vector machine model is applied to real-time monitoring to detect fault characteristics of wind turbine operation status;

[0120] When a fault is detected, an alarm signal is automatically issued to remind maintenance personnel to check and repair;

[0121] It should be noted that by extracting spectral features, time domain features and time-frequency features from time-frequency domain signals, and using multi-scale analysis, PCA and LLE methods for dimensionality reduction and processing, nonlinear structural features can be effectively captured and data complexity can be reduced. Subsequently, through training and classification of integrated classifiers and support vector machine models, the type and severity of wind turbine faults can be identified. Through cross-validation and real-time monitoring, the fault detection accuracy and prediction reliability can be improved. At the same time, an alarm signal is automatically issued to remind maintenance personnel to deal with the fault in a timely manner.

[0122] S4. Based on the fault characteristics, a fusion neural network model is constructed to generate fault predictions, output diagnostic reports, and make maintenance recommendations.

[0123] Furthermore, the input of the fusion neural network model is defined as the fault characteristics, and the output is the fault type and prediction result;

[0124] The convolution layer extracts spatial features from fault features and the convolution operation identifies local patterns between features.

[0125] By applying multiple convolution kernels (filters), the input data is convolved. Each convolution kernel slides on the input data, and the dot product between the convolution kernel and the local area of ​​the input data is calculated to extract local features.

[0126] The convolution operation generates feature maps that capture local patterns and features in the input data. For example, in fault detection, the convolution layer can identify abnormal peaks, periodic changes, and other features in the vibration signal.

[0127] The pooling layer is used to reduce the dimensionality of the spatial features extracted by the convolutional layer, reducing the complexity of the data while retaining important information;

[0128] Usually includes maximum pooling or average pooling, which reduces the dimension of features while retaining important information by applying pooling operations on feature maps;

[0129] Select the maximum value in each local area to reduce the size of the feature map. For example, a 2x2 maximum pooling operation will output the maximum of the four values.

[0130] Calculate the average value of each local area to reduce the size of the feature map;

[0131] Adaptive pooling layers are used to reduce the dimensionality of spatial features extracted by the convolutional layer. Pooling parameters are dynamically adjusted based on input fault characteristics to reduce data complexity while retaining important information, helping to prevent overfitting and improve the model's generalization capabilities.

[0132] Adaptive pooling is an advanced pooling method that can dynamically adjust the size and stride of the pooling window according to the input features to adapt to feature maps of different sizes. Through this dynamic adjustment, adaptive pooling can retain more feature information while reducing the dimension and avoid information loss.

[0133] Specifically, the adaptive pooling layer receives the output feature map of the convolutional layer and dynamically adjusts the pooling parameters (such as pooling window size, stride, etc.) according to the input fault characteristics;

[0134] For example, if the input feature map size is large, adaptive pooling may choose a larger pooling window and stride to reduce the size more effectively. If the input feature map size is small, it may choose a smaller pooling window and stride to preserve more details.

[0135] Through adaptive pooling, the size of the feature map is reduced, thereby reducing data complexity; this helps to reduce computational costs and improve the operating efficiency of the model;

[0136] Adaptive pooling can effectively reduce noise and irrelevant information in input features, which is crucial to preventing the model from overfitting during training;

[0137] By reducing the size of the feature map, adaptive pooling reduces the number of model parameters and reduces the risk of overfitting;

[0138] Adaptive pooling dynamically adjusts pooling parameters to ensure that important features are retained during dimensionality reduction, thereby improving the generalization ability of the model;

[0139] The long short-term memory network layer is used to capture the time series data in the fault characteristics, obtain the long-term dependencies and dynamic features in the time series data, and process the time correlation in the input features;

[0140] Long Short-Term Memory (LSTM) is a special type of recurrent neural network (RNN) that can capture and memorize long-term dependencies in time series data. LSTM controls the flow of information through input, forget, and output gates, solving the vanishing and exploding gradient problems found in traditional RNNs.

[0141] The Long Short-Term Memory (LSTM) layer processes input time series data and, through memory cells and gating mechanisms, captures the dynamic characteristics of data over time. For example, in wind turbine fault detection, the LSTM layer can track the long-term trends of vibration signals and identify fault modes with a long duration.

[0142] The hidden states and cell states generated by the LSTM layer of the long short-term memory network contain long-term dependencies and dynamic features in time series data, providing rich information for subsequent fault prediction;

[0143] The spatial features and dynamic features are integrated to generate the comprehensive fault feature vector, which is expressed as:

[0144] P t =tanh(CNN(x t )+LSTM(x t-1 )+b);

[0145] Among them, P t is the fault comprehensive feature vector, the hidden state vector at time t, tanh is the hyperbolic tangent activation function, which is used to introduce nonlinear characteristics, CNN(x t ) is the output of the convolutional layer, x t is the input feature at time t, LSTM(x t-1 ) is the output of the long short-term memory network layer, x t-1 is the input feature at time t-1, b is the hyperbolic tangent activation function, which is used to introduce nonlinear characteristics;

[0146] This formula combines the advantages of CNN and LSTM and introduces activation functions and bias terms to achieve a comprehensive and accurate representation of complex signals of wind turbines, improving the effectiveness of fault detection and the adaptability of the model.

[0147] The comprehensive fault feature vector is mapped to a multi-dimensional space through a fully connected layer to generate a fault prediction;

[0148] In this multidimensional space, the fully connected layer maps the input feature vector to the output feature space through weights and biases to generate the final fault prediction result;

[0149] The output layer outputs the fault type and prediction results based on the fault prediction;

[0150] Generate a fault diagnosis report with fault type, fault location, fault severity, timestamp and data chart based on the fault type and prediction results;

[0151] Fault type describes the specific fault type identified, such as bearing fault, gear fault, etc.

[0152] The fault location indicates the specific location where the fault occurred, such as which component or area of ​​the wind turbine has a problem;

[0153] Fault Severity assesses the severity of the fault, such as minor, moderate, or severe;

[0154] Evaluate the severity of the fault according to the fault severity, and set thresholds:

[0155] Minor fault: h t <If h < h1, the fault is evaluated as minor. Minor faults usually do not have a significant impact on the normal operation of the equipment, but should be monitored;

[0156] Moderate fault: h1 < h t <h2, the fault is evaluated as moderate. Moderate faults may affect the equipment performance and require regular inspections and preventive maintenance;

[0157] Severe fault: h t ≥ h2, the fault is evaluated as severe. Severe faults will significantly affect the equipment operation, may cause downtime or equipment damage, and immediate repair measures need to be taken;

[0158] It should be noted that the threshold setting for evaluating the severity of the fault can be set according to the actual situation;

[0159] The timestamp records the specific time point when the fault is detected;

[0160] The data chart includes the chart display of the fault characteristic data, such as time domain and frequency domain charts, to visually display the fault characteristics;

[0161] Through the fault diagnosis report, automatically generate maintenance suggestions and make maintenance measures;

[0162] Specifically, ​​​​​​​​​​​​​​​​​​​​​​Send an email with detailed fault information;

[0171] If the maintenance team uses a dedicated mobile app, send instant notifications;

[0172] It should be noted that by constructing a fusion neural network model, the spatial and temporal information of fault characteristics can be extracted to generate accurate fault prediction results. Combined with the convolutional layer and the long short-term memory network layer, the model can identify abnormal patterns and long-term dependencies of wind turbines. The adaptive pooling and fully connected layers are used to improve the effectiveness and generalization ability of feature extraction and classification. Diagnostic reports and maintenance recommendations can help to quickly take maintenance measures, reduce the risk of equipment failure, and ensure operational stability and reliability. The automatic alarm and real-time monitoring mechanism further improve the response speed and efficiency of fault management.

[0173] This embodiment also provides an intelligent wind turbine fault detection system, including: an acquisition module, a fusion module, a detection module and a prediction module; the acquisition module is used to acquire multidimensional data and perform preprocessing; the fusion module is used to extract multidimensional features from the preprocessed multidimensional data, and fuse the multidimensional features through graph neural networks and short-time Fourier transform to generate time-frequency domain signals; the detection module is used to extract key features of the time-frequency domain signals based on the time-frequency domain signals, and detect the fault characteristics of the wind turbine operating status through a support vector machine model; the prediction module is used to construct a fusion neural network model based on the fault characteristics, generate a fault prediction, output a diagnosis report, and make maintenance suggestions.

[0174] This embodiment also provides a computer device, which is applicable to a method for intelligent detection of wind turbine faults, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a method for intelligent detection of wind turbine faults as proposed in the above embodiment.

[0175] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0176] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements a wind turbine fault intelligent detection method 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), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0177] In summary, the present invention combines convolutional neural networks and long short-term memory networks through a fusion neural network model to extract spatial features and time series data, thereby realizing multi-dimensional characterization of wind turbine fault characteristics. The pooling layer performs dimensionality reduction to improve computational efficiency. After the features are fused, a comprehensive fault feature vector is generated and mapped to multi-dimensional space through a fully connected layer to generate accurate fault predictions. Finally, the output layer generates diagnostic reports and maintenance recommendations based on the fault predictions, which not only improves the accuracy and real-time performance of fault detection, but also significantly improves the overall reliability and maintenance efficiency.

[0178] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A wind turbine fault intelligent detection method, characterized by: include, Collect multidimensional data and perform preprocessing; Extract multidimensional features from preprocessed multidimensional data, fuse them through graph neural network and short-time Fourier transform to generate time-frequency domain signals; Extract key features from time-frequency domain signals and detect fault characteristics of wind turbine operation status using support vector machine model; Based on fault characteristics, a fusion neural network model is constructed to generate fault predictions, output diagnostic reports, and provide maintenance recommendations; The multidimensional data includes temperature data, vibration data and current data; Multidimensional features, including Extract average temperature, temperature change rate, and temperature peak value from temperature data as temperature features; Extract vibration amplitude, vibration frequency, and vibration energy from vibration data as vibration features; Extracting average current, current fluctuation, and current peak from current data as current features; The multi-dimensional features are fused through graph neural network and short-time Fourier transform to generate time-frequency domain signals. The specific steps are as follows: The temperature characteristics, vibration characteristics and current characteristics are used as nodes of the graph; The node features in the graph are fused through the graph neural network to generate a comprehensive feature set, which is expressed as: Among them, hv (k) is the integrated feature set after fusion, h is the feature vector, k is the kth iteration in the graph neural network, v is the node, σ is the activation function, N(v) is the set function, is the feature vector of the neighbor node u in the k-1th iteration of the graph neural network, where u is the neighbor node of node v; Based on the fused comprehensive feature set, short-time Fourier transform is performed to generate time-frequency domain signals.

2. The method for intelligent detection of wind turbine faults according to claim 1, characterized in that: The preprocessing includes denoising, filtering and outlier removal.

3. The intelligent detection method for wind turbine faults according to claim 1, characterized in that: The time-frequency domain signal is expressed as: Among them, Z(t,ω) is the time-frequency domain signal, t is the time, ω is the angular frequency, A(θ) is the time domain signal, θ is the integral variable, w(θ-t) is the window function, e -jωθ is a complex exponential function, j is an imaginary unit, is the integral symbol, ∞ is positive infinity, and -∞ is negative infinity.

4. The method for intelligent detection of wind turbine faults according to claim 1, characterized in that: The key features are extracted and the fault features of the wind turbine operating status are detected by the support vector machine model. The specific steps are: Extract spectral features, time domain features and time-frequency features from time-frequency domain signals; Through multi-scale analysis, the time-varying characteristics of different frequency components in the time-frequency domain signals are captured, and the spikes and abnormal changes in the time-frequency domain signals are identified; The dimensionality reduction of spectrum features, time domain features, time-frequency features and time-varying features is performed through principal component analysis; Use local linear embedding to find the neighbors of data points in the spectral features, time domain features, time-frequency features, and time-varying features after dimension reduction, and determine the local neighborhood; In the local neighborhood of each data point, the linear reconstruction weight is calculated to minimize the embedding reconstruction error; Find low-dimensional embedding coordinates that minimize the embedding-reconstruction error and capture nonlinear structural features; The spectrum features, time domain features, time-frequency features and time-varying features after principal component analysis and local linear embedding are input into multiple classifiers, and the classification results are fused by voting to generate a comprehensive feature vector; Use the pre-trained support vector machine model to train and classify the comprehensive feature vector to identify the fault type and severity of the wind turbine; Cross-validate the support vector machine model to evaluate performance; Continuously optimize feature extraction and classification based on verification results; The optimized support vector machine model is applied to real-time monitoring to detect fault characteristics of wind turbine operating status.

5. The intelligent fault detection method for a wind turbine according to claim 1, characterized in that: The specific steps of building a fusion neural network model, generating fault predictions, outputting diagnostic reports, and proposing maintenance recommendations are as follows: Define the fusion neural network model input as fault characteristics and output as fault type and prediction results; Extract spatial features from fault features through convolutional layers; Use the adaptive pooling layer to reduce the dimensionality of the spatial features extracted by the convolution layer and dynamically adjust the pooling parameters according to the input fault characteristics; Capture time series data in fault characteristics through the long short-term memory network layer to obtain long-term dependencies and dynamic features in time series data; The spatial features and dynamic features are integrated to generate the comprehensive fault feature vector, which is expressed as: P t =tanh(CNN(x t )+LSTM(x t-1 )+b); Among them, P t is the fault comprehensive feature vector, tanh is the hyperbolic tangent activation function, CNN(x t ) is the output of the convolutional layer, x t is the input feature at time t, LSTM(x t-1 ) is the output of the long short-term memory network layer, x t-1 is the input feature at time t-1, b is the hyperbolic tangent activation function; The comprehensive fault feature vector is mapped to a multi-dimensional space through a fully connected layer to generate a fault prediction; The output layer outputs the fault type and prediction results based on the fault prediction; Generate a fault diagnosis report with fault type, fault location, fault severity, timestamp and data chart based on the fault type and prediction results; Through fault diagnosis reports, maintenance suggestions are automatically generated and maintenance measures are taken.

6. An intelligent wind turbine fault detection system, based on the intelligent wind turbine fault detection method according to any one of claims 1 to 5, characterized in that: Including, acquisition module, fusion module, detection module and prediction module; The acquisition module is used to collect multidimensional data and perform preprocessing; The fusion module is used to extract multidimensional features from the preprocessed multidimensional data, and fuse the multidimensional features through graph neural network and short-time Fourier transform to generate time-frequency domain signals; The detection module is used to extract key features of the time-frequency domain signals based on the time-frequency domain signals and detect fault features of the wind turbine operating status through a support vector machine model; The prediction module is used to build a fusion neural network model based on fault characteristics, generate fault predictions, output a diagnosis report, and make maintenance recommendations.

7. 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 wind turbine fault intelligent detection method according to any one of claims 1 to 5 are implemented.

8. 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 intelligent detection method according to any one of claims 1 to 5 are implemented.

Images