Fault detection method and system for wind driven generator based on deep learning

Abstract

The invention relates to the technical field of wind power generation, in particular to a fault detection method and system for a wind driven generator based on deep learning. According to the wind driven generator fault detection method based on deep learning, a network model is constructed by introducing a bidirectional long-short-term memory network and a Softmax classifier through an improved convolutional neural network, and model pre-training is carried out based on operation historical data and a fault type so as to obtain a pre-trained fault detection network model. Introducing a transfer learning mechanism training model to obtain a trained fault detection network model; the trained fault detection network model can more comprehensively and accurately capture fault features by extracting multi-scale local features and carrying out time sequence feature learning, so that the fault detection precision is improved, and misinformation and missing report conditions are reduced. Meanwhile, by introducing a transfer learning mechanism, the model can quickly adapt to different types of wind driven generators.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a fault detection method and system for wind turbines based on deep learning. Background Technology

[0002] As a clean and renewable energy source, wind power is playing an increasingly important role in the energy structure. However, wind power plants operate in complex and ever-changing natural environments, making their key components prone to failure. Failure to detect and address these failures promptly can lead to shutdowns, causing significant economic losses and impacting power generation efficiency.

[0003] Traditional wind turbine fault detection methods mainly rely on manual inspections and simple sensor monitoring. This approach is overly dependent on human experience, resulting in low detection accuracy, poor real-time performance, and high labor intensity. With the development of deep learning technology, its powerful feature learning and pattern recognition capabilities have shown great potential in the field of fault detection. Currently, some research has applied deep learning to wind turbine fault detection; however, existing technologies still have shortcomings in terms of the comprehensiveness of feature extraction, the effectiveness of temporal feature processing, and adaptability to different wind turbines, making it difficult to meet the high-precision, high-real-time fault detection requirements of practical applications. Summary of the Invention

[0004] Based on the aforementioned problems in the existing technology, the purpose of this invention is to provide a fault detection method and system for wind turbines based on deep learning. By extracting multi-scale local features through an improved CNN and combining Bi-LSTM and multi-head self-attention mechanism to learn effective temporal features, the fault features can be captured more comprehensively and accurately, thereby improving the accuracy of fault detection and reducing false alarms and missed alarms.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A fault detection method for wind turbines based on deep learning, comprising: Step S1: Collect historical operating data and fault types of various wind turbines under different operating conditions; Step S2: Obtain real-time operating data of various wind turbines, preprocess the operating data to obtain an initial dataset, and divide the initial dataset into a training set, a validation set, and a test set according to the proportions. Step S3: Construct a network model based on the improved convolutional neural network and introduce a bidirectional long short-term memory network and a Softmax classifier, and pre-train the model based on the historical running data and the fault type to obtain a fault detection network model; Step S4: Input the training set into the fault detection network model, introduce the transfer learning mechanism to train the model, and adjust the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model. Step S5: Input the validation set into the trained fault detection network model for model evaluation; Step S6: Input the test set into the trained fault detection network model and output the fault detection results.

[0006] A further improvement of the present invention is that the step of acquiring various real-time operating data of the wind turbine in step S2 includes: collecting vibration data, temperature data, rotational speed data and current and voltage data of the wind turbine in real time through various sensors installed on various key components of the wind turbine; the various key components include: gearbox, generator, blades and bearings.

[0007] A further improvement of the present invention is that the step of preprocessing the running data in step S2 to obtain an initial dataset includes: performing data cleaning and data standardization on the running data; The data cleaning method uses a combination of the 3σ criterion and the isolated forest algorithm to remove noise and outliers; The data standardization adopts the z-score standardization method, and the formula is: ; in, The original data, The mean of the dimensional data. is the standard deviation of the data in the stated dimension.

[0008] A further improvement of the present invention is that the step of constructing the network model in step S3, which involves an improved convolutional neural network and the introduction of a bidirectional long short-term memory network and a softmax classifier, includes: The historical data of operation is convolved through the convolutional neural network to extract local features at different scales and integrate them to output fault features; The fault features are input into the bidirectional long short-term memory network that incorporates a multi-head self-attention mechanism for temporal feature learning. The fault features learned through time-series feature learning are input into the Softmax classifier, and the faults are classified based on the fault types to obtain the detection results of different types of faults of various wind turbines. The different types of faults include gearbox wear, gearbox tooth breakage, generator overheating, generator short circuit, blade cracks, blade breakage, bearing wear, bearing cage damage, and normal state.

[0009] A further improvement of the present invention is that the improved convolutional neural network includes an input layer, three convolutional layers, three pooling layers, and one fully connected layer; the input layer receives the historical running data; the three convolutional layers use different convolutional kernels to perform convolution operations to extract local features at different scales; the pooling layer uses max pooling; and the fully connected layer integrates the extracted local features. The multi-head self-attention mechanism has 8 heads, and each head has a dimension of 64; the bidirectional long short-term memory network includes 2 hidden layers, and each hidden layer has 128 units.

[0010] A further improvement of the present invention is that the step S4, in which the training set is input into the fault detection network model and a transfer learning mechanism is introduced for model training, includes: The training set is input into the fault detection network model, the transfer learning mechanism is introduced, and the parameters of the top-level network of the fault detection network model are fine-tuned by freezing the bottom-level network of the fault detection network model through a hierarchical fine-tuning strategy.

[0011] A further improvement of the present invention is that the step S4, which involves adjusting the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model, includes: The model was trained using the PyTorch framework with a learning rate of 0.0001, a batch size of 32, and 50 training epochs. The model was trained using a cross-entropy loss function and a domain-adaptive loss function. The Adam optimizer was used to optimize the model parameters. During the model training process, the model's performance is monitored in real time using the validation set. When the model's performance on the validation set no longer improves, training is stopped to prevent overfitting.

[0012] A further improvement of this invention is that step S5 includes: inputting the test set into the trained fault detection network model, and evaluating the model based on its precision P, recall R, and mean precision mAP; the formula for evaluating model performance is as follows:

[0013]

[0014]

[0015] Where TP is the number of correctly predicted positive samples, FN is the number of incorrectly predicted negative samples, and FP is the number of incorrectly predicted positive samples.

[0016] A fault detection system for wind turbines based on deep learning, comprising: The data acquisition module allows users to collect historical operating data and fault types of various wind turbines under different operating conditions. The dataset acquisition module is used to acquire real-time operating data of various wind turbines, preprocess the operating data to obtain an initial dataset, and divide the initial dataset into a training set, a validation set, and a test set according to a certain ratio. The model building module is used to build a network model based on an improved convolutional neural network and introduce a bidirectional long short-term memory network and a softmax classifier, and to pre-train the model based on the historical running data and the fault type to obtain a fault detection network model. The model training module is used to input the training set into the fault detection network model, introduce a transfer learning mechanism to train the model, and adjust the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model. The model validation module is used to input the validation set into the trained fault detection network model for model evaluation. The result generation test module is used to input the test set into the trained fault detection network model and output the fault detection results.

[0017] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the deep learning-based fault detection method for wind turbines.

[0018] Compared with the prior art, the present invention has at least the following beneficial technical effects: This invention provides a deep learning-based method and system for fault detection of wind turbines. The deep learning-based fault detection method constructs a network model using an improved convolutional neural network, incorporating a bidirectional long short-term memory network and a Softmax classifier. The model is pre-trained based on historical operating data and the fault types to obtain a pre-trained fault detection network model. A transfer learning mechanism is then introduced to train the model, resulting in a trained fault detection network model. This trained fault detection network model extracts multi-scale local features and performs temporal feature learning, enabling it to capture fault features more comprehensively and accurately, thereby improving fault detection accuracy and reducing false alarms and false negatives. Furthermore, the introduced transfer learning mechanism allows the model to quickly adapt to different types of wind turbines, leveraging existing knowledge to improve fault detection performance on new turbines and reducing reliance on large amounts of labeled fault data from new wind turbines. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the fault detection method for wind turbines based on deep learning according to Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of the module of the deep learning-based fault detection system for wind turbines according to Embodiment 2 of the present invention. Detailed Implementation

[0021] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0022] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0026] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] Example 1 like Figure 1 As shown in the figure, this embodiment provides a fault detection method for wind turbines based on deep learning. The fault detection method includes: Step S1: Collect historical operating data and fault types of various wind turbines under different operating conditions.

[0028] In this embodiment, the historical operating data includes system data of the wind turbine (vibration data, temperature data, speed and torque data), electrical system data (electrical parameters of the wind turbine, converter data, insulation resistance data, etc.), hydraulic and braking system data (hydraulic system pressure, hydraulic oil level, filter differential pressure, reflecting whether the hydraulic power transmission is normal and the status of the braking system, etc.), environmental condition data (meteorological data such as wind speed, wind direction, ambient temperature, humidity, air pressure, precipitation, solar radiation intensity, salt spray concentration, wave height, etc., as well as the current operating mode and altitude of the wind turbine), control and regulation system data (pitch data, yaw data, control system commands), performance and energy consumption data (power generation, equipment utilization rate, wind curtailment rate, and energy consumption of each component, etc.), and fault and maintenance record data. Among them, the fault record data includes historical fault types (such as gearbox high-speed shaft bearing failure, generator insulation breakdown, blade cracks, etc.), fault occurrence time, snapshots of key parameters at the time of fault (such as vibration values ​​and temperature values ​​10 minutes before the fault), and fault duration. Maintenance records include repair time, replaced parts (such as bearing type, gearbox oil replacement volume), post-repair parameter calibration values ​​(such as pitch angle zero-point calibration), and maintenance cycles (such as regular lubrication and filter replacement records). Fault types include gearbox wear, broken gears, generator overheating, generator short circuit, blade cracks, blade breakage, bearing wear, and bearing cage damage, among others.

[0029] Step S2: Obtain real-time operating data from various wind turbines, preprocess the operating data to obtain an initial dataset, and divide the initial dataset into a training set, a validation set, and a test set according to a ratio. Specifically, the initial dataset is randomly divided into the training set, validation set, and test set in a ratio of 7:2:1.

[0030] In this embodiment, the step of obtaining real-time operating data of various wind turbines in step S2 includes: collecting vibration data, temperature data, speed data, current and voltage data of the wind turbines in real time through various sensors installed on various key components of the wind turbines; the key components include: gearbox, generator, blades, and bearings.

[0031] As an optional embodiment, the step of preprocessing the running data in step S2 to obtain an initial dataset includes: data cleaning and data standardization of the running data; wherein, the data cleaning adopts a method based on the 3σ criterion (also known as the Laida criterion) and the isolated forest algorithm to remove noise and outliers; specifically, for vibration data, temperature data, etc. that follow a normal distribution, the 3σ criterion is used to identify and remove data that exceeds the mean ± 3 times the standard deviation; for non-normally distributed data such as rotational speed data, current and voltage data, the isolated forest algorithm is used to detect and remove outliers.

[0032] Data standardization uses the z-score standardization method, with the following formula: ; in, The original data, The mean of the dimensional data. is the standard deviation of the data in the stated dimension.

[0033] Step S3: Construct a network model based on an improved convolutional neural network and introduce a bidirectional long short-term memory network and a Softmax classifier. Pre-train the model based on historical data and fault types to obtain a fault detection network model.

[0034] In this embodiment, the step S3, which involves constructing a network model using an improved convolutional neural network and incorporating a bidirectional long short-term memory network and a Softmax classifier, includes: Historical operating data is convolutionally processed using a Convolutional Neural Network (CNN) to extract local features at different scales and integrate them to output fault features. These fault features are then input into a Bi-directional Long Short-Term Memory (BiLSTM) network with a multi-head self-attention mechanism for temporal feature learning. The fault features after temporal feature learning are then input into a Softmax classifier for classification based on fault type, thereby obtaining detection results for different types of faults in various wind turbines. These different fault types include gearbox wear, broken gearbox teeth, generator overheating, generator short circuit, blade cracks, blade breakage, bearing wear, bearing cage damage, and normal operation.

[0035] In existing technologies, CNN network models applied to wind turbine fault detection mostly use single-size convolutional kernels, such as 3×3 or 5×5 kernels. This structure has significant limitations in extracting local features from wind turbine operating data. Wind turbine operating data has multi-scale characteristics; for example, vibration signals contain both high-frequency subtle vibration information (such as slight wear on the gearbox) and low-frequency macroscopic vibration information (such as the overall oscillation of the blades). Single-size convolutional kernels cannot comprehensively capture these features at different scales, resulting in incomplete extraction of local features, which in turn affects the accuracy of subsequent fault detection.

[0036] As an optional embodiment, to solve the above problems, the improved convolutional neural network of this embodiment includes an input layer, three convolutional layers (3×3, 5×5, 7×7), three pooling layers, and one fully connected layer; the input layer receives historical running data; the three convolutional layers use different convolutional kernels to perform convolution operations to extract local features at different scales; the pooling layer uses max pooling, and the fully connected layer integrates the extracted local features.

[0037] Existing BiLSTM network models lack an effective attention mechanism when processing wind turbine time-series data. While BiLSTM networks can capture the dependencies between time-series data, they give equal attention to features at all time steps when dealing with large amounts of wind turbine operating data, failing to highlight important features at key time points. For example, in the period leading up to a wind turbine failure, certain operational data, such as sudden temperature increases or abnormal vibration frequency changes, are crucial for fault detection. However, existing BiLSTM networks fail to focus on this key information, resulting in poor learning performance for time-series features.

[0038] As an optional embodiment, to address the aforementioned issues, this embodiment introduces a multi-head self-attention mechanism into the Bi-LSTM network. This mechanism enables the model to focus on key time-point features in the wind turbine operation data. The mechanism includes eight attention heads, each with a dimension of 64; the Bi-LSTM network comprises two hidden layers, each with 128 units. In this embodiment, the temporal features processed by the Bi-LSTM network first undergo a linear transformation, then are distributed to the eight attention heads for parallel computation. Each attention head focuses on features at different locations. Finally, the computation results from each attention head are concatenated and linearly transformed to obtain the final temporal features.

[0039] Step S4: Input the training set into the fault detection network model, introduce the transfer learning mechanism to train the model, and adjust the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model.

[0040] In this embodiment, step S4, which involves inputting the training set into the fault detection network model and introducing a transfer learning mechanism for model training, includes: inputting the training set into the fault detection network model, introducing a transfer learning mechanism, and using a layered fine-tuning strategy to freeze the bottom layers of the fault detection network model and fine-tuning only the parameters of the top layers. Specifically, the first two convolutional layers of the CNN network and the bottom layer of the BiLSTM network are frozen to retain general feature extraction capabilities; simultaneously, the third convolutional layer of the CNN network, the fully connected layer, the top layer of the BiLSTM, and the attention module are unfrozen to adapt to the target domain features.

[0041] In the field of wind turbine fault detection, existing applications of transfer learning are relatively simple, often directly transferring models trained on data from one type of wind turbine to other types, with only minor adjustments to the model parameters. However, due to significant differences in structure, operating environment, and performance parameters among different types of wind turbines, the distribution of their fault data also varies. This simplistic transfer learning approach fails to adequately consider the differences in data distribution across different turbine models, making it difficult for the model to quickly and accurately adapt to new wind turbines. This results in poor transfer performance, requiring a large amount of new labeled data to achieve satisfactory detection results, thus increasing the cost of practical applications.

[0042] As an optional implementation, to address the aforementioned issues, a pre-trained fault detection network model is constructed based on a large amount of historical operating data (including faulty data) of different types of wind turbines. This model contains rich fault characteristics of wind turbines. When migrating the fault detection network model to new wind turbines, a hierarchical fine-tuning strategy is adopted, i.e., the bottom layer of the fault detection network model is frozen, and only the parameters of the top layer network are fine-tuned. Simultaneously, a domain-adaptive loss function is introduced to reduce the distributional differences between the source domain (pre-trained data) and the target domain (new turbine data).

[0043] As an optional embodiment, step S4, which involves adjusting the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model, includes: The model was trained using the PyTorch framework with a learning rate of 0.0001, a batch size of 32, and 50 training epochs. The model training loss function was based on cross-entropy loss and a domain-adaptive loss function. The Adam optimizer was used to optimize the model parameters.

[0044] Step S5: Input the validation set into the trained fault detection network model for model evaluation.

[0045] Step S6: Input the test set into the trained fault detection network model and output the fault detection results.

[0046] During model training, model performance is monitored in real time using a validation set. Training is stopped when the model's performance on the validation set no longer improves to prevent overfitting. Specifically, the test set is input into the trained fault detection network model, and its performance is evaluated based on the precision (P), recall (R), and mean precision (mAP). The formulas for evaluating model performance are as follows:

[0047]

[0048]

[0049] Where TP is the number of correctly predicted positive samples, FN is the number of incorrectly predicted negative samples, and FP is the number of incorrectly predicted positive samples.

[0050] This embodiment provides a deep learning-based fault detection method for wind turbines. The method constructs a network model using an improved convolutional neural network, incorporating a bidirectional long short-term memory network and a Softmax classifier. The model is pre-trained based on historical operating data and the fault types to obtain a pre-trained fault detection network model. A transfer learning mechanism is then introduced to train the model, resulting in a trained fault detection network model. This trained model extracts multi-scale local features and performs temporal feature learning, enabling it to capture fault features more comprehensively and accurately, thereby improving fault detection accuracy and reducing false alarms and false negatives. Furthermore, the introduced transfer learning mechanism allows the model to quickly adapt to different types of wind turbines, leveraging existing knowledge to improve fault detection performance on new turbines and reducing reliance on large amounts of labeled fault data from new wind turbines.

[0051] Example 2 like Figure 2 As shown, this embodiment provides a fault detection system 10 for wind turbines based on deep learning. The fault detection system includes: a data acquisition module 11, a dataset acquisition module 12, a model building module 13, a model training module 14, a model verification module 15, and a result generation and testing module 16.

[0052] Data acquisition module 11 allows users to collect historical operating data and fault types of various wind turbines under different operating conditions.

[0053] The dataset acquisition module 12 is used to acquire real-time operating data of various wind turbines, preprocess the operating data to obtain an initial dataset, and divide the initial dataset into training set, validation set and test set according to the proportion.

[0054] The model building module 13 is used to build a network model based on an improved convolutional neural network and introduce a bidirectional long short-term memory network and a Softmax classifier, and to pre-train the model based on historical running data and fault types to obtain a fault detection network model.

[0055] The model training module 14 is used to input the training set into the fault detection network model, introduce the transfer learning mechanism to train the model, and adjust the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model.

[0056] The model validation module 15 is used to input the validation set into the trained fault detection network model for model evaluation.

[0057] The result generation test module 16 is used to input the test set into the trained fault detection network model and output the fault detection results.

[0058] This embodiment provides a deep learning-based fault detection system for wind turbines. This fault detection system is based on the deep learning-based fault detection method for wind turbines proposed in Embodiment 1 above; specific details are not repeated here. This embodiment's deep learning-based fault detection system for wind turbines constructs a network model using an improved convolutional neural network and incorporating a bidirectional long short-term memory network and a Softmax classifier. The model is pre-trained based on historical operating data and the fault types to obtain a pre-trained fault detection network model. A transfer learning mechanism is then introduced to train the model, resulting in a trained fault detection network model. This trained fault detection network model extracts multi-scale local features and performs temporal feature learning, enabling it to capture fault features more comprehensively and accurately, thereby improving fault detection accuracy and reducing false alarms and false negatives. Simultaneously, the introduced transfer learning mechanism allows the model to quickly adapt to different types of wind turbines, leveraging existing knowledge to improve fault detection performance on new turbines and reducing reliance on large amounts of labeled fault data from new wind turbines.

[0059] Example 3 This embodiment relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the deep learning-based fault detection method for wind turbines described in Embodiment 1 above.

[0060] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0061] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0062] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A fault detection method for wind turbines based on deep learning, characterized in that, include: Step S1: Collect historical operating data and fault types of various wind turbines under different operating conditions; Step S2: Obtain real-time operating data of various wind turbines, preprocess the operating data to obtain an initial dataset, and divide the initial dataset into a training set, a validation set, and a test set according to the proportions. Step S3: Construct a network model based on the improved convolutional neural network and introduce a bidirectional long short-term memory network and a Softmax classifier, and pre-train the model based on the historical running data and the fault type to obtain a fault detection network model; Step S4: Input the training set into the fault detection network model, introduce the transfer learning mechanism to train the model, and adjust the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model. Step S5: Input the validation set into the trained fault detection network model for model evaluation; Step S6: Input the test set into the trained fault detection network model and output the fault detection results.

2. The fault detection method for wind turbines based on deep learning according to claim 1, characterized in that, The step of obtaining various real-time operating data of the wind turbine in step S2 includes: collecting vibration data, temperature data, speed data and current and voltage data of the wind turbine in real time through various sensors installed on various key components of the wind turbine; the various key components include: gearbox, generator, blades and bearings.

3. The fault detection method for wind turbines based on deep learning according to claim 2, characterized in that, The step of preprocessing the running data in step S2 to obtain an initial dataset includes: performing data cleaning and data standardization on the running data; The data cleaning method uses a combination of the 3σ criterion and the isolated forest algorithm to remove noise and outliers; The data standardization adopts the z-score standardization method, and the formula is: ; in, The original data, The mean of the dimensional data. is the standard deviation of the data in the stated dimension.

4. The fault detection method for wind turbines based on deep learning according to claim 1, characterized in that, The steps in step S3, which involve constructing a network model using an improved convolutional neural network and incorporating a bidirectional long short-term memory network and a softmax classifier, include: The historical data of operation is convolved through the convolutional neural network to extract local features at different scales and integrate them to output fault features; The fault features are input into the bidirectional long short-term memory network that incorporates a multi-head self-attention mechanism for temporal feature learning. The fault features learned through time-series feature learning are input into the Softmax classifier, and the faults are classified based on the fault types to obtain the detection results of different types of faults of various wind turbines. The different types of faults include gearbox wear, gearbox tooth breakage, generator overheating, generator short circuit, blade cracks, blade breakage, bearing wear, bearing cage damage, and normal state.

5. The fault detection method for wind turbines based on deep learning according to claim 4, characterized in that, The improved convolutional neural network includes an input layer, three convolutional layers, three pooling layers, and one fully connected layer. The input layer receives the historical running data. The three convolutional layers use different convolutional kernels to perform convolution operations to extract local features at different scales. The pooling layer uses max pooling, and the fully connected layer integrates the extracted local features. The multi-head self-attention mechanism has 8 heads, and each head has a dimension of 64; the bidirectional long short-term memory network includes 2 hidden layers, and each hidden layer has 128 units.

6. The fault detection method for wind turbines based on deep learning according to claim 1, characterized in that, The step S4, which involves inputting the training set into the fault detection network model and introducing a transfer learning mechanism for model training, includes: The training set is input into the fault detection network model, the transfer learning mechanism is introduced, and the parameters of the top-level network of the fault detection network model are fine-tuned by freezing the bottom-level network of the fault detection network model through a hierarchical fine-tuning strategy.

7. The fault detection method for wind turbines based on deep learning according to claim 6, characterized in that, The step S4, which involves adjusting the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model, includes: The model was trained using the PyTorch framework with a learning rate of 0.0001, a batch size of 32, and 50 training epochs. The model was trained using a cross-entropy loss function and a domain-adaptive loss function. The Adam optimizer was used to optimize the model parameters. During the model training process, the model's performance is monitored in real time using the validation set. When the model's performance on the validation set no longer improves, training is stopped to prevent overfitting.

8. The fault detection method for wind turbines based on deep learning according to claim 1, characterized in that, Step S5 includes: inputting the test set into the trained fault detection network model, and evaluating the model based on its precision P, recall R, and mean precision mAP; the formula for evaluating the model performance is as follows: Where TP is the number of correctly predicted positive samples, FN is the number of incorrectly predicted negative samples, and FP is the number of incorrectly predicted positive samples.

9. A fault detection system for wind turbines based on deep learning, characterized in that, include: The data acquisition module allows users to collect historical operating data and fault types of various wind turbines under different operating conditions. The dataset acquisition module is used to acquire real-time operating data of various wind turbines, preprocess the operating data to obtain an initial dataset, and divide the initial dataset into a training set, a validation set, and a test set according to a certain ratio. The model building module is used to build a network model based on an improved convolutional neural network and introduce a bidirectional long short-term memory network and a softmax classifier, and to pre-train the model based on the historical running data and the fault type to obtain a fault detection network model. The model training module is used to input the training set into the fault detection network model, introduce a transfer learning mechanism to train the model, and adjust the parameters of the fault detection network model based on the training results to obtain the trained fault detection network model. The model validation module is used to input the validation set into the trained fault detection network model for model evaluation. The result generation test module is used to input the test set into the trained fault detection network model and output the fault detection results.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based fault detection method for wind turbines as described in any one of claims 1 to 8.

Images