Wind turbine condition recognition model training method, recognition method, equipment and medium

Abstract

This application provides a training method, recognition method, device, and medium for a wind turbine state recognition model, relating to the field of data processing technology. This method expands the coverage of the parameter space by utilizing multiple mutually orthogonal initialized parameter spaces. Loss is calculated in each parameter space, and the loss results are normalized to determine the combination coefficients of each parameter space, thus forming a joint parameter space for meta-level updates. The joint parameter space is updated in the inner loop. In the outer loop, each parameter space is updated sequentially based on the updated gradient and its performance on the task query set. The parameter space used by the target recognition model has better generalization ability than any single target parameter space. Only a small amount of target domain training data is needed to obtain a target recognition model with good recognition performance. Furthermore, training with samples from multiple wind turbine operating conditions can further improve the model's generalization ability.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a wind turbine state recognition model training method, recognition method, equipment, and medium. Background Technology

[0002] Wind energy, as a clean and renewable energy source, plays a crucial role in reducing greenhouse gas emissions and addressing climate change. Against the backdrop of a global energy transition towards a low-carbon model, wind turbines are playing an increasingly important role in the new energy sector. In the daily management of wind turbines, condition identification is essential for operation and maintenance. It assesses the health status and power generation capacity of the turbine, enabling timely development of maintenance strategies and technical upgrades, reducing downtime, preventing major failures, and saving maintenance costs. However, the scarcity of wind turbine fault states in actual engineering projects makes obtaining large-scale labeled fault state data extremely difficult, resulting in limited data and hindering the training of robust models. Furthermore, since monitoring data may come from different operating domains of the wind turbine, these conditions may involve the coupling of multiple factors, such as wind speed, temperature, and altitude, which presents additional challenges to condition identification. Summary of the Invention

[0003] This application provides a wind turbine condition recognition model training method, recognition method, equipment and medium, which can eliminate the problem of poor recognition effect caused by the large difference in data distribution and small sample size of wind turbine parameter signals collected in actual engineering projects from different wind turbine operating conditions.

[0004] This application provides a method for training a wind turbine state recognition model, including:

[0005] Obtain the recognition model to be trained, obtain the source domain training samples for training the recognition model, obtain the initialized parameter space set, the parameter space set includes multiple parameter spaces of the recognition model, the source domain training samples include a support set and a query set, the source domain training samples include multiple wind turbine parameter signals, and the wind turbine state is marked as a label in the source domain training samples;

[0006] The iterative training process includes: sampling multiple batches of target training samples from the source domain training samples based on a preset number of support set samples and a preset number of query set samples, with each batch of target training samples corresponding to different operating conditions of the wind turbine; iteratively training multiple parameter spaces of the recognition model based on the multiple batches of target training samples; in the inner loop of the iterative training, determining the loss of each parameter space in the parameter space set in the support set, normalizing each loss to obtain a combination coefficient, combining each initialized parameter space into a joint parameter space based on the combination coefficient, and performing gradient update on the joint parameter space based on the support set to obtain a gradient-updated parameter space; in the outer loop of the iterative training, determining the loss of the gradient-updated parameter space in the query set to obtain a first loss, determining a second loss based on the current training round number and the first loss, determining the sum of the second losses of each batch of target training samples to obtain a third loss, and updating the parameter space in the parameter space set in a polling manner based on the third loss to obtain the parameter space updated in the current iteration.

[0007] Training stops when the current training round number equals the preset training round number, resulting in a space of multiple target parameters after training is complete.

[0008] Obtain training samples for the target domain;

[0009] Based on multiple target parameter spaces and the target domain training samples, a target recognition model is determined.

[0010] In one embodiment of this application, obtaining the initialized parameter space set includes:

[0011] The parameter spaces in the parameter space set are initialized using the Schmitt orthogonalization method to obtain multiple parameter spaces.

[0012] In one embodiment of this application, the combination coefficients are obtained by normalizing each loss, including:

[0013] The loss of each parameter space in the parameter space set is normalized based on the nonlinear activation function to obtain multiple spatial coefficients, and the set of multiple spatial coefficients is determined as the combined coefficients.

[0014] In one embodiment of this application, gradient update parameter space is obtained by performing gradient update on the joint parameter space based on the support set, including:

[0015] The joint parameter space is loaded into the recognition model to determine the fourth loss of the joint parameter space on the support set;

[0016] Based on the joint parameter space, the learning rate of the preset base learner, and the fourth loss, gradient update is performed to obtain the gradient update parameter space.

[0017] In one embodiment of this application, determining a second loss based on the current training epoch and the first loss includes:

[0018] When the current training round number is an integer multiple of the preset regularization round number, the dot product of each initial parameter space is determined as the regularization loss, and the sum of the first loss and the regularization loss is determined as the second loss, wherein the preset regularization round number is less than the preset training round number;

[0019] When the current training round number is not an integer multiple of the preset regularization round number, the first loss is determined as the second loss.

[0020] In one embodiment of this application, a target recognition model is determined based on multiple target parameter spaces and training samples of the target domain, including:

[0021] The loss value for each target parameter space is determined based on the training samples in the target domain;

[0022] The loss values ​​of each objective parameter space are normalized to obtain the objective combination coefficients;

[0023] Based on the target combination coefficients, the target parameter spaces are combined into a target joint parameter space;

[0024] The target model parameter space is obtained by performing gradient update on the target joint parameter space based on the target domain training samples;

[0025] The target model parameter space is loaded into the recognition model to obtain the target recognition model.

[0026] In one embodiment of this application, updating the parameter space in the parameter space set based on the third loss in a polling manner includes:

[0027] The parameter space is updated in the current iteration by using a polling method based on the parameter space in the parameter space set, the learning rate of the preset meta-learner, and the third loss.

[0028] To achieve the above and other related objectives, this application also provides a wind turbine state identification method, including:

[0029] Obtain the parameters of the wind turbine to be identified;

[0030] The parameters of the wind turbine to be identified are input into the target identification model to obtain the wind turbine status;

[0031] The target recognition model is a target recognition model trained by one or more of the aforementioned training methods.

[0032] To achieve the above and other related objectives, this application also provides an electronic device, the electronic device comprising:

[0033] One or more processors;

[0034] Memory used to store the executable program code of the processor;

[0035] The processor is configured to execute the program code to implement one or more of the methods described above.

[0036] To achieve the above and other related objectives, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a computer's processor, causes the computer to perform one or more of the aforementioned methods.

[0037] As described above, the wind turbine state recognition model training method, recognition method, equipment, and medium provided in this application have the following beneficial effects:

[0038] This application discloses a wind turbine state recognition model training method. This method expands the coverage of the parameter space by utilizing multiple mutually orthogonal initialized parameter spaces. Loss is calculated in each parameter space, and the loss results are normalized to determine the combination coefficients of each parameter space, thus forming a joint parameter space for meta-level updates. The joint parameter space is updated in the inner loop. In the outer loop, each parameter space is updated sequentially based on the updated gradient and its performance on the task query set. The target recognition model is determined using target domain training samples and multiple target parameter spaces. The parameter space used in the target recognition model has better generalization ability than any single target parameter space, requiring only a small amount of target domain training data to obtain a target recognition model with good recognition performance. Furthermore, training with samples from multiple wind turbine operating conditions can further improve the model's generalization ability.

[0039] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0040] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0041] Figure 1This is a flowchart illustrating a wind turbine state recognition model training method in an exemplary embodiment of this application;

[0042] Figure 2 This is a schematic diagram illustrating the training logic of the inner and outer loops, as shown in an exemplary embodiment of this application.

[0043] Figure 3 This is a flowchart illustrating the source domain training phase of a wind turbine state recognition model, as shown in an exemplary embodiment of this application.

[0044] Figure 4 This is a flowchart illustrating the target domain training phase of a wind turbine state recognition model, as shown in an exemplary embodiment of this application.

[0045] Figure 5 This is a flowchart illustrating a wind turbine state identification method in an exemplary embodiment of this application. Detailed Implementation

[0046] The embodiments of this application will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be understood that the preferred embodiments are only for illustrating this application and are not intended to limit the scope of protection of this application.

[0047] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0048] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the present application. However, it will be apparent to those skilled in the art that embodiments of the present application may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the present application.

[0049] Please see Figure 1 , Figure 1 This is a flowchart illustrating a wind turbine state recognition model training method in an exemplary embodiment of this application. (Reference) Figure 1 It can be seen that the training method for this wind turbine state recognition model can include:

[0050] Step S110: Obtain the recognition model to be trained, obtain the source domain training samples used to train the recognition model, and obtain the initialized parameter space set.

[0051] The parameter space set includes multiple parameter spaces of the identification model, the source domain training samples include a support set and a query set, the source domain training samples contain multiple wind turbine parameter signals, and the wind turbine state is marked as a label in the source domain training samples.

[0052] In one embodiment of this application, the recognition model to be trained can be an untrained neural network model. During iterative training of the recognition model, multiple batches of target training samples can be sampled from the source domain training samples. Each batch of target training samples represents a task, and each task comes from a different distribution domain. Tasks share common characteristics as well as their own specific characteristics. The goal of meta-learning is to learn the common characteristics of the training tasks from the source domain, i.e., general knowledge. When facing a specific task, only a small number of samples are needed to transfer this general knowledge to the specific distribution domain of the task, forming specific knowledge adapted to that task.

[0053] Wind turbine condition identification under various operating conditions can be considered a unique meta-learning task. Constrained by practical engineering conditions, multiple training tasks from different data distribution domains can be constructed based on limited data. The identification model trained through meta-learning can be applied to wind turbine condition identification tasks under different operating conditions, rather than being limited to condition identification tasks from a single data distribution domain. Furthermore, when faced with previously unseen condition identification tasks, the identification model only requires a small amount of data for adaptation to achieve satisfactory performance.

[0054] In one possible implementation, obtaining an initialized set of parameter spaces includes: initializing each parameter space in the set using Gram-Schmidt orthogonalization to obtain multiple parameter spaces. Gram-Schmidt orthogonalization is a method in mathematics and numerical analysis used to construct a set of orthogonal vectors from a set of linearly independent vectors. This process can be applied to any inner product space, but is most commonly used in Euclidean space (i.e., a space of real or complex vectors with a standard inner product). Vectors obtained through Gram-Schmidt orthogonalization are not only mutually orthogonal (i.e., their inner product is zero), but can also be further normalized so that each vector has a unit length, thus forming an orthogonal basis. Multiple randomly initialized first parameter spaces can be obtained first, and these first parameter spaces can be initialized using Gram-Schmidt orthogonalization to obtain multiple parameter spaces.

[0055] The support set can be used to update the parameter space of the recognition model, allowing it to quickly adapt to new tasks. After rapid adjustment based on the support set, the recognition model can be used to classify data on the query set, thereby evaluating its performance.

[0056] Wind turbine parameter signals can include parameters such as current, speed, vibration, or load. Wind turbine status can include normal operation or fault status. Fault status can include stator winding faults, rotor winding faults, demagnetization faults, etc. The characteristics of wind turbine parameter signals differ under each fault condition; therefore, the wind turbine status can be determined by identifying these parameter signals. Source domain training samples can include relevant parameters for wind turbine operation under no-load and relevant parameters for operation under load.

[0057] It should be noted that the executing entity of this application can be any electronic device with computing capabilities, such as a server. This electronic device can be deployed locally or in the cloud; this application does not limit its deployment.

[0058] Step S120: Perform the iterative training step.

[0059] The iterative training steps include: sampling multiple batches of target training samples from the source domain training samples based on a preset number of support set samples and query set samples, with each batch of target training samples corresponding to different operating conditions of the wind turbine; iteratively training multiple parameter spaces of the recognition model based on multiple batches of target training samples; in the inner loop of the iterative training, determining the loss of each parameter space in the support set, normalizing each loss to obtain a combination coefficient, combining each initialized parameter space into a joint parameter space based on the combination coefficient, and performing gradient update on the joint parameter space based on the support set to obtain the gradient-updated parameter space; in the outer loop of the iterative training, determining the loss of the gradient-updated parameter space in the query set to obtain the first loss, determining the second loss based on the current training round number and the first loss, determining the sum of the second losses of each batch of target training samples to obtain the third loss, and updating the parameter space in the parameter space set based on the third loss in a polling manner to obtain the parameter space updated in the current iteration.

[0060] In one embodiment of this application, each initialized parameter space can be sequentially loaded into the recognition model, and the loss corresponding to each support set can be determined based on a preset loss function. The gradient update parameter space can be loaded into the recognition model to determine the loss of the gradient update parameter space in the query set.

[0061] It should be noted that when training and updating the parameter space, the parameter space can be loaded into the network architecture of the recognition model, and then the parameters can be updated on the task. Each batch of target training samples corresponds to different operating conditions of the wind turbine. For example, batch one includes the parameter signals of each wind turbine under light load and high speed conditions, and batch two includes the parameter signals of each wind turbine under high load and high speed conditions.

[0062] The inner loop refers to the rapid adaptation process for a specific task. In this process, the model performs multiple gradient updates based on a support set for a given task to adjust its parameter space and better suit the task. The goal of the inner loop is to find a set of parameters that allows the model to perform well on the task with a small number of training samples by minimizing the loss function on the support set. The outer loop refers to the meta-learning process, in which the model adjusts its initial parameter space based on the average performance across all tasks. The aim is to find multiple sets of initial parameters so that the model can quickly adapt to any new task through a small number of updates in the inner loop.

[0063] In one possible implementation, normalizing each loss to obtain combination coefficients includes: normalizing the loss of each parameter space in the parameter space set based on a nonlinear activation function to obtain multiple spatial coefficients, and determining the set of multiple spatial coefficients as the combination coefficients. The nonlinear activation function can be the Softmax function.

[0064] For example, an initial parameter space set can be defined. N is the number of parameter spaces, θ (n) The parameter space is initialized for the nth parameter. Normalizing the loss of each parameter space in the parameter space set based on a nonlinear activation function yields multiple spatial coefficients, and the process of determining the set of these multiple spatial coefficients as the combined coefficients can include:

[0065]

[0066] in, Let n be the spatial coefficients when using the target training samples of the i-th batch. The loss is determined using the target training samples from the i-th batch when the model is loaded into the n-th parameter space. The combination coefficients corresponding to the target training samples from the i-th batch can be characterized as...

[0067] In one possible implementation, the process of combining the initialized parameter spaces into a joint parameter space based on the combination coefficients may include:

[0068]

[0069] in, For the joint parameter space.

[0070] In one possible implementation, the gradient update parameter space is obtained by performing gradient update on the joint parameter space based on the support set, including: loading the joint parameter space into the recognition model to determine the fourth loss of the joint parameter space on the support set; and performing gradient update based on the joint parameter space, the learning rate of the preset base learner, and the fourth loss to obtain the gradient update parameter space.

[0071] For example, the process of obtaining the gradient update parameter space may include:

[0072]

[0073] in, S is the gradient update parameter space corresponding to the target training samples in the i-th batch. i This represents the support set for the i-th batch of target training samples. The fourth loss is α, which is the learning rate of the preset base learner.

[0074] In one possible implementation, determining the second loss based on the current training epoch and the first loss includes: when the current training epoch is an integer multiple of a preset regularization epoch, determining the dot product of each initial parameter space as the regularization loss, and determining the sum of the first loss and the regularization loss as the second loss, wherein the preset regularization epoch is less than the preset training epoch; when the current training epoch is not an integer multiple of the preset regularization epoch, determining the first loss as the second loss.

[0075] For example, the process of determining the second loss as the sum of the first loss and the regularization loss may include:

[0076]

[0077] Among them, Q i Let i be the query set of the target training samples for the i-th batch. The loss for updating the parameter space for gradients in the query set is also known as the first loss. For regularization loss, This is the second loss.

[0078] It should be noted that periodically calculating the regularization loss can enhance the diversity of the parameter space and improve the generalization ability, which helps to expand the domain adaptation space.

[0079] In one possible implementation, the parameter space in the parameter space set is updated in a polling manner based on the third loss, including: performing gradient updates in a polling manner based on the parameter space in the parameter space set, the learning rate of the preset meta-learner, and the third loss to obtain the parameter space updated in the current iteration.

[0080] For example, the sum of the second losses of each batch of target training samples can be used to determine the third loss, and this can be recorded. for The third loss, or cumulative total loss, is calculated for each batch of target training samples. The parameter space in the initialized parameter space set is then polled based on the cumulative total loss. The formula is updated each time:

[0081]

[0082] Where, θ (n) 'This represents the parameter space after the current iteration update. The third loss is β, which is the preset learning rate of the meta-learner.

[0083] Step S130: Stop training when the current training round number is equal to the preset training round number, and obtain the training completed target parameter space.

[0084] In one embodiment of this application, a preset number of training rounds can be set before iterative training, and training can be stopped when the current number of training rounds is equal to the preset number of training rounds, thereby obtaining multiple target parameter spaces after training is completed.

[0085] Step S140: Obtain training samples for the target domain.

[0086] In one embodiment of this application, the data distributions of the source domain and the target domain are often not entirely the same. Obtaining training samples from the target domain helps the identification model understand and adapt to the differences in data distribution, ensuring that the identification model remains effective in new environments. The training samples from the target domain can be parameter signals corresponding to the wind turbine for which state identification needs to be performed. The parameter signals can include parameters such as the wind turbine's current, speed, vibration, and load.

[0087] Step S150: Determine the target recognition model based on multiple target parameter spaces and target domain training samples.

[0088] In one embodiment of this application, a target recognition model can be determined based on multiple target parameter spaces and target domain training samples.

[0089] In one embodiment, step S150, which determines the target recognition model based on multiple target parameter spaces and the target domain training samples, may include steps S151 to S155.

[0090] Step S151: Determine the loss value for each target parameter space based on the training samples in the target domain.

[0091] In one embodiment of this application, the loss value for each target parameter space on the support set of training samples in the target domain can be determined, thereby obtaining a set of loss values ​​corresponding to all target parameter spaces.

[0092] Step S152: Normalize the loss values ​​of each objective parameter space to obtain the objective combination coefficients.

[0093] In one embodiment of this application, the loss value of each target parameter space can be normalized by applying the Softmax nonlinear activation function to obtain the target combination coefficients.

[0094]

[0095] Where, σ (n) This refers to the nth spatial coefficient when training samples using the target domain. The loss value is determined when the model is loaded with the parameter space of the nth objective. The objective combination coefficients corresponding to the training samples in the objective domain can be characterized as...

[0096] Step S153: Combine the target parameter spaces into a target joint parameter space based on the target combination coefficients.

[0097] In one embodiment of this application, each target parameter space can be combined into a joint parameter space based on the target combination coefficients.

[0098]

[0099] Where, θ ＊ For the joint parameter space of the objective, θ (n)＊ Let n be the target parameter space.

[0100] Step S154: Based on the training samples in the target domain, perform gradient update on the joint parameter space of the target to obtain the parameter space of the target model.

[0101] In one embodiment of this application, the target model parameter space can be obtained by gradient updating of the target joint parameter space based on the support set of the target domain training samples.

[0102]

[0103] Where, φ ★ Let S be the parameter space of the target model, and S be the support set of the training samples in the target domain. Let α be the loss of the joint parameter space of the target domain on the support set of the training samples in the target domain, and let α be the learning rate of the preset base learner.

[0104] Step S155: Load the target model parameter space into the recognition model to obtain the target recognition model.

[0105] In one embodiment of this application, the target model parameter space can be loaded into the recognition model to obtain the target recognition model.

[0106] For example, please refer to Figure 2 This is a schematic diagram illustrating the training logic of the inner and outer loops in an exemplary embodiment of this application. Solid lines represent the inner loop, dotted dashed lines represent the outer loop, and dotted dashed lines represent the updates of each parameter space to the joint parameter space. When facing wind turbine state recognition tasks, the data distribution differences between the source and target domains often pose a challenge. In traditional model-independent meta-learning methods, two-layer updates are applied only to a single parameter space, which may lead to insufficient task space coverage and cause optimization to remain at a local optimum. The multi-parameter space exploration and expansion of task space coverage provided in this application embodiment is an effective solution to improve the model's generalization ability.

[0107] The recognition model training method provided in this application also follows the meta-learning task framework, allowing training in scenarios with limited data and multiple source domains. By utilizing multiple mutually orthogonal initialized parameter spaces, the coverage of the task space is expanded. In a specific task, this application calculates the loss in each model parameter space and applies a non-linear activation function to the loss results to determine the combined weights of each parameter space, thereby forming a joint parameter space for meta-level updates. The joint parameter space is updated in the inner loop. In the outer loop, each initial parameter space is updated sequentially based on the updated gradient and the performance of the parameter space on the task query set. During the update process, the scheme in this application introduces the dot product of each parameter space as a regularization term to enhance the diversity of the parameter space, thereby helping to expand the domain adaptation space.

[0108] For example, please refer to Figure 3 This is a flowchart illustrating the source domain training phase of a wind turbine state recognition model, as shown in an exemplary embodiment of this application. c represents the current training round number.

[0109] For example, please refer to Figure 4 This is a flowchart illustrating the target domain training phase of a wind turbine state recognition model, as shown in an exemplary embodiment of this application. Multiple target parameter spaces constitute a target parameter space set.

[0110] Figure 5 This is a flowchart illustrating a wind turbine state identification method according to an exemplary embodiment of this application. Figure 5 As shown, this exemplary wind turbine status identification method may include:

[0111] Step S510: Obtain the parameters of the wind turbine to be identified.

[0112] In one embodiment of this application, parameters of the wind turbine to be identified can be obtained. These parameters may include signals such as current, speed, vibration, and load of the wind turbine.

[0113] Step S520: Input the parameters of the wind turbine to be identified into the target identification model to obtain the wind turbine status.

[0114] The target recognition model is a target recognition model trained using the training method provided in any of the above embodiments.

[0115] It should be noted that a target recognition model can be used to identify the state of a wind turbine, and the target model parameter space has a better recognition effect than any target parameter space.

[0116] Embodiments of this application also provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by one or more processors, cause the electronic device to implement the methods provided in the above embodiments.

[0117] Another aspect of this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a computer's processor, causes the computer to perform the methods provided in the various embodiments described above. This computer-readable storage medium may be included in the electronic devices described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0118] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various embodiments described above.

[0119] In the embodiments of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The terms "comprising" and "including" as used throughout the specification and claims are open-ended terms and should therefore be interpreted as "comprising but not limited to".

[0120] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for training a wind turbine state recognition model, characterized in that, include: Obtain the recognition model to be trained, obtain the source domain training samples for training the recognition model, obtain the initialized parameter space set, the parameter space set includes multiple parameter spaces of the recognition model, the source domain training samples include a support set and a query set, the source domain training samples include multiple wind turbine parameter signals, the wind turbine parameter signals include the wind turbine current, speed, vibration or load, and the wind turbine state is marked as a label in the source domain training samples, the wind turbine state includes normal operation state or fault state; The iterative training process includes: sampling multiple batches of target training samples from the source domain training samples based on a preset number of support set samples and a preset number of query set samples, with each batch of target training samples corresponding to different operating conditions of the wind turbine; iteratively training multiple parameter spaces of the recognition model based on the multiple batches of target training samples; in the inner loop of the iterative training, determining the loss of each parameter space in the parameter space set in the support set, normalizing each loss to obtain a combination coefficient, combining each initialized parameter space into a joint parameter space based on the combination coefficient, and performing gradient update on the joint parameter space based on the support set to obtain a gradient-updated parameter space; in the outer loop of the iterative training, determining the loss of the gradient-updated parameter space in the query set to obtain a first loss, determining a second loss based on the current training round number and the first loss, determining the sum of the second losses of each batch of target training samples to obtain a third loss, and updating the parameter space in the parameter space set in a polling manner based on the third loss to obtain the parameter space updated in the current iteration. Training stops when the current training round number equals the preset training round number, resulting in a space of multiple target parameters after training is complete. Obtain training samples for the target domain; Based on multiple target parameter spaces and the target domain training samples, a target recognition model is determined; Obtain the parameters of the wind turbine to be identified; The wind turbine parameters to be identified are input into the target identification model to obtain the wind turbine status.

2. The wind turbine state recognition model training method according to claim 1, characterized in that, Obtain the initial parameter space set, including: The parameter spaces in the parameter space set are initialized using the Schmitt orthogonalization method to obtain multiple parameter spaces.

3. The wind turbine state recognition model training method according to claim 1, characterized in that, Normalizing each loss yields the combination coefficients, including: The loss of each parameter space in the parameter space set is normalized based on the nonlinear activation function to obtain multiple spatial coefficients, and the set of multiple spatial coefficients is determined as the combined coefficients.

4. The wind turbine state recognition model training method according to claim 1, characterized in that, The gradient update parameter space is obtained by performing gradient update on the joint parameter space based on the support set, including: The joint parameter space is loaded into the recognition model to determine the fourth loss of the joint parameter space on the support set; Based on the joint parameter space, the learning rate of the preset base learner, and the fourth loss, gradient update is performed to obtain the gradient update parameter space.

5. The wind turbine state recognition model training method according to claim 1, characterized in that, The second loss is determined based on the current training epoch and the first loss, including: When the current training round number is an integer multiple of the preset regularization round number, the dot product of each initial parameter space is determined as the regularization loss, and the sum of the first loss and the regularization loss is determined as the second loss, wherein the preset regularization round number is less than the preset training round number; When the current training round number is not an integer multiple of the preset regularization round number, the first loss is determined as the second loss.

6. The wind turbine state recognition model training method according to claim 1, characterized in that, Based on multiple target parameter spaces and training samples of the target domain, a target recognition model is determined, including: The loss value for each target parameter space is determined based on the training samples in the target domain; The loss values ​​of each objective parameter space are normalized to obtain the objective combination coefficients; Based on the target combination coefficients, the target parameter spaces are combined into a target joint parameter space; The target model parameter space is obtained by performing gradient update on the target joint parameter space based on the target domain training samples; The target model parameter space is loaded into the recognition model to obtain the target recognition model.

7. The wind turbine state recognition model training method according to claim 1, characterized in that, The parameter space in the parameter space set is updated based on the third loss in a polling manner, including: The parameter space is updated in the current iteration by using a polling method based on the parameter space in the parameter space set, the learning rate of the preset meta-learner, and the third loss.

8. An electronic device, characterized in that, The electronic device includes: One or more processors; Memory used to store the executable program code of the processor; The processor is configured to execute the program code to implement the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the computer's processor, causes the computer to perform the method described in any one of claims 1 to 7.

Images