Wind turbine yaw fault detection method, device and electronic equipment

Abstract

The present application proposes a method, device and electronic equipment for detecting yaw faults of a wind turbine, wherein the method for detecting yaw faults of a wind turbine comprises: obtaining one or more pre-trained first fault detection models; determining the precision and recall of the first fault detection model based on the detection results of each first fault detection model, and selecting a second fault detection model from the first fault detection model according to the precision and recall; migrating the second fault detection model to a new model for learning to obtain a target fault detection model; and performing yaw fault detection on the wind turbine according to the target fault detection model to solve the technical problem in the prior art of inaccurate yaw fault detection due to too little sample data.

Description

Technical Field

[0001] The present application relates to the technical field of fault detection, and in particular to a method, device and electronic equipment for detecting yaw faults of a wind turbine. Background Art

[0002] The yaw system of a wind turbine is an important component of the wind turbine. When the yaw system fails, it will have a great impact on the normal operation of the wind turbine. Currently, the yaw fault of the wind turbine can be detected based on the pre-trained LSTM neural network. However, the premise of accurate prediction of the LSTM neural network is based on sufficient training samples for training and learning. When there are few training samples or no labeled samples, the detection effect of the yaw fault of the wind turbine is poor. Summary of the invention

[0003] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0004] To this end, the first objective of the present application is to propose a method for detecting a yaw fault of a wind turbine generator system, so as to achieve accurate yaw fault detection.

[0005] The second objective of the present application is to provide a wind turbine yaw fault detection device.

[0006] The third objective of the present application is to provide an electronic device.

[0007] A fourth objective of the present application is to provide a computer-readable storage medium.

[0008] A fifth object of the present application is to provide a computer program product.

[0009] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a method for detecting a yaw fault of a wind turbine, comprising:

[0010] Obtain one or more pre-trained first fault detection models, wherein hyperparameters of the plurality of first fault detection models are different;

[0011] Based on the detection results of each of the first fault detection models, determine the precision and recall of the first fault detection model, and select a second fault detection model from the first fault detection models according to the precision and the recall;

[0012] Migrating the second fault detection model to a new model for learning to obtain a target fault detection model;

[0013] The yaw fault detection of the wind turbine is performed according to the target fault detection model.

[0014] To achieve the above-mentioned purpose, the second embodiment of the present application proposes a wind turbine yaw fault detection device, comprising:

[0015] A first acquisition module is used to acquire one or more pre-trained first fault detection models, wherein the hyperparameters of the multiple first fault detection models are different;

[0016] a second acquisition module, configured to determine the precision and recall of the first fault detection model based on the detection results of each of the first fault detection models, and select a second fault detection model from the first fault detection model according to the precision and the recall;

[0017] A third acquisition module is used to transfer the second fault detection model to a new model for learning to obtain a target fault detection model;

[0018] The fault detection module is used to perform yaw fault detection on the wind turbine according to the target fault detection model.

[0019] To achieve the above-mentioned purpose, the third aspect of the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0020] The memory stores computer-executable instructions;

[0021] The processor executes the computer-executable instructions stored in the memory to implement the method as described in the embodiment of the first aspect.

[0022] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present application proposes a computer-readable storage medium, in which computer execution instructions are stored. When the computer execution instructions are executed by a processor, they are used to implement the method described in the first aspect embodiment.

[0023] To achieve the above-mentioned purpose, the fifth aspect of the present application proposes a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0024] The wind turbine yaw fault detection method, device and electronic equipment provided in the present application calculate the precision and recall rate of multiple first fault detection models, thereby determining a second fault detection model based on the precision and recall rate, obtaining a target fault detection model by performing transfer learning on the second fault detection model, and performing yaw fault detection based on the target fault detection model, thereby improving detection accuracy when data is limited or lacking, achieving high-precision fault detection, and reducing maintenance costs.

[0025] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0027] Figure 1 A schematic diagram of a flow chart of a method for detecting a yaw fault of a wind turbine provided in an embodiment of the present application;

[0028] Figure 2 A schematic flow chart of another method for detecting yaw faults of a wind turbine provided in an embodiment of the present application.

[0029] Figure 3 A logical schematic diagram of a wind turbine yaw fault detection method provided in an embodiment of the present application.

[0030] Figure 4 A schematic structural diagram of a wind turbine yaw fault detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0032] The following describes a method, device and electronic device for detecting yaw faults of a wind turbine generator system according to an embodiment of the present application with reference to the accompanying drawings.

[0033] Figure 1 The following is a flow chart of a method for detecting yaw faults of a wind turbine provided in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0034] S101, obtaining one or more pre-trained first fault detection models.

[0035] In some implementations, the first fault detection model may be a conventional network model, such as a long short-term memory network model (LSTM), and the LSTM network is pre-trained using training data to obtain a pre-trained first fault detection model; the training data may be operating data of other units.

[0036] In some implementations, the initial LSTM model can be trained in different rounds based on the training data, each round corresponding to a first fault detection model, wherein the hyperparameters of the multiple first fault detection models are different, and the hyperparameter combinations of different training rounds are different; it can be understood that when the hyperparameter combinations of the models are different, the training results of the corresponding models are very different, so multiple first fault detection models can be trained through multiple sets of different hyperparameter combinations, but the basic model structures of the multiple first fault detection models are the same.

[0037] In some implementations, the first fault detection model can be used to perform fault detection. For example, the data to be detected is input into the first fault detection model, and the first fault detection model extracts and analyzes features of the data to be detected, outputs detection results of the data to be detected, and determines whether a fault occurs.

[0038] S102, based on the detection results of each first fault detection model, determine the precision and recall of the first fault detection model, and select a second fault detection model from the first fault detection model according to the precision and recall.

[0039] In some implementations, a test sample may be obtained and input into each first fault detection model respectively to obtain a detection result of each first fault detection model, and the detection result may include whether the current test sample has a fault or not.

[0040] In some implementations, the precision rate is used to reflect the ability of the model to correctly predict the accuracy of positive samples, that is, the proportion of real positive samples in the predicted positive samples. In this embodiment, the positive samples can be normal samples without faults. The precision rate is the ability of the first fault detection model to predict the accuracy of normal samples. For example, the ratio of the number of normal samples predicted correctly to the number of all samples predicted as normal samples is taken as the precision rate. The larger the precision rate, the better the model's prediction ability.

[0041] In some implementations, the recall rate is also called the recall rate or true positive rate, which is used to reflect the algorithm's ability to correctly predict the full range of positive samples, that is, the proportion of positive samples predicted to be positive samples to the total positive samples. For example, the ratio of the number of normal samples predicted correctly to the total number of normal samples is taken as the recall rate. The larger the recall rate, the better the model's prediction ability.

[0042] Optionally, the precision and the recall may be fused, for example, the precision and the recall are weightedly summed to obtain an evaluation score, and the first fault detection model with the highest evaluation score is selected as the second fault detection model.

[0043] S103, migrating the second fault detection model to the new model for learning to obtain a target fault detection model.

[0044] Transfer learning refers to using existing knowledge or existing models to try to solve problems in different fields. It can even use existing knowledge to solve learning problems with only a small number of or no labeled samples in the target sample. Therefore, for samples that lack labels or have only a small amount of data, you can use approximate samples to pre-train the model and transfer the pre-trained model for use.

[0045] Therefore, in this embodiment, the second fault detection model that has been pre-trained and has the best prediction effect can be migrated to the new model for learning, the hyperparameters and model parameters of the new model can be updated to the second fault detection model, and the target fault detection model can be determined. The target fault detection model can achieve the same prediction effect as the second fault detection model.

[0046] S104, performing yaw fault detection on the wind turbine generator set according to the target fault detection model.

[0047] In some cases, the real-time operation data of the wind turbine is input into the target fault detection model, and the target fault detection model outputs whether the current wind turbine has a yaw fault. Optionally, the current wind turbine may be a unit with less operation data, that is, it is impossible to determine the fault detection model corresponding to the wind turbine according to the operation data of the current wind turbine. Therefore, the target fault detection model of transfer learning is used to perform yaw fault detection on the current wind turbine to ensure the accuracy of fault detection.

[0048] In this embodiment, multiple pre-trained first fault detection models are obtained, and the second fault detection model with the best prediction effect is selected according to the precision and recall rate of the first fault detection model. The target fault detection model is obtained by performing transfer learning on the second fault detection model. Yaw fault detection is performed based on the target fault detection model, which improves detection accuracy when data is limited or lacking, achieves high-precision fault detection, and reduces maintenance costs.

[0049] Figure 2 This is a flow chart of another method for detecting yaw faults of wind turbines provided in an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0050] S201, obtaining one or more pre-trained first fault detection models.

[0051] In some implementations, a first data set may be obtained, and training features may be determined from the first data set; training features refer to input variables, that is, variables related to the prediction results. For example, to predict a yaw system failure, the variables related to the prediction results are determined to be wind direction and wind speed, and the training features are determined from the first data set to be wind direction and wind speed. Optionally, the first data set may be the operating data of any wind turbine in a wind farm, and the data in the first data set is relatively rich and can be used to train a fault detection model.

[0052] Determine the training data according to the training features, and preprocess the training data; that is, select the wind direction and wind speed data from the first data set as the training data; optionally, the preprocessing of the training data may include outlier removal, redundant data removal, and data balancing and standardization; for example, remove outliers based on the boxplot function, remove redundant data based on the similarity discrimination method, balance the training data set based on the oversampling technology, so that the normal samples and abnormal samples in the training data are 1:1, and further map all the feature data to the range of [0,1] based on the minimum-maximum value Min-Max standardization method.

[0053] Furthermore, an initial LSTM model is trained according to the preprocessed training data, where the training data may include training samples and label samples, which are decomposed into a training set, a test set, and a validation set in a ratio of 8:1:1. For each training round, a corresponding first fault detection model can be obtained, and the first fault detection model of each round is analyzed.

[0054] In the embodiment of the present application, the implementation method of step S201 can be implemented by any of the methods in the embodiments of the present disclosure, which is not limited here and will not be repeated.

[0055] S202, comparing the detection result with the corresponding label result to obtain a matching result.

[0056] It can be understood that the label result refers to the standard result, that is, the actual result corresponding to each test sample. The detection result is the result of the test sample data input into the first fault detection model output. The detection result is compared with the corresponding label result to obtain the matching result of each test sample. The matching result includes at least the detection correctness of each test sample. For example, the matching result includes whether the detection result of the test sample is consistent with the label result.

[0057] S203: Determine a first fault detection precision and recall rate according to the recognition results of the normal samples and the recognition results of the abnormal samples in the matching results.

[0058] It can be understood that the normal samples are actually normal data samples in the training data, and the abnormal samples are actually faulty data samples in the training data. The normal samples and the abnormal samples are sampled from the first data set, and the ratio of the normal samples to the abnormal samples is 1:1.

[0059] Optionally, based on the recognition results of normal samples, the number of normal samples that are correctly identified can be determined as a first number, that is, the number of positive samples that are correctly identified; based on the recognition results of abnormal samples, the number of abnormal samples that are correctly identified can be determined as a second number, that is, the number of negative samples that are correctly identified, and the number of abnormal samples that are incorrectly identified can be determined as a third number, that is, the number of falsely reported negative samples.

[0060] The precision is obtained based on the first number and the third number; the recall is obtained based on the first number and the second number.

[0061] Optionally, the accuracy can be calculated as:

[0062]

[0063] Among them, precision is the accuracy; TP is the first quantity; FP is the third quantity.

[0064] Optionally, the recall rate can be calculated as:

[0065]

[0066] Among them, recall is the recall rate; FN is the second number.

[0067] S204: Determine an evaluation score according to the precision and recall.

[0068] Optionally, the evaluation score can be expressed as:

[0069]

[0070] Among them, F1-Score is the evaluation score; precision is the accuracy; recall is the recall rate.

[0071] S205 , obtaining a first fault detection model having an evaluation score greater than or equal to a first preset value as a candidate fault detection model.

[0072] It can be understood that the evaluation score of each first fault detection model can represent its prediction ability. The higher the evaluation score, the better the prediction effect. Therefore, the first fault detection model with an evaluation score greater than or equal to the first preset value can be selected, that is, the first fault detection model with better prediction effect can be selected, and the selected first fault detection model with better prediction effect can be used as a candidate fault detection model.

[0073] S206, obtaining the return rate of each candidate fault detection model, and determining a second fault detection model among the candidate fault detection models according to the return rate.

[0074] In some implementations, a fourth number of normal samples and a fifth number of abnormal samples in the detection results, a sixth number of normal samples and a seventh number of abnormal samples in the label results can be obtained; and a return rate can be obtained based on the fourth number, the fifth number, the sixth number, and the seventh number.

[0075] It can be understood that the fourth number of normal samples in the test results, that is, the number of normal test results, and the fifth number of abnormal samples in the test results, that is, the number of faulty test results; the sixth number of normal samples and the seventh number of abnormal samples in the label results refer to the number of samples whose actual values ​​are normal and the number of samples whose actual values ​​are abnormal in the label samples.

[0076] Optionally, the rate of return can be calculated as:

[0077]

[0078] Among them, P reward represents the rate of return; N pre_0 Indicates the fourth number of normal samples in the test results; N sum_0 Indicates the sixth number of normal samples in the label result; N pre_1 Indicates the fifth number of abnormal samples in the test results; N sum_1 Indicates the seventh number of abnormal samples in the label results.

[0079] Furthermore, the return rates of all candidate fault detection models are obtained, and the second fault detection model is selected from the candidate fault detection models according to the return rates; it can be understood that during the training of the initialization model, by calculating the evaluation score of each first fault detection model, the first fault detection models with evaluation scores greater than a preset threshold are screened out as candidate fault detection models, and it is further determined whether the return rate of the candidate fault detection model in the current round is greater than that of the candidate fault detection model in the previous round, thereby determining the candidate fault detection model with the largest return rate as the second fault detection model.

[0080] It can be understood that if the evaluation score of the first fault detection model in the current round is less than the first preset value, that is, there is no candidate fault detection model in the current round, then the rate of return calculation can be omitted and the next round of training can be carried out directly to obtain the first fault detection model of the next round for analysis.

[0081] In some implementations, a training round threshold can also be set in advance to avoid long-term model training and increase the amount of calculation. When the training round reaches the training round threshold, the candidate fault detection model corresponding to the current training round is the model with the highest return rate, that is, the first fault detection model obtained by the current round of training is the optimal model, that is, the second fault detection model.

[0082] S207, migrating the second fault detection model to the new model for learning to obtain a target fault detection model.

[0083] In some implementations, the training features can be transferred for learning to obtain feature data in a second data set, where the second data set and the first data set are operating data of different wind turbines; that is, feature data corresponding to the training features are determined from the second data set, and the feature data is extracted from the second data set as current verification data. Optionally, the current feature data can also be preprocessed, such as de-redundancy and normalization, to reduce interference from abnormal data.

[0084] Furthermore, the feature data is input into the target fault detection model to obtain the output result, that is, the verification data is input into the target fault detection model to obtain the corresponding output result, that is, the prediction result of the target fault detection model for each verification sample; the migration performance of the target fault detection model is determined according to the output result.

[0085] Optionally, the accuracy of the target fault detection model in identifying the feature data can be obtained; the number of correct detections in the output results is obtained, and the ratio of the number of correct detections to the total number of input samples is calculated as the accuracy rate. The higher the accuracy rate, the better the prediction effect of the target fault detection model.

[0086] Optionally, the rate of return of the target fault detection model for identifying feature data can be obtained; the number of normal samples and the number of abnormal samples in the output results and the labels of the input samples are obtained respectively to determine the rate of return. The acquisition of the rate of return can refer to the calculation formula of step S206, which will not be repeated here.

[0087] Optionally, the false negative rate of the target fault detection model for the characteristic data identification can be obtained; the number of yaw faults in the output prediction results and the number of yaw faults actually existing in the input samples are obtained, and the ratio of the number of predicted yaw faults to the number of actual yaw faults is calculated, and the false negative rate is determined based on the ratio. The calculation of the false negative rate can be:

[0088]

[0089] Among them, P miss represents the underreporting rate; Indicates the number of yaw faults identified in the prediction results; Nfailure is the number of yaw faults actually existing in the sample.

[0090] Furthermore, a migration score of the target fault detection model is determined according to the accuracy, the return rate, and the false negative rate; and a migration performance of the target fault detection model is determined according to the migration score.

[0091] Alternatively, the migration performance can be expressed as:

[0092] P=α*P right +β*P reward -γ*P miss

[0093] Where P is the migration performance; P right Indicates accuracy; P reward represents the rate of return; P miss represents the false negative rate; α, β and γ are the weight coefficients of accuracy, return rate and false negative rate respectively, which are preset values ​​greater than zero and α+β+γ=1.

[0094] Based on the calculation formula of the above migration performance, the prediction effect of the target fault detection model on the feature data can be determined, that is, the prediction performance of the target fault detection model migrated to the current data set. The greater the migration performance, the more accurate the prediction and the better the migration effect. If the migration performance is low, the target fault detection model can be fine-tuned to achieve better prediction results and reduce the amount of calculation for retraining.

[0095] It can be understood that in actual scenarios, the second fault detection model with better prediction effect is selected according to the evaluation score, and the second fault detection model is transferred and learned to obtain the target fault detection model. The target fault detection model will learn the better prediction results of the second fault detection model, and its corresponding migration performance will also be greater.

[0096] In the embodiment of the present application, the implementation method of step S207 can be implemented by any of the methods in the embodiments of the present disclosure, which is not limited here and will not be repeated.

[0097] S208, performing yaw fault detection on the wind turbine generator set according to the target fault detection model.

[0098] In the embodiment of the present application, the implementation method of step S208 can be implemented by any of the methods in the embodiments of the present disclosure, which is not limited here and will not be repeated.

[0099] In this embodiment, multiple pre-trained first fault detection models are obtained, and evaluation scores are determined according to the precision and recall of the first fault detection models. Candidate fault detection models are selected based on the evaluation scores, and a second fault detection model is selected based on the return rate of each candidate fault detection model to ensure that the prediction effect of the second fault detection model of transfer learning is better. Transfer learning is further performed on the second fault detection model to obtain a target fault detection model, and the migration performance of the target fault detection model is determined based on the evaluation of the accuracy, return rate and omission rate. Yaw fault detection is performed based on the target fault detection model to improve detection accuracy when data is limited or lacking, achieve high-precision fault detection, and reduce maintenance costs.

[0100] Figure 3 The following is a logical diagram of a method for detecting a yaw fault of a wind turbine provided in an embodiment of the present application. Figure 3 As shown, data is read from unit A and preprocessed, the data is divided into a training set, a test set, and a validation set, the training features are determined, and data extraction and balanced sampling are performed based on the features to obtain training data, and the initially constructed LSTM network is trained. When the training rounds do not exceed the preset rounds, the neural network is trained based on the training data to obtain the first fault detection model of the current round, and the evaluation score of the first fault detection model is calculated. Based on the evaluation score, the candidate fault detection model is determined to calculate the rate of return, and it is determined whether the rate of return of the current candidate fault detection model is better than the rate of return of the previous round. If If yes, the next round of training is carried out until the highest return rate is determined or the training round is reached to obtain the best model. If the return rate of the current round is not better than that of the previous round, the candidate fault detection model of the previous round is determined to be the best model, that is, the second fault detection model. The best model is migrated to unit B. At the same time, the training features of unit A are migrated and the verification data is read from the data of unit B. The data is preprocessed and the accuracy, return rate and missed reporting rate are calculated to obtain the migration performance. Based on the migration performance, the comparison between the actual yaw fault result of unit B and the predicted result can be reflected.

[0101] In order to implement the above embodiment, the present application also proposes a wind turbine yaw fault detection device.

[0102] Figure 4 The following is a schematic diagram of the structure of a wind turbine yaw fault detection device provided in an embodiment of the present application. Figure 4 As shown, the wind turbine yaw fault detection device 400 includes:

[0103] A first acquisition module 401 is used to acquire one or more pre-trained first fault detection models, wherein the hyperparameters of the multiple first fault detection models are different;

[0104] A second acquisition module 402 is used to determine the precision and recall of the first fault detection model based on the detection results of each first fault detection model, and select a second fault detection model from the first fault detection model according to the precision and recall;

[0105] A third acquisition module 403 is used to transfer the second fault detection model to the new model for learning to obtain a target fault detection model;

[0106] The fault detection module 404 is used to perform yaw fault detection on the wind turbine according to the target fault detection model.

[0107] Furthermore, in a possible implementation of the embodiment of the present application, the second acquisition module 402 includes:

[0108] Compare the detection results with the corresponding label results to obtain the matching results;

[0109] The first fault detection precision and recall rate are determined according to the recognition results of the normal samples and the recognition results of the abnormal samples in the matching results.

[0110] Furthermore, in a possible implementation of the embodiment of the present application, the second acquisition module 402 includes:

[0111] According to the recognition results of the normal samples, determining the number of correctly recognized normal samples as a first number;

[0112] According to the recognition results of the abnormal samples, determining the number of abnormal samples that are correctly recognized as a second number, and the number of abnormal samples that are incorrectly recognized as a third number;

[0113] According to the first quantity and the third quantity, obtaining the accuracy rate;

[0114] According to the first quantity and the second quantity, the recall rate is obtained.

[0115] Furthermore, in a possible implementation of the embodiment of the present application, the second acquisition module 402 includes:

[0116] Determine the evaluation score based on precision and recall;

[0117] Acquire a first fault detection model having an evaluation score greater than or equal to a first preset value as a candidate fault detection model;

[0118] The return rate of each candidate fault detection model is obtained, and a second fault detection model among the candidate fault detection models is determined according to the return rate.

[0119] Furthermore, in a possible implementation of the embodiment of the present application, the second acquisition module 402 includes:

[0120] Obtaining a fourth number of normal samples and a fifth number of abnormal samples in the test results, a sixth number of normal samples and a seventh number of abnormal samples in the label results;

[0121] Based on the fourth quantity, the fifth quantity, the sixth quantity and the seventh quantity, the rate of return is obtained.

[0122] Furthermore, in a possible implementation of the embodiment of the present application, the first acquisition module 401 includes:

[0123] Obtaining a first data set, and determining training features from the first data set;

[0124] Determine the training data according to the training characteristics, and pre-process the training data;

[0125] The fault detection model is trained according to the preprocessed training data to obtain one or more first fault detection models.

[0126] Furthermore, in a possible implementation of the embodiment of the present application, the third acquisition module further includes:

[0127] Perform transfer learning on the training features to obtain feature data in a second data set, wherein the second data set and the first data set are operating data of different wind turbines;

[0128] Input the characteristic data into the target fault detection model to obtain the output result;

[0129] The migration performance of the target fault detection model is determined based on the output results.

[0130] Furthermore, in a possible implementation of the embodiment of the present application, the third acquisition module includes:

[0131] Obtain the accuracy of the target fault detection model in identifying feature data;

[0132] Obtain the return rate of the target fault detection model on the feature data identification;

[0133] Obtain the underreporting rate of the target fault detection model for feature data identification;

[0134] Determine the migration score of the target fault detection model based on the accuracy, return rate and false negative rate;

[0135] The migration performance of the target fault detection model is determined according to the migration score.

[0136] It should be noted that the above explanations and descriptions of the embodiment of the method for detecting a yaw fault of a wind turbine generator set are also applicable to the device for detecting a yaw fault of a wind turbine generator set of this embodiment, and will not be described in detail herein.

[0137] In an embodiment of the present application, multiple pre-trained first fault detection models are obtained, and a second fault detection model with the best prediction effect is selected according to the precision and recall rate of the first fault detection model. A target fault detection model is obtained by performing transfer learning on the second fault detection model. Yaw fault detection is performed based on the target fault detection model, so as to improve detection accuracy when data is limited or lacking, achieve high-precision fault detection, and reduce maintenance costs.

[0138] In order to implement the above embodiments, the present application also proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method provided by the above embodiments.

[0139] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0140] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0141] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0142] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0143] The present application is expected to provide an implementation scheme for users to selectively block the use or access of personal information data. That is, the present disclosure is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, the risk can be minimized by limiting data collection and deleting the data. In addition, when applicable, such personal information is de-identified to protect the privacy of the user.

[0144] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0145] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0146] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0147] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0148] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0149] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0150] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0151] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for detecting yaw fault of a wind turbine, characterized in that: The following steps are involved: Acquire a plurality of pre-trained first fault detection models, wherein the plurality of first fault detection models have different hyperparameters; Based on the detection results of each of the first fault detection models, determine the precision and recall of the first fault detection model, and select a second fault detection model from the first fault detection models according to the precision and the recall; Migrating the second fault detection model to a new model for learning to obtain a target fault detection model; Performing yaw fault detection on the wind turbine according to the target fault detection model; The obtaining of a plurality of pre-trained first fault detection models comprises: Acquire a first data set, and determine training features from the first data set; Determining training data according to the training features, and preprocessing the training data; Training a fault detection model according to the preprocessed training data to obtain a plurality of first fault detection models, wherein the fault detection model is trained in different rounds based on the training data, each round corresponds to a first fault detection model, wherein hyperparameter combinations of different training rounds are different, and the basic model structures of the plurality of first fault detection models are the same; After the second fault detection model is transferred to the new model for learning to obtain the target fault detection model, the method further includes: Performing transfer learning on the training features to obtain feature data in a second data set, wherein the second data set and the first data set are operating data of different wind turbines; Inputting the characteristic data into the target fault detection model to obtain an output result; The migration performance of the target fault detection model is determined according to the output result.

2. The method according to claim 1, characterized in that The determining, based on the detection result of each of the first fault detection models, the precision and recall of the first fault detection model comprises: Compare the detection result with the corresponding label result to obtain a matching result; The precision and recall of the first fault detection model are determined according to the recognition results of the normal samples and the recognition results of the abnormal samples in the matching results.

3. The method according to claim 2, characterized in that Determining the precision and recall of the first fault detection model according to the recognition results of the normal samples and the recognition results of the abnormal samples in the matching results includes: According to the recognition results of the normal samples, determining the number of correctly recognized normal samples as a first number; According to the recognition results of the abnormal samples, determining the number of abnormal samples that are correctly recognized as a second number, and the number of abnormal samples that are incorrectly recognized as a third number; Obtaining the accuracy rate according to the first number and the third number; The recall rate is obtained according to the first number and the second number.

4. The method according to claim 3, characterized in that The selecting a second fault detection model from the first fault detection model according to the precision and the recall rate includes: Determining an evaluation score according to the precision and the recall; Acquire a first fault detection model having an evaluation score greater than or equal to a first preset value as a candidate fault detection model; The return rate of each of the candidate fault detection models is obtained, and a second fault detection model among the candidate fault detection models is determined according to the return rate.

5. The method according to claim 4, characterized in that The obtaining of the rate of return of each candidate fault detection model comprises: Obtaining a fourth number of normal samples and a fifth number of abnormal samples in the test results, a sixth number of normal samples and a seventh number of abnormal samples in the label results; The rate of return is obtained according to the fourth quantity, the fifth quantity, the sixth quantity and the seventh quantity.

6. The method according to claim 1, characterized in that The determining the migration performance of the target fault detection model according to the output result includes: Obtaining the accuracy of the target fault detection model in identifying the characteristic data; Obtaining a rate of return of the target fault detection model for identifying the feature data; Obtaining an underreporting rate of the target fault detection model for identifying the characteristic data; Determining a migration score of the target fault detection model according to the accuracy rate, the return rate, and the false negative rate; The migration performance of the target fault detection model is determined according to the migration score.

7. A wind turbine yaw fault detection device, characterized in that: include: A first acquisition module is used to acquire one or more pre-trained first fault detection models, wherein the hyperparameters of the multiple first fault detection models are different; a second acquisition module, configured to determine the precision and recall of the first fault detection model based on the detection results of each of the first fault detection models, and select a second fault detection model from the first fault detection model according to the precision and the recall; A third acquisition module is used to transfer the second fault detection model to a new model for learning to obtain a target fault detection model; A fault detection module, used for performing yaw fault detection on the wind turbine according to the target fault detection model; The first acquisition module is further used to acquire a first data set and determine training features from the first data set; Determining training data according to the training features, and preprocessing the training data; Training a fault detection model according to the preprocessed training data to obtain a plurality of first fault detection models, wherein the fault detection model is trained in different rounds based on the training data, each round corresponds to a first fault detection model, wherein hyperparameter combinations of different training rounds are different, and the basic model structures of the plurality of first fault detection models are the same; The third acquisition module is used to perform transfer learning on the training features to obtain feature data in a second data set, wherein the second data set and the first data set are operating data of different wind turbines; Inputting the characteristic data into the target fault detection model to obtain an output result; The migration performance of the target fault detection model is determined according to the output result.

8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 6.

Images