Wind turbine generator fault early warning method, device, equipment and storage medium

Abstract

Embodiments of the present disclosure relate to a wind turbine fault early warning method, device, equipment and storage medium, the method comprising: using a shared layer in a multi-task prediction model to perform feature extraction on an input vector to obtain shared features, and copying the shared features to obtain shared features corresponding to a plurality of target variables respectively; for the shared features corresponding to each target variable, using a sub-task feature layer in the multi-task prediction model corresponding to the shared features to perform feature extraction to obtain specific features matching the corresponding target variable; for the specific features matching each target variable, using a fully connected layer in the multi-task prediction model corresponding to the specific features to perform mapping to obtain a predicted value of the corresponding target variable in a current period; obtaining measurement values of the plurality of target variables in the current period, and performing fault prediction on the wind turbine based on the measurement values and the predicted values of the plurality of target variables in the current period, and issuing an alarm when a fault is predicted.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method, apparatus, device, and storage medium for early warning of wind turbine faults. Background Technology

[0002] With the rapid development of large-scale wind turbines, the single-unit capacity of wind turbines is increasing, and the losses to wind farms due to failures are also growing. In the era of grid parity for wind power, the safe and stable operation of wind turbines is particularly crucial. Due to the complex operating environment and variable operating conditions, wind turbines are prone to deterioration and failures, especially offshore wind turbines. These offshore turbines have large installed capacities, high utilization hours, high average wind speeds, and operate in environments with high humidity and salinity, making them more susceptible to failures and incurring higher maintenance costs, resulting in significant economic losses for wind farms. Therefore, wind turbine fault early warning systems are becoming increasingly important. Fault early warning systems allow staff to promptly inspect and repair wind turbines, effectively reducing losses caused by wind turbine downtime.

[0003] Currently, common methods for wind turbine fault early warning mainly involve physical modeling and data-driven approaches. Physical modeling refers to constructing an accurate mathematical model by thoroughly understanding the physical structure and operating principles of wind turbines. However, the physical structure of wind turbines is complex, making it difficult to establish an accurate physical model. Compared to physical modeling methods, data-driven methods not only avoid the tedious process of model building but also possess greater universality and flexibility, although their fault early warning accuracy is lower. Therefore, there is an urgent need for a simple and accurate new method for wind turbine fault early warning. Summary of the Invention

[0004] To solve the above-mentioned technical problems, or at least partially solve them, this disclosure provides a method, apparatus, device, and storage medium for early warning of wind turbine faults.

[0005] A first aspect of this disclosure provides a wind turbine fault early warning method, the method comprising:

[0006] The measurement values ​​of relevant variables in the current period are obtained and preprocessed to obtain an input vector. The relevant variables are correlated with multiple target variables, and the relevant variables and the multiple target variables are all monitoring variables of wind turbine units.

[0007] By utilizing the shared layer in the multi-task prediction model, feature extraction is performed on the input vector to obtain shared features, and the shared features are copied to obtain shared features corresponding to the multiple target variables respectively;

[0008] For each target variable, the shared features corresponding to the target variable are extracted using the sub-task feature layer in the multi-task prediction model to obtain specific features that match the corresponding target variable.

[0009] For each target variable, a specific feature that matches it is used to map it using the fully connected layer in the multi-task prediction model to obtain the predicted value of the target variable in the current period.

[0010] The system acquires the measured values ​​of the multiple target variables in the current period, performs fault prediction on the wind turbine based on the measured and predicted values ​​of the multiple target variables in the current period, and issues an alarm when a fault is predicted.

[0011] A second aspect of this disclosure provides a wind turbine fault early warning device, the device comprising:

[0012] The first acquisition module is used to acquire the measurement values ​​of relevant variables in the current period and preprocess them to obtain an input vector. The relevant variables are correlated with multiple target variables, and the relevant variables and the multiple target variables are all monitoring variables of wind turbine units.

[0013] The first extraction module is used to extract features from the input vector using the shared layer in the multi-task prediction model to obtain shared features, and to copy the shared features to obtain shared features corresponding to the multiple target variables respectively.

[0014] The second extraction module is used to extract specific features that match the corresponding target variables by using the sub-task feature layer in the multi-task prediction model for the shared features corresponding to each target variable.

[0015] The first mapping module is used to map the corresponding target variable to the specific feature that matches each target variable using the fully connected layer in the multi-task prediction model to obtain the predicted value of the target variable in the current period.

[0016] The second acquisition module is used to acquire the measured values ​​of the multiple target variables in the current period, and to perform fault prediction on the wind turbine based on the measured values ​​and predicted values ​​of the multiple target variables in the current period, and to issue an alarm when a fault is predicted.

[0017] A third aspect of this disclosure provides an electronic device, the server comprising: a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the method of the first aspect described above.

[0018] A fourth aspect of this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the method of the first aspect described above.

[0019] The technical solution provided in this disclosure has the following advantages compared with the prior art:

[0020] This embodiment of the disclosure eliminates the need for a complex physical model. By employing a multi-task prediction model that has already learned complex mapping relationships between large amounts of data, it achieves accurate prediction of multiple target variables for wind turbine units. This model is easy to model and simpler. Furthermore, it comprehensively considers the mutual influence between monitoring variables of the wind turbine unit. The measured values ​​of related variables that are correlated with multiple target variables in the current cycle are preprocessed to obtain an input vector, which serves as the input to the multi-task prediction model. This allows the shared layer in the multi-task prediction model to extract and replicate features from the input vector, obtaining shared features—that is, shared information between different prediction tasks. Moreover, the feature layers of each sub-task corresponding to each prediction task in the multi-task prediction model can extract specific features more closely related to their respective prediction tasks from the shared features. Consequently, the fully connected layers corresponding to each prediction task can more accurately map the prediction results of their respective prediction tasks (i.e., the predicted values ​​of the corresponding target variables). Therefore, this embodiment of the disclosure enables feature or information sharing between different prediction tasks, enhancing the learning ability of the multi-task prediction model and thus improving fault early warning performance, thereby increasing the accuracy and reliability of predictions for multiple target variables. Thus, this embodiment of the disclosure has the advantages of simplicity, accuracy, and reliability. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0022] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a wind turbine fault early warning method provided in an embodiment of this disclosure;

[0024] Figure 2 This is a schematic diagram of the structure of a multi-task prediction model provided in an embodiment of this disclosure;

[0025] Figure 3 This is a flowchart of a wind turbine fault early warning process provided in this public embodiment;

[0026] Figure 4 This is a schematic diagram of the structure of a wind turbine fault early warning device provided in an embodiment of this disclosure;

[0027] Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this disclosure. Detailed Implementation

[0028] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0029] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0030] Figure 1 This is a flowchart illustrating a wind turbine fault early warning method provided in an embodiment of this disclosure. This method can be executed by an electronic device. The electronic device can be exemplarily understood as a device such as a mobile phone, tablet computer, laptop computer, desktop computer, or smart TV. Figure 1 As shown, the method provided in this embodiment includes the following steps:

[0031] S110. Obtain the measured values ​​of relevant variables in the current period and preprocess them to obtain the input vector. The relevant variables are correlated with multiple target variables, and the relevant variables and multiple target variables are all monitoring variables of the wind turbine.

[0032] In this embodiment of the disclosure, the wind turbine includes a Supervisory Control and Data Acquisition (SCADA) system. The SCADA system can periodically measure the parameter values ​​of each monitored variable of the wind turbine, thereby obtaining the measurement value of each monitored variable in the current period.

[0033] Specifically, the monitored variables are those measured by the SCADA system in the wind turbine and capable of characterizing the operating status of the wind turbine. For example, the monitored variables may include: wind turbine winding temperature, wind turbine front bearing temperature, wind turbine rear bearing temperature, wind speed, power, generator speed, gearbox oil temperature, gearbox front bearing temperature, gearbox rear bearing temperature, nacelle temperature, ambient temperature, wind angle, wind direction, blade angle, generator current, voltage, etc., but are not limited to these.

[0034] Specifically, the target variable is the monitoring variable corresponding to the component in the wind turbine that needs to be monitored for fault early warning.

[0035] Specifically, the relevant variables are monitoring variables that are different from the target variables and are correlated with all target variables.

[0036] For example, when fault warning monitoring is required for the generator windings, front bearing, and rear bearing of a wind turbine, multiple target variables include the temperature of the wind turbine windings, the temperature of the front bearing, and the temperature of the rear bearing. Correspondingly, the relevant variables are monitoring variables that are correlated with the temperatures of the wind turbine windings, the front bearing, and the rear bearing. However, this is not the only possible scenario.

[0037] Specifically, there are various ways to determine the relevant variables. The following are typical examples, but they do not constitute a limitation of this application.

[0038] In some embodiments, the process of determining the relevant variables may include:

[0039] S1. Obtain multiple candidate variables, where the candidate variables are other monitoring variables besides the target variable.

[0040] Specifically, among all the monitoring variables of the wind turbine, all monitoring variables other than the target variable can be used as candidate variables.

[0041] S2. For each target variable, calculate the similarity value between the target variable and each candidate variable.

[0042] Specifically, there are various ways to calculate the similarity value. The following are typical examples, but they do not constitute a limitation of this application.

[0043] In one example, the process of calculating the similarity value between the target variable and the candidate variable includes: using the Copula function to calculate the similarity value between the candidate variable and the target variable based on the marginal distribution function of the candidate variable and the marginal distribution function of the target variable, wherein the marginal distribution function of the candidate variable and the marginal distribution function of the target variable are calculated using the kernel density estimation method.

[0044] Specifically, the kernel density estimation method is used to obtain the marginal distribution function of the candidate variable based on multiple sample measurements of the candidate variable. Similarly, the kernel density estimation method is used to obtain the marginal distribution function of the target variable based on multiple sample measurements of the target variable.

[0045] Specifically, the sample measurement values ​​are obtained by the SCADA system from measuring the parameter values ​​of the monitored variables, and are used to calculate the marginal distribution function. Correspondingly, the sample measurement values ​​of the candidate variables are obtained by the SCADA system from measuring the parameter values ​​of the candidate variables, and the same applies to the sample measurement values ​​of the target variables, which will not be elaborated here.

[0046] Specifically, the kernel density estimation method uses the following formula: substituting multiple sample measurements of the candidate variable into the formula yields the marginal distribution function of the candidate variable. Similarly, substituting multiple sample measurements of the target variable into the formula yields the marginal distribution function of the target variable.

[0047]

[0048] Where f is the marginal distribution function, h is a hyperparameter called bandwidth or window, n1 is the number of sample measurements, and x i Let x be the measurement value of the i-th sample, K be the kernel function, and K(x) ≥ 0.

[0049]

[0050] Specifically, the Copula function is calculated as follows: by substituting the marginal distribution functions of the candidate variable and the target variable into the following formula, the similarity value between the candidate variable and the target variable can be obtained:

[0051]

[0052] Where τ is the similarity value, u is the marginal distribution function of the candidate variable, v is the marginal distribution function of the target variable, and C(u,v) is the pair-Copula function of u and v.

[0053] S3. Select candidate variables whose similarity values ​​with multiple target variables are all greater than a preset threshold as relevant variables.

[0054] For example, multiple target variables include wind turbine winding temperature, wind turbine front bearing temperature, and wind turbine rear bearing temperature. If candidate variables that are correlated with wind turbine winding temperature form a set A, candidate variables that are correlated with wind turbine front bearing temperature form a set B, and candidate variables that are correlated with wind turbine rear bearing temperature form a set C, then the intersection of set A, set B, and set C is the correlated variable.

[0055] Understandably, using the Copula function to calculate the Kendall correlation coefficient (i.e., correlation value) between candidate and target variables aligns with the distribution characteristics of different variables, takes into account the randomness and coupling of variables, avoids the problems of too many parameters and missing parameters, and helps to improve the accuracy of similarity values, thereby improving the accuracy of related variables.

[0056] In other embodiments, the process of determining relevant variables may include: for each target variable, querying a preset correlation mapping table to obtain monitoring variables that are correlated with the target variable; and selecting monitoring variables that are correlated with all target variables from the monitoring variables that are correlated with each target variable as relevant variables.

[0057] The correlation mapping table contains multiple pairs of monitoring variables, and the two monitoring variables in a pair are correlated.

[0058] Specifically, preprocessing may include, but is not limited to, data resampling, missing data handling, and / or data normalization.

[0059] It is understandable that, compared to preprocessing the measurement values ​​of all monitoring variables other than the target variable in the current period to obtain the input vector, the present embodiment preprocesses the measurement values ​​of relevant variables in the current period to obtain the input vector, which can reduce the redundancy and complexity of the input vector.

[0060] S120. Using the shared layer in the multi-task prediction model, feature extraction is performed on the input vector to obtain shared features, and the shared features are copied to obtain shared features corresponding to multiple target variables.

[0061] In this embodiment, the multi-task prediction model can predict the parameter values ​​of each target variable based on the input vector, thereby outputting the predicted values ​​of each target variable. The predicted value of the target variable refers to the parameter value of the target variable when the component corresponding to the target variable does not fail, as predicted by the multi-task prediction model. Since the multi-task prediction model needs to predict the parameter values ​​of multiple target variables, it has multiple prediction tasks. The multi-task prediction model includes a shared layer, multiple sub-task feature layers, and multiple fully connected layers. Of course, the multi-task prediction model may also include an input layer and an output layer.

[0062] Specifically, shared features are features shared by multiple prediction tasks and are obtained by feature extraction from the input vector.

[0063] Optionally, the shared layer may include a TCN layer and a Lambda layer. The TCN layer can receive the input vector transmitted from the input layer and extract features from the input vector to obtain shared features. The Lambda layer can copy the shared features to obtain multiple copies of the shared features. However, it is not limited to this.

[0064] S130. For the shared features corresponding to each target variable, use the feature layer of the corresponding sub-task in the multi-task prediction model to extract specific features that match the corresponding target variable.

[0065] In this embodiment of the disclosure, each prediction task (or each target variable) corresponds to a sub-task feature layer, which can extract specific features that are more closely related to the prediction task from the shared features.

[0066] Specifically, the specific features that match the target variable refer to the shared features that are more closely related to, more relevant to, and more matched with the target variable.

[0067] Optionally, the subtask feature layer may include a GRU layer. However, it is not limited to this.

[0068] S140. For each target variable, a specific feature that matches it is used to map it to the corresponding fully connected layer in the multi-task prediction model to obtain the predicted value of the target variable in the current period.

[0069] In this embodiment of the disclosure, each prediction task (or each target variable) corresponds to a fully connected layer, which can map the received specific features to obtain the predicted value of the corresponding target variable in the current period.

[0070] For example, Figure 2 This is a schematic diagram of the structure of a multi-task prediction model provided in an embodiment of this disclosure. See also... Figure 2A multi-task prediction model is constructed using a temporal convolutional network (TCN), gated recurrent units (GRU), and multi-task learning fusion (MTL-TCN-GRU). The shared layer consists of a TCN and a Lambda layer. The TCN extracts time-series features from the input vector, completing the extraction of shared features for different prediction tasks. The Lambda layer copies the shared features extracted by the TCN and passes them to the sub-task feature layers of each prediction task, completing the data transfer of shared features from the shared layer to the sub-task feature layers (i.e., GRU). The task-specific layer consists of a GRU layer and a fully connected layer. During the training of the multi-task prediction model, the GRU layer can learn the shared information between different prediction tasks and the specific features of specific tasks, fusing the correlation between multiple prediction tasks and adaptively adjusting its state and weights according to the needs of different prediction tasks, thereby improving the performance of multi-task learning. The fully connected layer maps the specific features extracted by the GRU to the output of the prediction task to obtain the predicted value, realizing the prediction of the parameter values ​​of the target variable. For example, multiple target variables include wind turbine winding temperature (i.e., monitoring variable 1), wind turbine front bearing temperature (i.e., monitoring variable 2), and wind turbine rear bearing temperature (i.e., monitoring variable 3). Preprocessing the measurements of related variables that are correlated with these three target variables in the current cycle yields an input vector. The input layer can transmit the input vector to the TCN layer, which can extract shared features based on the input vector. The Lambda layer can then replicate these shared features to obtain three copies. For the wind turbine winding temperature prediction task, the corresponding GRU layer can extract specific features matching the wind turbine winding temperature from the shared features. The fully connected layer corresponding to this task can then map these specific features to obtain the predicted value of the wind turbine winding temperature, which is then output through the corresponding output layer. For the task of predicting the temperature of the front bearing of a wind turbine generator, the corresponding GRU layer can extract specific features that match the temperature of the front bearing from the shared features. Then, the corresponding fully connected layer can map these specific features to obtain the predicted value of the front bearing temperature, which is then output through the corresponding output layer. Similarly, for the task of predicting the temperature of the rear bearing of a wind turbine generator, the corresponding GRU layer can extract specific features that match the temperature of the rear bearing from the shared features. Then, the corresponding fully connected layer can map these specific features to obtain the predicted value of the rear bearing temperature, which is then output through the corresponding output layer.

[0071] S150. Obtain the measured values ​​of multiple target variables in the current period, and perform fault prediction on the wind turbine based on the measured and predicted values ​​of multiple target variables in the current period, and issue an alarm when a fault is predicted.

[0072] Specifically, there are various implementation methods for predicting the faults of wind turbines based on the measured and predicted values ​​of multiple target variables in the current period. Typical examples are described below, but they do not constitute a limitation of this disclosure.

[0073] In some embodiments, fault prediction of wind turbines is performed based on measured and predicted values ​​of multiple target variables in the current period, including:

[0074] S151. For each target variable, calculate the residual of the target variable in the current period based on the measured and predicted values ​​of the target variable in the current period.

[0075] In some examples, S151 includes: for each target variable, subtracting the measured and predicted values ​​of the target variable in the current period to obtain a first residual, and weighting and summing the first residual and the residual of the target variable in the previous period to obtain the residual of the target variable in the current period.

[0076] Optionally, the weighting coefficient of the first residual is greater than the weighting coefficient of the residual of the target variable in the previous period.

[0077] Specifically, the formula for calculating the residual of the target variable in the current period is as follows:

[0078] R t =λe t +(1-λ)R t-1

[0079] e t =|y′ t -R t |

[0080] Among them, R t e represents the residual of the target variable in the current period. t Let λ be the residual of the target variable in the previous period, and e be the value of λ. t Weighting coefficients; R t-1 The first residual; y′ t y represents the predicted value of the target variable in the current period. t This is the measured value of the target variable in the current period.

[0081] It is understandable that wind turbines operate in complex and variable environments. If the residual of a single point in the target variable is used as a representation of the operating status of the corresponding component, random errors are likely to occur, leading to a decrease in early warning accuracy. However, in the embodiments of this disclosure, when calculating the residual of the target variable in the current period, the first residual and the residual of the target variable in the previous period can be weighted and summed, and the first residual closer to the current moment can be assigned a larger weight, reducing the impact of random errors and thus improving the early warning accuracy.

[0082] In other examples, for each target variable, the residual of the target variable in the current period can also be obtained by subtracting the measured and predicted values ​​of the target variable in the current period.

[0083] S152. If the residual of at least one target variable in the current period is greater than its corresponding residual threshold, then the wind turbine unit is determined to have failed.

[0084] Specifically, the residual thresholds for different target variables can be the same or different. The residual threshold for the same target variable can be a fixed preset value (e.g., set based on engineering experience) or it can be calculated in real time. This disclosure does not limit this.

[0085] In one example, the process of determining the residual threshold corresponding to the target variable includes: obtaining multiple historical predicted values ​​of the target variable and calculating the average and standard deviation of the multiple historical predicted values; and determining the residual threshold corresponding to the target variable based on the average, standard deviation, preset width coefficient and preset weight value of the residual threshold corresponding to the target variable.

[0086] Specifically, the multiple historical predicted values ​​of the target variable mentioned here refer to the measured value of the target variable in the current period, as well as the measured values ​​in previous historical periods. The specific number of historical predicted values ​​can be set by those skilled in the art according to the actual situation, and is not limited here.

[0087] For example, the residual threshold corresponding to the target variable can be calculated using the following formula:

[0088]

[0089] in, The residual threshold for the target variable in the current period; is the average of multiple historical predicted values ​​of the target variable; c is the width coefficient (i.e., the preset width coefficient) for triggering the residual threshold control limit; Let be the standard deviation of multiple historical predicted values ​​of the target variable; λ is e t The weighting coefficients; t is the current time; n2 is the number of historical predicted values ​​for the target variable.

[0090] It can be understood that by calculating the residual threshold in real time, the residual threshold can be made more consistent with the current operating conditions of the wind turbine, which is conducive to improving the accuracy of the residual threshold and thus improving the accuracy of early warning.

[0091] It is also understandable that when the early warning monitoring components in a wind turbine are operating normally, the predicted values ​​of the target variables output by the multi-task prediction model are close to the measured values, with small residuals. However, when the early warning monitoring components are operating abnormally, the predicted values ​​of the target variables output by the multi-task prediction model differ significantly from the measured values, resulting in larger residuals. Therefore, the magnitude of the residuals can be used to determine whether the early warning monitoring components are in an abnormal state, thereby enabling early fault warnings for these components.

[0092] In other embodiments, the measured and predicted values ​​of multiple target variables in the current period are input into the operation status prediction model to obtain the operation status prediction results output by the operation status prediction model, wherein the operation status prediction results include normal operation status and abnormal operation status.

[0093] Optionally, after determining that a wind turbine has failed, the method further includes: for each target variable, obtaining multiple historical measurement values ​​of the target variable, and determining the fluctuation coefficient of the target variable based on the multiple historical measurement values; and identifying the component corresponding to the target variable with the largest fluctuation coefficient as the failed component.

[0094] Specifically, the multiple historical measurements of the target variable mentioned here refer to the measurement of the target variable in the current period, as well as the measurement values ​​in previous historical periods. The specific number of historical measurements can be set by those skilled in the art according to the actual situation, and is not limited here.

[0095] Specifically, the volatility coefficient can be calculated using the following formula:

[0096]

[0097] Among them, y i Let be the i-th historical measurement; μ be the average of the multiple historical measurements of the target variable; and n3 be the number of the multiple historical measurements of the target variable.

[0098] Understandably, when a faulty component occurs within a wind turbine, the measured value of the target variable corresponding to that component first experiences a sudden change, resulting in significant fluctuations. Subsequently, heat is conducted to other components, causing abnormal fluctuations in the measured values ​​of the target variables corresponding to those components. Due to heat loss during heat conduction, in the early stages of the fault, the measured values ​​of the target variables corresponding to non-faulty components show less fluctuation compared to those corresponding to the faulty component. Therefore, the degree of fluctuation across multiple target variables can be used to locate the fault and pinpoint the faulty component.

[0099] This embodiment of the disclosure eliminates the need for a complex physical model. By employing a multi-task prediction model that has already learned complex mapping relationships between large amounts of data, it achieves accurate prediction of multiple target variables for wind turbine units. This model is easy to model and simpler. Furthermore, it comprehensively considers the mutual influence between monitoring variables of the wind turbine unit. The measured values ​​of related variables that are correlated with multiple target variables in the current cycle are preprocessed to obtain an input vector, which serves as the input to the multi-task prediction model. This allows the shared layer in the multi-task prediction model to extract and replicate features from the input vector, obtaining shared features—that is, shared information between different prediction tasks. Moreover, the feature layers of each sub-task corresponding to each prediction task in the multi-task prediction model can extract specific features more closely related to their respective prediction tasks from the shared features. Consequently, the fully connected layers corresponding to each prediction task can more accurately map the prediction results of their respective prediction tasks (i.e., the predicted values ​​of the corresponding target variables). Therefore, this embodiment of the disclosure enables feature or information sharing between different prediction tasks, enhancing the learning ability of the multi-task prediction model and thus improving fault early warning performance, thereby increasing the accuracy and reliability of predictions for multiple target variables. Thus, this embodiment of the disclosure has the advantages of simplicity, accuracy, and reliability.

[0100] The wind turbine fault early warning method provided in this disclosure will be described in detail below with reference to a specific example. For example, Figure 3 This is a flowchart illustrating a wind turbine fault early warning process provided in an embodiment of this disclosure. See also... Figure 3A multi-task prediction model is trained using offline training samples. Specifically, data preprocessing involves preprocessing the measured values ​​of monitoring variables of the wind turbine collected by the SCADA system over a period of time (including multiple acquisition cycles). Then, based on Copula correlation analysis, the relevant variables corresponding to multiple target variables are determined. The multi-task prediction model is then trained using the measured values ​​of the relevant variables and the measured values ​​of the multiple target variables. Alternatively, residual thresholds can be calculated based on the measured values ​​of the multiple target variables. After the multi-task prediction model is trained, online sample detection is performed. Specifically, real-time data is acquired, i.e., the measured values ​​of multiple target variables in the current cycle are obtained. Then, the residuals between the predicted values ​​output by the multi-task prediction model and the measured values ​​are calculated and smoothed. If the residual of at least one target variable is greater than the corresponding residual threshold, the wind turbine is determined to be in an abnormal state; otherwise, the wind turbine is determined to be in a normal state. When the wind turbine is determined to be in an abnormal state, further fault location can be performed. It should be noted that as time goes on, the more measurement values ​​of the monitoring variables of the wind turbine units collected by the SCADA system, the more samples can be accumulated for training the multi-task prediction model. Therefore, the multi-task prediction model can be trained periodically (for example, once a month, but not limited to this) to make the prediction results of the multi-task prediction model more accurate.

[0101] This embodiment of the disclosure considers the correlation between different prediction tasks, introduces multi-prediction task learning into the multi-task prediction model of wind turbines, balances the learning speed and accuracy of each prediction task, fully extracts the correlation information of each prediction task, realizes information sharing among multiple prediction tasks, improves the model's learning ability, and enables monitoring of multiple components. This improves the accuracy of fault early warning and provides a reliable basis for wind farm operation and maintenance.

[0102] Figure 4 This is a schematic diagram of a wind turbine fault early warning device provided in an embodiment of this disclosure. This wind turbine fault early warning device can be understood as the aforementioned electronic device or a functional module within the aforementioned electronic device. For example... Figure 4 As shown, the wind turbine fault early warning device 400 includes:

[0103] The first acquisition module 410 is used to acquire the measurement value of the relevant variable in the current period and preprocess it to obtain an input vector. The relevant variable is correlated with multiple target variables, and the relevant variable and the multiple target variables are all monitoring variables of the wind turbine.

[0104] The first extraction module 420 is used to extract features from the input vector using the shared layer in the multi-task prediction model to obtain shared features, and to copy the shared features to obtain shared features corresponding to the multiple target variables respectively.

[0105] The second extraction module 430 is used to extract specific features that match the corresponding target variables by using the sub-task feature layer in the multi-task prediction model for the shared features corresponding to each target variable.

[0106] The first mapping module 440 is used to map the corresponding target variable in the current period to the specific features that match each target variable using the fully connected layer in the multi-task prediction model.

[0107] The second acquisition module 450 is used to acquire the measured values ​​of the multiple target variables in the current period, and to perform fault prediction on the wind turbine based on the measured values ​​and predicted values ​​of the multiple target variables in the current period, and to issue an alarm when a fault is predicted.

[0108] In other embodiments, the apparatus further includes: a first determining module for determining relevant variables, wherein the first determining module includes:

[0109] The first acquisition submodule is used to acquire multiple candidate variables, wherein the candidate variables are monitoring variables other than the target variable;

[0110] The first calculation submodule is used to calculate the similarity value between the target variable and each candidate variable for each target variable;

[0111] The first determining submodule is used to select candidate variables whose similarity values ​​with the plurality of target variables are all greater than a preset threshold as the relevant variables.

[0112] In some other embodiments, the device further includes a first calculation module for calculating the similarity value between the target variable and the candidate variable. Specifically, the first calculation module is used to calculate the similarity value between the candidate variable and the target variable using a Copula function based on the marginal distribution function of the candidate variable and the marginal distribution function of the target variable. The marginal distribution function of the candidate variable and the marginal distribution function of the target variable are calculated using kernel density estimation.

[0113] In some other implementations, the shared layer includes a TCN layer and a Lambda layer, and the subtask feature layer includes a GRU layer.

[0114] In some other embodiments, the second acquisition module 450 includes a first prediction submodule, used to perform fault prediction on the wind turbine based on the measured and predicted values ​​of the plurality of target variables in the current period, wherein the first prediction submodule includes:

[0115] The first calculation unit is used to calculate the difference between the measured value and the predicted value of the target variable in the current period for each target variable to obtain a first residual, and to perform a weighted summation of the first residual and the residual of the target variable in the previous period to obtain the residual of the target variable in the current period.

[0116] The first determining unit is configured to determine that the wind turbine has failed if the residual of at least one of the target variables in the current period is greater than its corresponding residual threshold.

[0117] In another embodiment of this disclosure, the apparatus further includes a second determining module for determining a residual threshold corresponding to the target variable, wherein the second determining module includes:

[0118] The second calculation submodule is used to obtain multiple historical predicted values ​​of the target variable and calculate the average and standard deviation of the multiple historical predicted values;

[0119] The second determining submodule is used to determine the residual threshold corresponding to the target variable based on the average value, the standard deviation, the preset width coefficient and the preset weight value of the residual threshold corresponding to the target variable.

[0120] In another embodiment of this disclosure, the device further includes: for

[0121] The third acquisition module is used to acquire multiple historical measurement values ​​of each target variable of the wind turbine after determining that the wind turbine has failed, and to determine the fluctuation coefficient of the target variable based on the multiple historical measurement values.

[0122] The third determination module is used to identify the component corresponding to the target variable with the largest fluctuation coefficient as the faulty component.

[0123] The apparatus provided in this embodiment can execute the methods of any of the above embodiments, and its execution method and beneficial effects are similar, so they will not be described again here.

[0124] This disclosure also provides an electronic device, which includes: a memory storing a computer program; and a processor for executing the computer program, wherein when the computer program is executed by the processor, it can implement the methods of any of the above embodiments.

[0125] Example, Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of this disclosure. See below for details. Figure 5The diagram illustrates a structural schematic suitable for implementing the electronic device 500 in the embodiments of this disclosure. The electronic device 500 in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0126] like Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processor, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0127] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0128] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, it performs the functions defined in the methods of embodiments of this disclosure.

[0129] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0130] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0131] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0132] The aforementioned computer-readable medium carries one or more programs. When the electronic device executes the aforementioned one or more programs, the electronic device causes the following to occur: First, it acquires the measured values ​​of relevant variables in the current period and preprocesses them to obtain an input vector, wherein the relevant variables are correlated with multiple target variables, and the relevant variables and multiple target variables are all monitoring variables of the wind turbine. Second, it uses a shared layer in a multi-task prediction model to extract features from the input vector to obtain shared features, and copies the shared features to obtain shared features corresponding to multiple target variables. Third, for each shared feature corresponding to a target variable, it uses the corresponding sub-task feature layer in the multi-task prediction model to extract features to obtain specific features matching the corresponding target variable. Fourth, for each specific feature matching a target variable, it uses the corresponding fully connected layer in the multi-task prediction model to map it to obtain the predicted value of the corresponding target variable in the current period. Fifth, it acquires the measured values ​​of multiple target variables in the current period, and performs fault prediction of the wind turbine based on the measured and predicted values ​​of multiple target variables in the current period, and issues an alarm when a fault is predicted.

[0133] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0134] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0135] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0136] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0137] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0138] This disclosure also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement the methods of any of the above embodiments. The execution method and beneficial effects are similar, and will not be described again here.

[0139] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0140] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for early warning of wind turbine faults, characterized in that, include: The measured values ​​of relevant variables in the current period are obtained and preprocessed to obtain an input vector. The relevant variables are correlated with multiple target variables. The relevant variables and the multiple target variables are all monitoring variables of the wind turbine. The monitoring variables include: wind turbine winding temperature, wind turbine front bearing temperature, wind turbine rear bearing temperature, wind speed, wind turbine power, wind turbine speed, gearbox oil temperature, gearbox front bearing temperature, gearbox rear bearing temperature, nacelle temperature, ambient temperature, wind angle, wind direction, blade angle, wind turbine current, and wind turbine voltage. Using the shared layer in the multi-task prediction model, feature extraction is performed on the input vector to obtain shared features, and the shared features are copied to obtain shared features corresponding to the multiple target variables respectively. The shared layer includes a TCN layer and a Lambda layer. For each target variable, the shared features corresponding to it are extracted using the sub-task feature layer in the multi-task prediction model to obtain specific features that match the corresponding target variable. The sub-task feature layer includes a GRU layer. For each target variable, a specific feature that matches it is used to map it using the fully connected layer in the multi-task prediction model to obtain the predicted value of the target variable in the current period. The system acquires the measured values ​​of the multiple target variables in the current period, performs fault prediction on the wind turbine based on the measured and predicted values ​​of the multiple target variables in the current period, and issues an alarm when a fault is predicted. The process of determining the relevant variables includes: Multiple candidate variables are obtained, wherein the candidate variables are other monitoring variables besides the target variable; For each target variable, calculate the similarity value between the target variable and each candidate variable; Candidate variables whose similarity values ​​to the multiple target variables are all greater than a preset threshold are selected as the relevant variables; The step of predicting faults in the wind turbine based on the measured and predicted values ​​of the multiple target variables in the current period includes: For each target variable, the difference between the measured value and the predicted value of the target variable in the current period is used to obtain the first residual. The first residual and the residual of the target variable in the previous period are then weighted and summed to obtain the residual of the target variable in the current period. If the residual of at least one of the target variables in the current period is greater than its corresponding residual threshold, then the wind turbine is determined to have failed. The wind turbine fault early warning method further includes, after determining that a fault has occurred in the wind turbine, the wind turbine fault warning method also includes: For each target variable, multiple historical measurements of the target variable are obtained, and the fluctuation coefficient of the target variable is determined based on the multiple historical measurements; The component corresponding to the target variable with the largest fluctuation coefficient is designated as the faulty component.

2. The method according to claim 1, characterized in that, The process of calculating the similarity value between the target variable and the candidate variable includes: The similarity value between the candidate variable and the target variable is calculated using the Copula function based on the marginal distribution functions of the candidate variable and the target variable. The marginal distribution functions of the candidate variable and the target variable are calculated using the kernel density estimation method.

3. The method according to claim 1, characterized in that, The process of determining the residual threshold corresponding to the target variable includes: Obtain multiple historical predicted values ​​of the target variable, and calculate the mean and standard deviation of the multiple historical predicted values; The residual threshold corresponding to the target variable is determined based on the mean, the standard deviation, the preset width coefficient and the preset weight value of the residual threshold corresponding to the target variable.

4. A wind turbine fault early warning device, characterized in that, include: The first acquisition module is used to acquire the measured values ​​of relevant variables in the current period and preprocess them to obtain an input vector. The relevant variables are correlated with multiple target variables. The relevant variables and the multiple target variables are all monitoring variables of the wind turbine. The monitoring variables include: wind turbine winding temperature, wind turbine front bearing temperature, wind turbine rear bearing temperature, wind speed, wind turbine power, wind turbine speed, gearbox oil temperature, gearbox front bearing temperature, gearbox rear bearing temperature, nacelle temperature, ambient temperature, wind angle, wind direction, blade angle, wind turbine current, and wind turbine voltage. The first extraction module is used to extract features from the input vector using the shared layer in the multi-task prediction model to obtain shared features, and to copy the shared features to obtain shared features corresponding to the multiple target variables respectively. The shared layer includes a TCN layer and a Lambda layer. The second extraction module is used to extract specific features that match the corresponding target variables from the shared features corresponding to each target variable using the sub-task feature layer in the multi-task prediction model. The sub-task feature layer includes a GRU layer. The first mapping module is used to map the corresponding target variable to the specific feature that matches each target variable using the fully connected layer in the multi-task prediction model to obtain the predicted value of the target variable in the current period. The second acquisition module is used to acquire the measured values ​​of the multiple target variables in the current period, and to perform fault prediction on the wind turbine based on the measured values ​​and predicted values ​​of the multiple target variables in the current period, and to issue an alarm when a fault is predicted. The wind turbine fault early warning device further includes: a first determining module, used to determine relevant variables, wherein the first determining module includes: The first acquisition submodule is used to acquire multiple candidate variables, wherein the candidate variables are monitoring variables other than the target variable; The first calculation submodule is used to calculate the similarity value between the target variable and each candidate variable for each target variable; The first determining submodule is used to select candidate variables whose similarity values ​​with the plurality of target variables are all greater than a preset threshold as the relevant variables; The second acquisition module includes a first prediction submodule, which includes: The first calculation unit is used to calculate the difference between the measured value and the predicted value of the target variable in the current period for each target variable to obtain a first residual, and to perform a weighted summation of the first residual and the residual of the target variable in the previous period to obtain the residual of the target variable in the current period. The first determining unit is configured to determine that the wind turbine has failed if the residual of at least one of the target variables in the current period is greater than its corresponding residual threshold. The wind turbine fault early warning device further includes a third acquisition module, which, after determining that the wind turbine has failed, acquires multiple historical measurement values ​​of each target variable of the wind turbine, and determines the fluctuation coefficient of the target variable based on the multiple historical measurement values. The third determination module is used to identify the component corresponding to the target variable with the largest fluctuation coefficient as the faulty component.

5. An electronic device, characterized in that, include: A processor and a memory, wherein the memory stores a computer program that, when executed by the processor, performs the method of any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-3.

Images