Wind turbine bearing lubrication fault prediction method based on active learning

Abstract

The present application relates to a method for predicting lubrication faults of wind turbine bearings based on active learning, and relates to the field of computer technology; obtaining working data corresponding to a bearing lubrication system of a wind turbine, the working data including multiple historical lubrication working data, eliminating failed working data based on the multiple historical lubrication working data, and obtaining target historical lubrication working data; determining a fault prediction result of the bearing lubrication system through a pre-trained support vector machine model and the target historical lubrication working data; wherein the training data of the pre-trained support vector machine model includes training data obtained through active learning and failure sample screening, so that the accuracy of the support vector machine model is higher, and thus the accuracy of the fault prediction result is also higher. Therefore, the technical solution can achieve accurate and effective wind turbine bearing lubrication fault prediction based on active learning.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular, to a method for predicting lubrication faults of wind turbine bearings based on active learning. Background Art

[0002] With the development of renewable energy technology, wind turbines have also developed. Wind turbines can convert the kinetic energy of wind into electrical energy. Wind turbines include wind rotors and generators. The wind rotors are composed of blades, hubs, and reinforcements. Among the hubs and wind rotors, bearing components are involved. The bearing components can realize the transmission of the corresponding components so that the wind turbine can operate stably.

[0003] The bearing lubrication of wind turbines is a key factor in ensuring their long-term stable operation. The lubrication system usually includes main bearing lubrication, yaw bearing lubrication, and pitch bearing lubrication. A good lubrication system can reduce bearing friction, reduce wear, extend service life, and ensure the safe and efficient operation of wind turbines. Summary of the invention

[0004] The purpose of the present application is to provide a wind turbine bearing lubrication fault prediction method based on active learning, which realizes accurate and effective wind turbine bearing lubrication fault prediction based on active learning.

[0005] In order to achieve the above-mentioned purpose, the present application provides a wind turbine bearing lubrication fault prediction method based on active learning, including: obtaining working data corresponding to the wind turbine bearing lubrication system, the working data including: multiple historical lubrication working data; eliminating failed working data from the working data corresponding to the wind turbine bearing lubrication system to obtain target historical lubrication working data; determining the fault prediction result corresponding to the wind turbine bearing lubrication system according to the target historical lubrication working data and a pre-trained support vector machine model, wherein the training data of the pre-trained support vector machine model includes training data obtained through active learning and failure sample screening.

[0006] Optionally, each historical lubrication work data includes: lubrication start time, lubrication end time and lubricant usage, and the work data corresponding to the wind turbine bearing lubrication system is subjected to invalid work data elimination to obtain target historical lubrication work data, including: determining the total lubrication time corresponding to multiple historical lubrication work data according to the lubrication start time and lubrication end time respectively included in multiple historical lubrication work data; determining a benchmark total lubrication time according to the total lubrication time corresponding to multiple historical lubrication work data; determining a benchmark lubricant usage according to the lubricant usage respectively included in multiple historical lubrication work data; and eliminating invalid work data from the work data corresponding to the wind turbine bearing lubrication system according to the benchmark total lubrication time and the benchmark lubricant usage to obtain target historical lubrication work data.

[0007] Optionally, the invalid working data corresponding to the working data of the wind turbine bearing lubrication system is eliminated according to the benchmark total lubrication time and the benchmark lubricant usage, including: comparing the total lubrication times corresponding to multiple historical lubrication working data with the benchmark total lubrication time, and comparing the lubricant usage respectively included in multiple historical lubrication working data with the benchmark lubricant usage; eliminating historical lubrication working data with a total lubrication time less than the benchmark total lubrication time and a lubricant usage less than the benchmark lubricant usage; and / or eliminating historical lubrication working data with a total lubrication time greater than or equal to the benchmark total lubrication time and a lubricant usage greater than or equal to the benchmark lubricant usage.

[0008] Optionally, the wind turbine bearing lubrication fault prediction method also includes: obtaining original training samples, the original training samples include multiple first lubrication work data samples and real labels corresponding to the multiple first lubrication work data samples; training the support vector machine model to be trained according to the original training samples to obtain the original trained support vector machine model; obtaining active learning samples, the active learning samples include multiple second lubrication work data samples and real labels corresponding to the multiple second lubrication work data samples; determining the response values ​​corresponding to the multiple first lubrication work data samples, and determining the response values ​​corresponding to the multiple second lubrication work data samples, any response value is used to characterize the bearing lubrication fault response speed; according to the response values ​​corresponding to the multiple first lubrication work data samples and the response values ​​corresponding to the multiple second lubrication work data samples, the multiple first lubrication work data samples and the multiple second lubrication work data samples are screened for failure samples to determine the target active learning samples; according to the target active learning samples, the originally trained support vector machine model is updated and trained to obtain the pre-trained support vector machine model.

[0009] Optionally, obtaining active learning samples includes: querying a preset bearing operation database to obtain abnormal bearing operation data; obtaining associated bearing lubrication work data of the abnormal bearing operation data, wherein the lubrication start time and lubrication end time corresponding to the associated bearing lubrication work data satisfy a preset time relationship with the recording time of the abnormal bearing operation data; determining a plurality of initial second lubrication work data samples from the associated bearing lubrication work data; and performing sample expansion on the plurality of initial second lubrication work data samples to obtain the active learning samples.

[0010] Optionally, determining the response values ​​corresponding to the multiple second lubrication work data samples respectively includes: inputting the multiple second lubrication work data samples into the originally trained support vector machine model to obtain the fault prediction results output by the originally trained support vector machine model; determining the prediction labels corresponding to the multiple second lubrication work data samples respectively according to the fault prediction results output by the originally trained support vector machine model; determining the response values ​​corresponding to the multiple second lubrication work data samples respectively according to the prediction labels corresponding to the multiple second lubrication work data samples respectively, the real labels corresponding to the multiple second lubrication work data samples respectively, and the pre-trained response value determination model.

[0011] Optionally, the wind turbine bearing lubrication fault prediction method also includes: obtaining a training data set, the training data set including multiple third lubrication work data samples, real fault labels and predicted fault labels corresponding to the multiple third lubrication work data samples respectively; performing a fault processing simulation according to the real fault labels corresponding to the multiple third lubrication work data samples respectively, to obtain a first response value; performing a fault processing simulation according to the predicted fault labels corresponding to the multiple third lubrication work data samples respectively, to obtain a second response value; determining the response value prediction labels corresponding to the multiple third lubrication work data samples respectively according to the first response value and the second response value; training the response value determination model to be trained according to the training data set and the response value prediction labels corresponding to the multiple third lubrication work data samples respectively, to obtain the pre-trained response value determination model.

[0012] Optionally, the method of screening the multiple first lubrication working data samples and the multiple second lubrication working data samples for failed samples according to the response values ​​respectively corresponding to the multiple first lubrication working data samples and the multiple second lubrication working data samples to determine the target active learning sample includes: sorting the multiple first lubrication working data samples according to the response values ​​respectively corresponding to the multiple first lubrication working data samples to obtain a first sorting result; sorting the multiple second lubrication working data samples according to the response values ​​respectively corresponding to the multiple second lubrication working data samples to obtain a second sorting result; sorting the multiple first lubrication working data samples and the multiple second lubrication working data samples according to the response values ​​respectively corresponding to the multiple first lubrication working data samples and the multiple second lubrication working data samples to obtain a third sorting result; sampling the lubrication working data samples according to the first sorting result, the second sorting result and the third sorting result, and determining the sampled lubrication working data samples as failed samples; and eliminating the failed samples to obtain the target active learning sample.

[0013] Optionally, the wind turbine bearing lubrication fault prediction method also includes: if the fault prediction result indicates that there is no fault in the wind turbine bearing lubrication system, determining a reference fault prediction result based on the eliminated failed working data and the pre-trained support vector machine model; if the reference fault prediction result indicates that there is a fault in the wind turbine bearing lubrication system, determining the eliminated failed working data as a new active learning sample; and updating the pre-trained support vector machine model based on the new active learning sample.

[0014] Optionally, each historical lubrication work data includes: lubrication start time, lubrication end time and lubricant usage, and the wind turbine bearing lubrication fault prediction method also includes: if the fault prediction result indicates that there is a fault in the wind turbine bearing lubrication system, determine a fault handling strategy based on at least one piece of information including the lubrication start time, lubrication end time and lubricant usage included in the target historical lubrication work data, and execute the fault handling strategy.

[0015] Through the above technical scheme, the working data corresponding to the wind turbine bearing lubrication system is obtained, and the working data includes multiple historical lubrication working data. Based on the multiple historical lubrication working data, the failed working data is eliminated to obtain the target historical lubrication working data; the fault prediction result of the bearing lubrication system is determined through the pre-trained support vector machine model and the target historical lubrication working data; wherein, the training data of the pre-trained support vector machine model includes the training data obtained by active learning and failure sample screening, so that the accuracy of the support vector machine model is higher, and thus the accuracy of the fault prediction result is also higher. Therefore, this technical scheme can realize accurate and effective wind turbine bearing lubrication fault prediction based on active learning.

[0016] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present application but do not constitute a limitation to the present application. In the accompanying drawings:

[0018] Figure 1 It is a structural block diagram of an intelligent bearing lubrication system according to an exemplary embodiment.

[0019] Figure 2 The present invention is a flowchart of a method for predicting lubrication faults of a wind turbine bearing based on active learning according to an exemplary embodiment.

[0020] Figure 3 The figure is a flowchart of a model training process according to an exemplary embodiment.

[0021] Figure 4 The present invention is a flowchart of determining a response value of an active learning sample according to an exemplary embodiment.

[0022] Figure 5 It is a block diagram of a wind turbine bearing lubrication fault prediction device based on active learning according to an exemplary embodiment.

[0023] Figure 6 It is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0024] The specific implementation of the present application is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present application, and is not used to limit the present application.

[0025] In various application scenarios of wind turbines, bearing lubrication of wind turbines is a key factor in ensuring their long-term stable operation. The lubrication system usually includes main bearing lubrication, yaw bearing lubrication, and pitch bearing lubrication. A good lubrication system can reduce bearing friction, reduce wear, extend service life, and ensure the safe and efficient operation of wind turbines.

[0026] Therefore, in these application scenarios, fault prediction can be performed on the wind turbine bearing lubrication system, and the wind turbine bearing lubrication system can be repaired or replaced according to the fault prediction results to ensure the stability and reliability of the wind turbine generator set.

[0027] Figure 1 is a structural block diagram of an intelligent bearing lubrication system according to an exemplary embodiment. Figure 1 As shown, the lubrication system mainly includes a lubrication oil pump, an oil filter, a lubrication pipeline, a metering device and a monitoring device.

[0028] Among them, the lubricating oil pump, oil filter, lubricating pipeline and metering device are the basic components of the lubrication system, which can realize the delivery of lubricant to the wind turbine bearing. For the specific implementation of these components, reference can be made to the mature technology in the field.

[0029] With regard to the monitoring device, fault prediction of bearing lubrication can be realized so as to perform fault handling on the lubrication system according to the fault prediction result.

[0030] Therefore, the technical solution provided in the embodiments of the present application can be applied to a monitoring device.

[0031] In the related art, in order to realize fault prediction of the bearing lubrication system, monitoring personnel are usually configured to regularly check the monitoring images or videos of the bearings, and determine whether the bearings are faulty based on the monitoring images or videos. This fault monitoring method is not automatic and has low accuracy.

[0032] Based on this, an embodiment of the present application provides a technical solution, in which a support vector machine model is used to achieve automatic and accurate fault prediction, thereby ensuring that the wind turbine bearing lubrication can effectively lubricate the wind turbine and improve the reliability and stability of the wind turbine generator set.

[0033] Figure 2 is a flow chart of a method for predicting lubrication faults of a wind turbine bearing based on active learning according to an exemplary embodiment. Figure 2 As shown, the method comprises the following steps:

[0034] Step S21, obtaining working data corresponding to the lubrication system of the wind turbine bearing, the working data including: multiple historical lubrication working data.

[0035] Step S22, eliminating invalid working data from working data corresponding to the lubrication system of the wind turbine bearing to obtain target historical lubrication working data.

[0036] Step S23, determining the fault prediction result corresponding to the wind turbine bearing lubrication system according to the target historical lubrication work data and the pre-trained support vector machine model, wherein the training data of the pre-trained support vector machine model includes training data obtained through active learning and failure sample screening.

[0037] In step S21, the working data corresponding to the wind turbine bearing lubrication system may be the recorded historical lubrication working data. Thus, after the bearing lubrication system completes the lubrication work each time, the relevant data of this work may be recorded, and fault prediction may be performed through multiple historical lubrication working data.

[0038] In some embodiments, the fault prediction may be periodic, for example, the fault prediction is performed once every other week, and the corresponding working data is the lubrication working data within that week.

[0039] In some embodiments, each historical lubrication work data includes: lubrication start time, lubrication end time and lubricant amount. The lubrication start time can be understood as the time when the bearing lubrication system starts the lubrication function, the lubrication end time can be understood as the time when the lubricant delivery is stopped, and the lubricant amount can be determined by statistically analyzing the lubricant reduction of the lubrication pump. For example, when lubrication starts, the lubricant remaining in the lubrication pump is A, and after lubrication ends, the lubricant remaining in the lubrication pump is B, then the lubricant amount can be AB.

[0040] In step S22, based on multiple historical lubrication work data, invalid work data may be eliminated first to obtain target historical lubrication work data.

[0041] In some embodiments, the failed operation data may be historical lubrication operation data that obviously has no fault prediction value.

[0042] As an optional implementation, step S22 includes: determining the total lubrication time corresponding to the multiple historical lubrication work data according to the lubrication start time and the lubrication end time respectively included in the multiple historical lubrication work data; determining the benchmark total lubrication time according to the total lubrication time corresponding to the multiple historical lubrication work data; determining the benchmark lubricant usage according to the lubricant usage respectively included in the multiple historical lubrication work data; eliminating failed working data of the working data corresponding to the wind turbine bearing lubrication system according to the benchmark total lubrication time and the benchmark lubricant usage, to obtain the target historical lubrication working data.

[0043] In this embodiment, the total lubrication durations corresponding to the multiple historical lubrication work data are first determined, so as to determine a reference total lubrication duration according to the total lubrication durations corresponding to the multiple historical lubrication work data, and the reference lubricant dosage is determined according to the lubricant dosages respectively included in the multiple historical lubrication work data.

[0044] Furthermore, based on the baseline total lubrication time and the baseline lubricant usage, the failed working data can be determined and eliminated.

[0045] In some embodiments, the reference lubrication total duration may be the mean, median or minimum of the lubrication total durations corresponding to multiple historical lubrication work data, etc. The reference lubricant dosage may be the mean, median or minimum of the lubricant dosages included in multiple historical lubrication work data, etc.

[0046] In some embodiments, based on a benchmark total lubrication time and a benchmark lubricant usage, invalid working data is eliminated from working data corresponding to the lubrication system of a wind turbine bearing, which may include: comparing the total lubrication time corresponding to multiple historical lubrication working data with the benchmark total lubrication time, and comparing the lubricant usage included in multiple historical lubrication working data with the benchmark lubricant usage; eliminating historical lubrication working data whose total lubrication time is less than the benchmark total lubrication time and whose lubricant usage is less than the benchmark lubricant usage; and / or eliminating historical lubrication working data whose total lubrication time is greater than or equal to the benchmark total lubrication time and whose lubricant usage is greater than or equal to the benchmark lubricant usage.

[0047] In this embodiment, if the total lubrication time is less than the reference lubrication time, and the lubricant amount is less than the reference lubricant amount, it means that the lubrication effect of the historical lubrication work data is not significant, so its reference value for fault prediction is not great, and it can be eliminated as failed work data. Also, if the total lubrication time is greater than or equal to the reference lubrication time, and the lubricant amount is greater than or equal to the reference lubricant amount, it means that the lubrication effect of the historical lubrication work data is too significant, so its reference value for fault prediction is not great, and it can also be eliminated as failed work data.

[0048] It can be understood that the above two conditions can be determined at the same time, or only one of them can be determined.

[0049] In step S23, the fault prediction result corresponding to the wind turbine bearing lubrication system is determined according to the target historical lubrication work data and the pre-trained support vector machine model.

[0050] In some embodiments, the target historical lubrication work data is converted into a vector sequence, and then the vector sequence is input into a pre-trained support vector machine model, and the pre-trained support vector machine model can output a category prediction result, which can be represented by a probability value, and based on the category prediction result, a fault prediction result can be determined. The fault prediction result is used to characterize whether a bearing lubrication system has a fault (occurrence).

[0051] In an embodiment of the present application, the training data of the pre-trained support vector machine model includes training data obtained through active learning and failure sample screening, that is, the training data of the support vector machine model is obtained by two methods: active learning and failure sample screening, wherein the rules for failure sample screening and the principles for eliminating failed work may be inconsistent.

[0052] As an optional implementation, the training of the support vector machine model includes: obtaining original training samples, the original training samples include multiple first lubrication work data samples and multiple real labels corresponding to the first lubrication work data samples; training the support vector machine model to be trained according to the original training samples to obtain the original trained support vector machine model; obtaining active learning samples, the active learning samples include multiple second lubrication work data samples and multiple real labels corresponding to the second lubrication work data samples; determining response values ​​corresponding to the multiple first lubrication work data samples, and determining response values ​​corresponding to the multiple second lubrication work data samples, any response value is used to characterize the response speed of the bearing lubrication fault; according to the response values ​​corresponding to the multiple first lubrication work data samples and the response values ​​corresponding to the multiple second lubrication work data samples, the multiple first lubrication work data samples and the multiple second lubrication work data samples are screened for failure samples to determine the target active learning samples; according to the target active learning samples, the originally trained support vector machine model is updated and trained to obtain a pre-trained support vector machine model.

[0053] In this implementation, three types of training samples are involved: original training samples, active learning samples, and target active learning samples.

[0054] Figure 3 is a flowchart of a model training process according to an exemplary embodiment. Figure 3 As shown in the figure, the support vector machine model is first trained based on the original training samples to obtain the original trained support vector machine model. Then, the response values ​​corresponding to the original training samples and the active learning samples are determined. Based on the response values, the original training samples and the active learning samples are screened for invalid samples to obtain the target active learning samples. Then, the support vector machine model is updated and trained based on the active learning samples to obtain the final support vector machine model.

[0055] In the related art, active learning generally occurs after the application of the support vector machine model. In the embodiment of the present application, active learning occurs before the application of the support vector machine model. Therefore, the method of obtaining active learning samples is different from the method of obtaining active learning samples in the related art.

[0056] As an optional implementation, obtaining active learning samples includes: querying a preset bearing operation database to obtain abnormal bearing operation data; obtaining associated bearing lubrication work data of the abnormal bearing operation data, wherein the lubrication start time and lubrication end time corresponding to the associated bearing lubrication work data satisfy a preset time relationship with the recording time of the abnormal bearing operation data; determining a plurality of initial second lubrication work data samples from the associated bearing lubrication work data; and performing sample expansion on the plurality of initial second lubrication work data samples to obtain active learning samples.

[0057] In the embodiment of the present application, considering the relationship between bearing lubrication failure and bearing operation, bearing operation data is used to acquire active learning samples.

[0058] Therefore, a bearing operation database is pre-configured, and the bearing operation database may include relevant data of the bearing operation. By querying the database, abnormal bearing operation data can be obtained.

[0059] In some embodiments, when configuring the bearing operation database, an abnormal or normal identification is configured for the bearing operation data stored in the bearing operation database, so that abnormal bearing operation data can be directly queried based on the data identification.

[0060] In some embodiments, based on the abnormal bearing operation data, the associated bearing lubrication work data can be obtained. It can be understood that the abnormal bearing operation data also corresponds to the data recording time, and the lubrication work data also corresponds to the lubrication start time and the lubrication end time. The associated bearing lubrication work data can be determined by using this time relationship.

[0061] In some embodiments, the preset time relationship may be: the recording time of the abnormal bearing operation data is between the lubrication start time and the lubrication end time; or the recording time of the abnormal bearing operation data is before the lubrication start time or after the lubrication end time. Regarding before or after, the time interval condition needs to be met, for example: one hour. If the time interval is too long, there is no influence relationship.

[0062] Furthermore, data sampling is performed from the associated bearing lubrication work data to determine a plurality of initial second lubrication work data samples, wherein the data sampling rule may be random sampling, interval sampling, and the like.

[0063] It can be understood that the prediction label corresponding to the initial second lubrication work data sample is determined to be a fault.

[0064] Furthermore, sample expansion may be performed on a plurality of initial second lubrication work data samples to obtain the active learning samples.

[0065] The sample expansion method includes: changing at least one of the lubrication start time, lubrication end time and lubricant remaining amount, and changing the corresponding real label to obtain an extended sample. The changed value cannot be too large, so that the data is not referenceable.

[0066] It can be understood that since the active learning samples correspond to abnormal bearing operation data, the corresponding prediction label is the presence of a fault. Therefore, sample expansion is required to improve the richness of the samples.

[0067] The original training samples may be obtained by using a mature model training data method in the art, for example, by manual labeling.

[0068] Thus, the original training samples include: multiple first lubrication work data samples and multiple real labels corresponding to the first lubrication work data samples; the active learning samples include: multiple second lubrication work data samples and multiple real labels corresponding to the second lubrication work data samples.

[0069] On the basis of the two samples, the corresponding response values ​​can be determined respectively, so as to screen the failed samples based on the response values.

[0070] In some embodiments, the response value is used to characterize the bearing lubrication fault response speed.

[0071] The response value of the original training sample can be determined by querying a preset bearing operation database. For example, first query the abnormal bearing operation data that matches the lubrication time of the original training sample, then determine the normal bearing operation data whose recording time is after the abnormal bearing operation data, and determine the response value based on the time interval between the normal bearing operation data and the abnormal bearing operation data. Among them, the abnormal bearing operation data that matches the lubrication time of the original training sample can be recorded at the same time as the lubrication end time, or substantially the same time.

[0072] In some embodiments, the longer the time interval is, the larger the response value is, that is, the time interval is proportional to the response value, and the corresponding response value can be determined based on this proportional relationship.

[0073] In some embodiments, determining response values ​​corresponding to multiple second lubrication work data samples respectively includes: inputting multiple second lubrication work data samples into an originally trained support vector machine model to obtain a fault prediction result output by the originally trained support vector machine model; determining prediction labels corresponding to multiple second lubrication work data samples respectively based on the fault prediction result output by the originally trained support vector machine model; determining a model based on the prediction labels corresponding to multiple second lubrication work data samples respectively, the true labels corresponding to multiple second lubrication work data samples respectively, and the pre-trained response value determination, and determining the response values ​​corresponding to multiple second lubrication work data samples respectively.

[0074] Figure 4 is a flowchart of determining a response value of an active learning sample according to an exemplary embodiment. Figure 4 As shown, it is necessary to first input multiple second lubrication work data samples in the active learning samples into the originally trained support vector machine model to obtain the fault prediction result output by the originally trained support vector machine model; based on the fault prediction result, the prediction label can be determined; thus, combining the prediction labels corresponding to the multiple second lubrication work data samples and the true label, the pre-trained response value determination model is used to determine the response value.

[0075] In some embodiments, the pre-trained response value determination model can analyze the response value based on the true label and the predicted label.

[0076] The training process corresponding to the response value determination model may include: obtaining a training data set, the training data set including multiple third lubrication work data samples, and real fault labels and predicted fault labels corresponding to the multiple third lubrication work data samples respectively; performing a fault processing simulation according to the real fault labels corresponding to the multiple third lubrication work data samples respectively, to obtain a first response value; performing a fault processing simulation according to the predicted fault labels corresponding to the multiple third lubrication work data samples respectively, to obtain a second response value; determining the response value prediction labels corresponding to the multiple third lubrication work data samples respectively according to the first response value and the second response value; training the response value determination model to be trained according to the training data set and the response value prediction labels corresponding to the multiple third lubrication work data samples respectively, to obtain a pre-trained response value determination model.

[0077] In this embodiment, the method for obtaining multiple third lubrication work data samples can be the same as that of the original training samples. The real fault labels and predicted fault labels corresponding to the multiple third lubrication work data samples can be labeled by the user, or they can be determined according to different fault prediction methods. There may be differences between the two.

[0078] In the embodiment of the present application, the first lubrication work data sample, the second lubrication work data sample and the third lubrication work data sample may have some repeated samples, but most of the samples are non-repeated samples.

[0079] Furthermore, a fault handling simulation is performed according to real fault labels corresponding to the plurality of third lubrication work data samples to obtain a first response value, and a fault handling simulation is performed according to predicted fault labels corresponding to the plurality of third lubrication work data samples to obtain a second response value.

[0080] In some embodiments, fault handling simulation can be implemented based on a simulation model, such as a simulation model. In some embodiments, both the real fault label and the predicted fault label can represent the corresponding fault probability. Based on different fault probabilities, the response speed in the simulation model is different. Therefore, the first response value and the second response value can be determined according to the response speed of the simulation model.

[0081] Furthermore, according to the first response value and the second response value, response value prediction labels respectively corresponding to the plurality of third lubrication work data samples may be determined.

[0082] In some embodiments, the smaller response value between the first response value and the second response value may be determined as the response value prediction label. Alternatively, the first response value and the second response value may be summed and averaged, and the summed average may be determined as the response value prediction label.

[0083] Thus, the response value determination model to be trained is trained according to the response value prediction labels corresponding to the training data set and multiple third lubrication work data samples, so that the pre-trained response value determination model can predict the response value based on the real label and the predicted label.

[0084] In this embodiment, the response value tag is implemented based on a fault handling simulation.

[0085] In other embodiments, the response value label may also be a manually marked label, or a label obtained by other means, which is not limited here.

[0086] Further, according to the response values ​​respectively corresponding to the multiple first lubrication working data samples and the multiple second lubrication working data samples, the multiple first lubrication working data samples and the multiple second lubrication working data samples are screened for failure samples to determine the target active learning samples, including: sorting the multiple first lubrication working data samples according to the response values ​​respectively corresponding to the multiple first lubrication working data samples to obtain a first sorting result; sorting the multiple second lubrication working data samples according to the response values ​​respectively corresponding to the multiple second lubrication working data samples to obtain a second sorting result; sorting the multiple first lubrication working data samples and the multiple second lubrication working data samples according to the response values ​​respectively corresponding to the multiple first lubrication working data samples and the multiple second lubrication working data samples to obtain a third sorting result; sampling the lubrication working data samples according to the first sorting result, the second sorting result and the third sorting result, and determining the sampled lubrication working data samples as failure samples; and eliminating the failure samples to obtain the target active learning samples.

[0087] In this embodiment, the first sorting result may be a result obtained by sorting the response values ​​corresponding to the plurality of first lubrication work data samples in descending order, and the second sorting result may be a result obtained by sorting the response values ​​corresponding to the plurality of second lubrication work data samples in descending order. As for the third sorting result, the plurality of first lubrication work data samples and the plurality of second lubrication work data samples may be deduplicated first, and then the deduplicated work data samples may be sorted in descending order according to the response values ​​to obtain the third sorting result.

[0088] In some embodiments, the sampling rules for sampling the lubrication work data samples may include random sampling, pre-sequence sampling, post-sequence sampling, etc. Regarding random sampling, random sampling may be performed from the first sorting result, the second sorting result, and the third sorting result. Regarding pre-sequence sampling, sampling may be performed from samples with a higher order; regarding post-sequence sampling, sampling may be performed from samples with a lower order. The multiple samples sampled each time are different from each other. The last sampled failure samples do not include duplicate samples.

[0089] In some embodiments, the number of failed samples may be preset, for example, 1 / 4 of the total number of samples, and reasonable sampling is performed according to the number to eliminate failed samples.

[0090] Therefore, based on the target active learning samples, the support vector machine model is actively learned and updated, so that the generalization ability of the support vector machine model is improved and the accuracy is higher.

[0091] After the fault prediction result is obtained, it can be fed back to the relevant user for processing. Of course, if the fault prediction result indicates that there is no fault, no feedback is required.

[0092] In some embodiments, the wind turbine bearing lubrication fault prediction method may further include: if the fault prediction result indicates that there is no fault in the wind turbine bearing lubrication system, determining a reference fault prediction result based on the eliminated failed working data and a pre-trained support vector machine model; if the reference fault prediction result indicates that there is a fault in the wind turbine bearing lubrication system, determining the eliminated failed working data as a new active learning sample; and updating the pre-trained support vector machine model based on the new active learning sample.

[0093] In this implementation, since the aforementioned active learning occurs before the support vector machine model is applied, an active learning process after the application may be added, and the sample basis of this active learning is the failed working data eliminated in the early stage.

[0094] In addition, active learning after application is performed only when the fault prediction results indicate that there is no fault in the wind turbine bearing lubrication system. If a fault exists, there is no need for active learning again.

[0095] In some embodiments, the wind turbine bearing lubrication fault prediction method also includes: if the fault prediction result indicates that there is a fault in the wind turbine bearing lubrication system, a fault handling strategy is determined based on at least one piece of information including lubrication start time, lubrication end time and lubricant usage included in the target historical lubrication work data, and the fault handling strategy is executed.

[0096] In this implementation, a fault handling strategy can also be actively formulated and executed. The fault handling strategy can be targeted at the bearing lubrication system as a kind of fault pre-processing before the user handles the fault.

[0097] For example, determine the latest lubrication end time in the target historical lubrication work data, compare the time interval between the latest lubrication end time and the current time, if the time interval is long, then the lubrication pump can be turned off first to avoid lubrication again in the event of a fault. If the time interval is short, the ratio of lubricant usage to lubricant remaining can be compared. If the ratio is large, then extend the time for the next lubrication work to avoid lubrication in the event of a fault as much as possible. If the ratio is small, then it can be temporarily ignored and wait for the user to handle the fault.

[0098] Alternatively, other fault handling strategies may be used and may be formulated in advance. When a fault is detected, a reasonable handling strategy is selected from the pre-formulated fault handling strategies according to the specific situation.

[0099] Through the above technical scheme, the working data corresponding to the wind turbine bearing lubrication system is obtained, and the working data includes multiple historical lubrication working data. Based on the multiple historical lubrication working data, the failed working data is eliminated to obtain the target historical lubrication working data; the fault prediction result of the bearing lubrication system is determined through the pre-trained support vector machine model and the target historical lubrication working data; wherein, the training data of the pre-trained support vector machine model includes the training data obtained by active learning and failure sample screening, so that the accuracy of the support vector machine model is higher, and thus the accuracy of the fault prediction result is also higher. Therefore, this technical scheme can realize accurate and effective wind turbine bearing lubrication fault prediction based on active learning.

[0100] Figure 5 is a block diagram of a wind turbine bearing lubrication fault prediction device 500 based on active learning according to an exemplary embodiment. Figure 5 As shown, the device comprises:

[0101] The acquisition module 501 is used to acquire the working data corresponding to the lubrication system of the wind turbine bearing, and the working data includes: multiple historical lubrication working data.

[0102] The fault prediction module 502 is used to eliminate failed working data of the working data corresponding to the wind turbine bearing lubrication system to obtain target historical lubrication working data; determine the fault prediction result corresponding to the wind turbine bearing lubrication system based on the target historical lubrication working data and a pre-trained support vector machine model, wherein the training data of the pre-trained support vector machine model includes training data obtained through active learning and failure sample screening.

[0103] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0104] Figure 6 FIG. 6 is a block diagram of an electronic device 600 according to an exemplary embodiment. Figure 6 As shown, the electronic device 600 may include: a processor 601 , a memory 602 . The electronic device 600 may also include one or more of a multimedia component 603 , an input / output (I / O) interface 604 , and a communication component 605 .

[0105] The processor 601 is used to control the overall operation of the electronic device 600 to complete all or part of the steps in the above-mentioned active learning-based wind turbine bearing lubrication fault prediction method. The memory 602 is used to store various types of data to support the operation of the electronic device 600, and these data may include, for example, instructions for any application or method used to operate on the electronic device 600, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 602 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. The multimedia component 603 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, which is used to receive external audio signals. The received audio signal may be further stored in the memory 602 or sent through the communication component 605. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 604 provides an interface between the processor 601 and other interface modules, and the above-mentioned other interface modules may be a keyboard, a mouse, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 605 is used for wired or wireless communication between the electronic device 600 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 605 may include: Wi-Fi module, Bluetooth module, NFC module.

[0106] In an exemplary embodiment, the electronic device 600 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned active learning-based wind turbine bearing lubrication fault prediction method.

[0107] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned wind turbine bearing lubrication fault prediction method based on active learning are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 602 including program instructions, and the above-mentioned program instructions can be executed by the processor 601 of the electronic device 600 to complete the above-mentioned wind turbine bearing lubrication fault prediction method based on active learning.

[0108] In another exemplary embodiment, a computer program product is further provided. The computer program product includes a computer program executable by a processor. When the computer program is executed by the processor, the steps of the above-mentioned active learning-based wind turbine bearing lubrication fault prediction method are implemented.

[0109] The preferred embodiments of the present application are described in detail above in conjunction with the accompanying drawings; however, the present application is not limited to the specific details in the above embodiments. Within the technical concept of the present application, a variety of simple modifications can be made to the technical solution of the present application, and these simple modifications all fall within the protection scope of the present application.

[0110] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this application will not further describe various possible combinations.

[0111] In addition, the various implementation modes of the present application may be arbitrarily combined, and as long as they do not violate the concept of the present application, they should also be regarded as the contents disclosed in the present application.

Claims

1. A wind turbine bearing lubrication fault prediction method based on active learning, characterized in that: include: Obtaining working data corresponding to the lubrication system of the wind turbine bearing, the working data including: multiple historical lubrication working data, each historical lubrication working data including: lubrication start time, lubrication end time and lubricant usage; Eliminating failed working data from working data corresponding to the lubrication system of the wind turbine bearing to obtain target historical lubrication working data; Determine a fault prediction result corresponding to the wind turbine bearing lubrication system according to the target historical lubrication work data and a pre-trained support vector machine model, wherein the training data of the pre-trained support vector machine model includes training data obtained through active learning and failure sample screening; The training process of the pre-trained support vector machine model includes: Acquire original training samples, where the original training samples include a plurality of first lubrication work data samples and real labels corresponding to the plurality of first lubrication work data samples respectively; Training the support vector machine model to be trained according to the original training samples to obtain the original trained support vector machine model; Acquire active learning samples, wherein the active learning samples include a plurality of second lubrication work data samples and real labels corresponding to the plurality of second lubrication work data samples respectively; Determine the response values ​​corresponding to the plurality of first lubrication operation data samples, and determine the response values ​​corresponding to the plurality of second lubrication operation data samples, wherein any response value is used to characterize the bearing lubrication fault response speed; According to the response values ​​respectively corresponding to the plurality of first lubrication work data samples and the response values ​​respectively corresponding to the plurality of second lubrication work data samples, the plurality of first lubrication work data samples and the plurality of second lubrication work data samples are screened for failure samples to determine a target active learning sample; The originally trained support vector machine model is updated and trained according to the target active learning sample to obtain the pre-trained support vector machine model.

2. The wind turbine bearing lubrication fault prediction method according to claim 1, characterized in that: The step of eliminating invalid working data from the working data corresponding to the bearing lubrication system of the wind turbine generator to obtain target historical lubrication working data includes: Determine the total lubrication durations corresponding to the multiple historical lubrication work data respectively according to the lubrication start time and lubrication end time respectively included in the multiple historical lubrication work data; Determine the benchmark total lubrication time based on the total lubrication time corresponding to multiple historical lubrication work data; Determine the base lubricant dosage according to the lubricant dosage respectively included in multiple historical lubrication work data; According to the benchmark total lubrication time and the benchmark lubricant usage, invalid working data is eliminated from the working data corresponding to the wind turbine bearing lubrication system to obtain target historical lubrication working data.

3. The wind turbine bearing lubrication fault prediction method according to claim 2, characterized in that: The step of eliminating invalid working data corresponding to the working data of the wind turbine bearing lubrication system according to the benchmark total lubrication time and the benchmark lubricant usage includes: Comparing the total lubrication durations respectively corresponding to the multiple historical lubrication work data with the benchmark total lubrication duration, and comparing the lubricant usages respectively included in the multiple historical lubrication work data with the benchmark lubricant usage; Eliminate historical lubrication work data whose total lubrication time is less than the benchmark total lubrication time and whose lubricant usage is less than the benchmark lubricant usage; and / or eliminate historical lubrication work data whose total lubrication time is greater than or equal to the benchmark total lubrication time and whose lubricant usage is greater than or equal to the benchmark lubricant usage.

4. The method for predicting lubrication failure of a wind turbine bearing according to claim 1, characterized in that: The obtaining of active learning samples includes: Query the preset bearing operation database to obtain abnormal bearing operation data; Acquire bearing lubrication work data associated with the abnormal bearing operation data, wherein the lubrication start time and lubrication end time corresponding to the associated bearing lubrication work data satisfy a preset time relationship with the recording time of the abnormal bearing operation data; Determining a plurality of initial second lubrication operation data samples from the associated bearing lubrication operation data; The multiple initial second lubrication work data samples are subjected to sample expansion to obtain the active learning samples.

5. The wind turbine bearing lubrication fault prediction method according to claim 1, characterized in that: The determining of the response values ​​respectively corresponding to the plurality of second lubrication operation data samples comprises: Inputting the plurality of second lubrication work data samples into the originally trained support vector machine model to obtain a fault prediction result output by the originally trained support vector machine model; Determining prediction labels corresponding to the plurality of second lubrication work data samples respectively according to the fault prediction results output by the originally trained support vector machine model; A model is determined according to the predicted labels respectively corresponding to the multiple second lubrication work data samples, the real labels respectively corresponding to the multiple second lubrication work data samples and the pre-trained response value, and the response values ​​respectively corresponding to the multiple second lubrication work data samples are determined.

6. The wind turbine bearing lubrication fault prediction method according to claim 5, characterized in that: The wind turbine bearing lubrication fault prediction method further includes: Acquire a training data set, wherein the training data set includes a plurality of third lubrication work data samples, and real fault labels and predicted fault labels respectively corresponding to the plurality of third lubrication work data samples; Performing a fault processing simulation according to the real fault labels respectively corresponding to the plurality of third lubrication work data samples to obtain a first response value; Performing a fault processing simulation according to the predicted fault labels respectively corresponding to the plurality of third lubrication work data samples to obtain a second response value; Determining response value prediction labels corresponding to the plurality of third lubrication work data samples respectively according to the first response value and the second response value; The response value determination model to be trained is trained according to the response value prediction labels respectively corresponding to the training data set and the plurality of third lubrication work data samples to obtain the pre-trained response value determination model.

7. The wind turbine bearing lubrication fault prediction method according to claim 1, characterized in that: The method of screening the plurality of first lubrication work data samples and the plurality of second lubrication work data samples for failure samples according to the response values ​​respectively corresponding to the plurality of first lubrication work data samples and the response values ​​respectively corresponding to the plurality of second lubrication work data samples, and determining the target active learning sample, comprises: sorting the plurality of first lubrication work data samples according to the response values ​​respectively corresponding to the plurality of first lubrication work data samples to obtain a first sorting result; sorting the plurality of second lubrication work data samples according to the response values ​​respectively corresponding to the plurality of second lubrication work data samples to obtain a second sorting result; sorting the plurality of first lubrication operation data samples and the plurality of second lubrication operation data samples according to the response values ​​respectively corresponding to the plurality of first lubrication operation data samples and the response values ​​respectively corresponding to the plurality of second lubrication operation data samples to obtain a third sorting result; Sampling lubrication work data samples according to the first sorting result, the second sorting result and the third sorting result, and determining the sampled lubrication work data samples as failure samples; The failed samples are eliminated to obtain the target active learning samples.

8. The method for predicting lubrication failure of a wind turbine bearing according to claim 1, characterized in that: The wind turbine bearing lubrication fault prediction method further includes: If the fault prediction result indicates that there is no fault in the wind turbine bearing lubrication system, determining a reference fault prediction result according to the eliminated failed working data and the pre-trained support vector machine model; If the reference fault prediction result indicates that there is a fault in the lubrication system of the wind turbine bearing, the eliminated failed working data is determined as a new active learning sample; Based on the new active learning sample, the pre-trained support vector machine model is updated.

9. The wind turbine bearing lubrication fault prediction method according to claim 1, characterized in that: The wind turbine bearing lubrication fault prediction method further includes: If the fault prediction result indicates that there is a fault in the wind turbine bearing lubrication system, a fault handling strategy is determined based on at least one piece of information including lubrication start time, lubrication end time and lubricant usage included in the target historical lubrication work data, and the fault handling strategy is executed.

Images