WO2024077753A1

WO2024077753A1 - Wind turbine fault prediction method based on multi-source heterogeneous data

Info

Publication number: WO2024077753A1
Application number: PCT/CN2022/137737
Authority: WO
Inventors: 杨之乐; 安钊; 郭媛君; 胡天宇; 吴承科
Original assignee: 深圳先进技术研究院
Priority date: 2022-10-10
Filing date: 2022-12-08
Publication date: 2024-04-18
Also published as: CN115293057A; CN115293057B

Abstract

Disclosed in the present invention is a wind turbine fault prediction method based on multi-source heterogeneous data. The method comprises: acquiring historical data of a target wind turbine, wherein the historical data of the target wind turbine at least comprises deformation data of each blade of the target wind turbine within a preset time period and vibration data of the target wind turbine, and the preset time period is a preset duration before the current moment; and inputting the historical data of the target wind turbine into a trained neural network, and acquiring a fault prediction result of the target wind turbine outputted by the neural network, wherein the neural network is trained on the basis of a training data set, the training data set comprises sample historical data and sample expansion data, the sample historical data is actually collected historical data of the wind turbine, and the sample expansion data is data generated by performing sample expansion processing on the sample historical data. The present invention can improve the fault detection efficiency of the wind turbine.

Description

A wind turbine fault prediction method based on multi-source heterogeneous data

Technical Field

The present invention relates to the field of new energy technology, and in particular to a wind turbine fault prediction method based on multi-source heterogeneous data.

Background technique

Existing wind turbines require regular manual inspections to troubleshoot wind turbine faults, but manual inspections require the wind turbines to be shut down for maintenance, which is not only inconvenient but also wastes clean energy.

Therefore, the prior art still needs to be improved and enhanced.

technical problem

In view of the above-mentioned defects of the prior art, the present invention provides a wind turbine fault prediction method based on multi-source heterogeneous data, aiming to solve the problem of low efficiency of fault detection of wind turbines requiring manual inspection in the prior art.

Technical Solutions

In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

A first aspect of the present invention provides a wind turbine fault prediction method based on multi-source heterogeneous data, the method comprising:

Acquire historical data of a target wind turbine, wherein the historical data of the target wind turbine at least includes deformation data of each blade of the target wind turbine within a preset time period and vibration data of the target wind turbine, wherein the preset time period is a time period of a preset duration before a current moment;

Inputting historical data of the target wind turbine into a trained neural network, and obtaining a fault prediction result of the target wind turbine output by the neural network;

Wherein, the neural network is trained based on a training data set, and the training data set includes sample historical data and sample expansion data, the sample historical data is historical data of wind turbines actually collected, and the sample expansion data is data generated by performing sample expansion processing on the sample historical data.

The wind turbine fault prediction method based on multi-source heterogeneous data, wherein the acquiring of historical data of the target wind turbine comprises:

The nose position of the target wind turbine is acquired, and deformation data of each blade of the target wind turbine is acquired according to the nose position and a reflection signal of a light signal sent to the target wind turbine.

The wind turbine fault prediction method based on multi-source heterogeneous data, wherein the step of acquiring deformation data of each blade of the target wind turbine according to the position of the wind turbine head and the reflection signal of the light signal sent to the target wind turbine, comprises:

Acquire a nose reference position of the target wind turbine and reference reflection signal data corresponding to the nose reference position;

The deformation data of each blade of the target wind turbine is acquired according to the difference between the nose position and the nose reference position, and the reflected signal of the light signal sent to the target wind turbine and the reference reflected signal data.

The wind turbine fault prediction method based on multi-source heterogeneous data, wherein, after obtaining the deformation data of each blade of the target wind turbine according to the position of the wind turbine head and the reflection signal of the light signal sent to the target wind turbine, comprises:

According to the current blade deformation data of the target wind turbine, updating the three-dimensional model in the digital twin model of the target wind turbine based on the current blade deformation data of the target wind turbine;

After obtaining the fault prediction result of the target wind turbine generator output by the neural network, the method further comprises:

The fault prediction result of the target wind turbine is updated into the digital twin model of the target wind turbine.

The wind turbine fault prediction method based on multi-source heterogeneous data, wherein the neural network includes a feature extraction module and a prediction module, and the training process of the neural network is:

Select a portion of the sample history data from each of the sample history data to form a target training batch, and perform the following steps for the target sample history data in the target training batch:

Inputting the target sample historical data into a sample expansion module to generate target sample expansion data;

The feature extraction module extracts the features of the target sample historical data and the target sample expanded data respectively to obtain a first feature and a second feature;

Inputting the first feature and the second feature into the prediction module respectively, and obtaining a first prediction result and a second prediction result output by the prediction module;

Obtaining a batch training loss according to the first feature, the second feature, the first prediction result, the second prediction result, and the fault label corresponding to each sample historical data in the target training batch;

The parameters of the sample expansion module, the feature extraction module and the prediction module are updated according to the batch training loss, and the step of selecting part of the sample historical data from each of the sample historical data to form a target training batch is re-executed until the parameters converge.

The wind turbine fault prediction method based on multi-source heterogeneous data, wherein the batch training loss is obtained according to the first feature, the second feature, the first prediction result, the second prediction result and the fault label corresponding to each sample historical data in the target training batch, including:

Obtaining a first loss according to the fault label corresponding to the target sample historical data and the first prediction result;

Inputting the first feature and the second feature into a discriminator respectively, obtaining a discrimination result output by the discriminator, and obtaining a second loss according to the discrimination result;

Obtaining a sample loss corresponding to the target sample historical data according to the first loss and the second loss;

Obtaining the probability distribution of the first fault prediction result corresponding to each sample historical data in the training batch as a first probability distribution, and obtaining the probability distribution of the second fault prediction result corresponding to each sample historical data in the training batch as a second probability distribution;

Obtaining a first batch loss according to the first probability distribution and the second probability distribution;

The sample losses corresponding to each sample historical data in the training batch are summed to obtain the second batch loss;

Obtaining the batch training loss according to the first batch loss and the second batch loss;

The updating of the parameters of the sample expansion module, the feature extraction module and the prediction module according to the batch training loss includes:

The parameters of the sample expansion module, the feature extraction module, the prediction module and the discriminator are updated according to the batch training loss.

The wind turbine fault prediction method based on multi-source heterogeneous data, wherein the features of the target sample historical data and the target sample expanded data are respectively extracted by the feature extraction module to obtain the first feature and the second feature, including:

Respectively performing dimensionality reduction processing on the target sample historical data and the target sample expanded data to obtain first dimensionality reduction data and second dimensionality reduction data;

The first dimension reduction data and the second dimension reduction data are respectively input into the feature extraction module to obtain the first feature and the second feature output by the feature extraction module.

A second aspect of the present invention provides a wind turbine fault prediction device based on multi-source heterogeneous data, comprising:

A data acquisition module, the data acquisition module is used to acquire historical data of a target wind turbine, the historical data of the target wind turbine at least including deformation data of each blade of the target wind turbine within a preset time period and vibration data of the target wind turbine, the preset time period being a period of preset duration before a current moment;

A prediction module, the prediction module is used to input the historical data of the target wind turbine into a trained neural network, and obtain a fault prediction result of the target wind turbine output by the neural network;

According to a third aspect of the present invention, a terminal is provided, comprising a processor and a computer-readable storage medium communicatively connected to the processor, wherein the computer-readable storage medium is suitable for storing a plurality of instructions, and the processor is suitable for calling the instructions in the computer-readable storage medium to execute the steps of implementing any one of the above-mentioned methods for predicting wind turbine faults based on multi-source heterogeneous data.

A fourth aspect of the present invention provides a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the steps of any of the above-mentioned methods for predicting wind turbine faults based on multi-source heterogeneous data.

Beneficial Effects

Compared with the prior art, the present invention provides a wind turbine fault prediction method based on multi-source heterogeneous data, which predicts wind turbine faults through a neural network. In order to address the problem of insufficient real label data of existing wind turbines, the real data is expanded and processed, so that the amount of data in the training data set of the neural network is larger, thereby improving the accuracy of fault prediction of the trained neural network. There is no need to manually inspect the wind turbine to troubleshoot the fault, thereby improving the fault detection efficiency of the wind turbine.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of an embodiment of a method for predicting wind turbine faults based on multi-source heterogeneous data provided by the present invention;

FIG2 is a structural principle diagram of an embodiment of a wind turbine fault prediction device based on multi-source heterogeneous data provided by the present invention;

FIG3 is a schematic diagram showing the principles of an embodiment of a terminal provided by the present invention.

Embodiments of the present invention

In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

The wind turbine fault prediction method based on multi-source heterogeneous data provided by the present invention can be applied to a terminal with computing capabilities. The terminal can execute the wind turbine fault prediction method based on multi-source heterogeneous data provided by the present invention to perform power grid peak-shaving scheduling. The terminal can be but is not limited to various computers, mobile terminals, smart home appliances, wearable devices, etc.

Embodiment 1

As shown in FIG1 , in one embodiment of the wind turbine fault prediction method based on multi-source heterogeneous data, the steps include:

S100, acquiring historical data of a target wind turbine, wherein the historical data of the target wind turbine at least includes deformation data of each blade of the target wind turbine within a preset time period and vibration data of the target wind turbine, wherein the preset time period is a period of preset duration before a current moment.

The preset time length may be one week, one month, etc., and the preset time length may be determined through experiments, that is, different time lengths are used to conduct experiments using the method provided in this embodiment to obtain a more accurate fault prediction result as the preset time length.

The vibration data in the historical data of the target wind turbine may be obtained by a sensor installed on the target wind turbine. The blade deformation data in the historical data of the target wind turbine may be obtained by optical signal detection. Specifically, the historical data of the target wind turbine may be obtained by:

In this embodiment, an optical signal generator and an optical signal detector may be arranged around the target wind turbine. The number of the optical signal generator and the optical signal detector may be multiple so that the detection range covers the entire blade of the target wind turbine. The optical signal generator sends an optical signal to the target wind turbine and is reflected and received by the optical signal detector. The optical signal detector receives the reflected signal and analyzes it to obtain the deformation data of the blade. Specifically, the reflected signal pre-set when the blade of the wind turbine is not deformed can be used as a reference, and the deformation data is determined according to the difference between the actually received reflected signal and the reflected signal when it is not deformed. During the operation of the wind turbine, as the wind direction changes, the head of the wind turbine will rotate accordingly to keep the blade in the right wind. Therefore, in this embodiment, when obtaining the deformation data of the blade, it is also necessary to combine the real-time head position of the wind turbine.

The step of acquiring deformation data of each blade of the target wind turbine according to the position of the wind turbine head and the reflected signal of the light signal sent to the target wind turbine includes:

The deformation data of each blade of the target wind turbine is acquired according to the difference between the nose position and the nose reference position, and the reflection signal of the light signal sent to the target wind turbine and the standard reflection signal data.

Specifically, the nose reference position refers to the pre-set position of the nose of the target wind turbine relative to the optical signal generator and the optical signal detector. The positions of the optical signal generator and the optical signal detector remain unchanged, while the nose position of the target wind turbine will change with different wind directions. The reflected signal data received by the optical signal detector after the optical signal emitted by the optical signal generator is reflected by the blades of the target wind turbine when the nose of the target wind turbine is at the nose reference position is obtained in advance as the reference reflected signal data. According to the difference between the nose position and the nose reference position and the reference transmitted signal, the reflected signal of the optical signal sent to the blades of the target wind turbine when the blades of the target wind turbine are not deformed can be obtained. By comparing with the actually received reflected signal, the deformation data of the blades of the target wind turbine can be determined.

After acquiring deformation data of each blade of the target wind turbine according to the position of the wind turbine head and the reflected signal of the light signal sent to the target wind turbine, the method includes:

According to the current blade deformation data of the target wind turbine, the three-dimensional model in the digital twin model of the target wind turbine is updated based on the current blade deformation data of the target wind turbine.

In order to facilitate the health management of wind turbines, in this embodiment, a digital twin model of the target wind turbine is set up, and the digital twin model includes a three-dimensional model of the target wind turbine. After obtaining the current blade deformation data of the target wind turbine, the three-dimensional model of the target wind turbine is updated according to the blade deformation data, so that maintenance personnel can intuitively obtain the deformation condition of the blades through the three-dimensional model.

Please refer to FIG. 1 again. The method provided in this embodiment further includes the steps of:

S200: Inputting historical data of the target wind turbine into a trained neural network, and obtaining a fault prediction result of the target wind turbine output by the neural network.

In the prior art, there is no mature database on wind turbines. The amount of real data on wind turbines is relatively small, while the training of neural network models requires a large amount of data. Using a small amount of data to train the neural network model will lead to unsatisfactory training results of the model, affecting the accuracy of fault prediction. To address this problem, the method provided in this embodiment uses sample expansion to expand the data set. Specifically, in this embodiment, the neural network is trained based on the training data set, and the training data set includes sample historical data and sample expansion data. The sample historical data is the historical data of the wind turbine that is actually collected, and the sample expansion data is the data generated by performing sample expansion processing on the sample historical data.

The neural network includes a feature extraction module and a prediction module, and the training process of the neural network is:

S001. Selecting a portion of the sample history data from each of the sample history data to form a target training batch, and executing the following steps for the target sample history data in the target training batch:

S002, inputting the target sample historical data into a sample expansion module to generate target sample expansion data;

S003, extracting features of the target sample historical data and the target sample expanded data respectively through the feature extraction module to obtain a first feature and a second feature;

S004, inputting the first feature and the second feature into the prediction module respectively, and obtaining a first prediction result and a second prediction result output by the prediction module;

S005, obtaining a batch training loss according to the first feature, the second feature, the first prediction result, the second prediction result, and the fault label corresponding to each sample historical data in the target training batch;

S006. Update the parameters of the sample expansion module, the feature extraction module and the prediction module according to the batch training loss, and re-execute the step of selecting part of the sample historical data from each of the sample historical data to form a target training batch until the parameters converge.

In the existing data expansion schemes, most of them use adversarial networks to expand samples according to labels, and use adversarial networks to train separately. For example, data with A label (taking the fault prediction task in this application as an example, A label is a fault category label) is generated through adversarial network training to achieve data expansion. However, in the case of a small amount of data, for example, there are multiple fault category labels, and the actual amount of labeled data of all wind turbines is not much, and the amount of data split into each fault category is even less. In this way, the training process of the adversarial network is not ideal, and the fault prediction network trained with the expanded data generated by the adversarial network after training is not high in the final prediction accuracy. In this embodiment, a new training method for a data expansion network is proposed, and the data expansion network and the fault prediction network are jointly trained.

Specifically, in this embodiment, when performing data expansion, the label of a single data is not restricted, but a part of all the real labeled data is selected. The sample historical data is the real historical data of the wind turbine, and the sample historical data has corresponding labels, that is, the actual fault type of the wind turbine and the corresponding real historical data when the fault type occurs are collected to obtain the sample historical data and the corresponding fault label label. The selected part of the sample historical data is used as the target training batch, and each of the sample historical data in the target training batch is input into the sample expansion module respectively. Through the above steps S002-S005, the batch training loss of the target training batch is obtained, and then the parameters of each network module are updated based on the batch training loss, and then part of the sample historical data is randomly removed as the new target training batch, and this is iterated multiple times until the parameters converge. In this process, the expanded data is not labeled. Instead, the outputs of multiple network modules are used to calculate multiple losses to update the parameters of the sample expansion module, so that the sample expansion module can learn the essence of the feature space where data of different fault types are located. Compared with the singleness of the loss of the existing adversarial network, the constraints of multiple types of losses on the update of module parameters can make up for the defect of small data volume and prevent the parameter optimization results of the network model from falling into the local optimum, resulting in reduced accuracy.

Specifically, obtaining the batch training loss according to the first feature, the second feature, the first prediction result, the second prediction result and the fault label corresponding to each sample historical data in the target training batch includes:

The batch training loss is obtained according to the first batch loss and the second batch loss.

In this embodiment, multiple losses are set to constrain the optimization direction of the network module parameters. The first loss can enable the feature extraction module and the prediction module to learn the intrinsic relationship between the historical data and faults of the wind turbine, and the second loss can enable the feature extraction module and the sample expansion module to learn the feature space of the real data of the wind turbine for predicting faults, so that the generated expanded data can be consistent with the real data in terms of the features used to predict faults. Since the sample expansion data does not carry labels, in order to use the generated sample expansion data to train the prediction module and further improve the fault prediction ability of the prediction module, in this embodiment, the module parameters are also updated by constraining the distance between the probability distribution of the fault prediction results of the real data and the probability distribution of the fault prediction results of the expanded data.

Specifically, for all the sample historical data in each of the training batches during the training process, the first probability distribution is obtained according to the corresponding first prediction results, and for all the expanded sample data generated in each of the training batches, the corresponding second features are input into the prediction module, and the second probability distribution is obtained based on the multiple second prediction results output by the prediction module. If the expanded data generated by the sample expansion module has the same feature space as the real data, and the prediction module has fully learned the intrinsic connection between the feature space and the fault category, then the first probability distribution and the second probability distribution should be consistent. Therefore, calculating the first batch distribution loss based on the difference between the first probability distribution and the second probability distribution for updating the module parameters can effectively improve the accuracy of the model.

From the foregoing description, it can be seen that for each sample historical data in the target training batch, the corresponding first loss and second loss can be obtained, the first loss and the second loss are summed to obtain a sample loss corresponding to the sample historical data, the sample losses corresponding to each sample historical data in the target training batch are summed to obtain the second batch loss, the first batch loss and the second batch loss are summed to obtain the batch training loss corresponding to the target training batch, and the parameters of the sample expansion module, the feature extraction module, the prediction module and the discriminator are updated based on the batch training loss.

Furthermore, in this embodiment, each time the sample historical data is selected to form the target training batch, it is selected randomly, that is, the data combination in the target training batch used to update the module parameters each time is different, so that the probability distribution is also different each time, which can have a similar effect as expanding the training data.

Furthermore, since the amount of data of the historical data is large and the dimension is large, in order to reduce the amount of calculation of the neural network, in this embodiment, before the historical data of the target wind turbine is input into the trained neural network, the historical data of the target wind turbine is also subjected to dimensionality reduction processing. Similarly, the features of the target sample historical data and the target sample expanded data are respectively extracted by the feature extraction module to obtain the first feature and the second feature, including:

After the neural network training is completed, the fault prediction result of the target wind turbine generator can be input into the trained neural network, and the neural network outputs the fault prediction result of the target wind turbine generator. In order to further facilitate maintenance personnel to obtain the fault prediction result, after obtaining the fault prediction result of the target wind turbine generator output by the neural network, the method includes:

Displaying the fault prediction result in the digital twin model of the target wind turbine may be based on the components corresponding to the fault prediction result, and displaying the fault prediction result on the three-dimensional model of the corresponding components, which is more intuitive.

In summary, this embodiment provides a wind turbine fault prediction method based on multi-source heterogeneous data, which predicts wind turbine faults through a neural network, and in order to address the problem of insufficient real label data of existing wind turbines, expands the real data, so that the amount of data in the training data set of the neural network is larger, thereby improving the accuracy of fault prediction of the trained neural network, eliminating the need for manual inspection of wind turbines to troubleshoot faults, and improving the fault detection efficiency of wind turbines.

It should be understood that, although the steps in the flowcharts given in the accompanying drawings of the present specification are displayed in sequence according to the indications of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Embodiment 2

Based on the above embodiments, the present invention further provides a wind turbine fault prediction device based on multi-source heterogeneous data. As shown in FIG2 , the wind turbine fault prediction device based on multi-source heterogeneous data includes:

A data acquisition module, the data acquisition module is used to acquire historical data of a target wind turbine, the historical data of the target wind turbine at least includes deformation data of each blade of the target wind turbine within a preset time period and vibration data of the target wind turbine, the preset time period is a period of preset duration before the current moment, as described in the first embodiment;

A prediction module, the prediction module is used to input the historical data of the target wind turbine into a trained neural network, and obtain a fault prediction result of the target wind turbine output by the neural network, as specifically described in the first embodiment;

Wherein, the neural network is trained based on a training data set, and the training data set includes sample historical data and sample expansion data, the sample historical data is historical data of wind turbines actually collected, and the sample expansion data is data generated by performing sample expansion processing on the sample historical data, as specifically described in Example 1.

Embodiment 3

Based on the above embodiments, the present invention also provides a terminal, as shown in Figure 3, the terminal includes a processor 10 and a memory 20. Figure 3 only shows some components of the terminal, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead.

In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Further, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a wind turbine fault prediction program 30 based on multi-source heterogeneous data is stored on the memory 20, and the wind turbine fault prediction program 30 based on multi-source heterogeneous data can be executed by the processor 10, thereby realizing the wind turbine fault prediction method based on multi-source heterogeneous data in the present application.

In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other chip, used to run the program code or process data stored in the memory 20, such as executing the wind turbine fault prediction method based on multi-source heterogeneous data.

In one embodiment, when the processor 10 executes the wind turbine fault prediction program 30 based on multi-source heterogeneous data in the memory 20, the following steps are implemented:

Embodiment 4

The present invention also provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the wind turbine fault prediction method based on multi-source heterogeneous data as described above.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

A wind turbine fault prediction method based on multi-source heterogeneous data, characterized in that the method comprises:

Acquire historical data of a target wind turbine, wherein the historical data of the target wind turbine at least includes deformation data of each blade of the target wind turbine within a preset time period and vibration data of the target wind turbine, wherein the preset time period is a time period of a preset duration before a current moment;

Inputting historical data of the target wind turbine into a trained neural network, and obtaining a fault prediction result of the target wind turbine output by the neural network;

Wherein, the neural network is trained based on a training data set, and the training data set includes sample historical data and sample expansion data, the sample historical data is historical data of wind turbines actually collected, and the sample expansion data is data generated by performing sample expansion processing on the sample historical data.
The method for predicting wind turbine faults based on multi-source heterogeneous data according to claim 1 is characterized in that the step of obtaining historical data of the target wind turbine comprises:

The nose position of the target wind turbine is acquired, and deformation data of each blade of the target wind turbine is acquired according to the nose position and a reflection signal of a light signal sent to the target wind turbine.
The wind turbine fault prediction method based on multi-source heterogeneous data according to claim 2 is characterized in that the deformation data of each blade of the target wind turbine is obtained according to the head position and the reflection signal of the light signal sent to the target wind turbine, including:

Acquire a nose reference position of the target wind turbine and reference reflection signal data corresponding to the nose reference position;

The deformation data of each blade of the target wind turbine is acquired according to the difference between the nose position and the nose reference position, and the reflected signal of the light signal sent to the target wind turbine and the reference reflected signal data.
The wind turbine fault prediction method based on multi-source heterogeneous data according to claim 2 is characterized in that after obtaining the deformation data of each blade of the target wind turbine according to the head position and the reflection signal of the light signal sent to the target wind turbine, it includes:

According to the current blade deformation data of the target wind turbine, updating the three-dimensional model in the digital twin model of the target wind turbine based on the current blade deformation data of the target wind turbine;

After obtaining the fault prediction result of the target wind turbine generator output by the neural network, the method further comprises:

The fault prediction result of the target wind turbine is updated into the digital twin model of the target wind turbine.
The wind turbine fault prediction method based on multi-source heterogeneous data according to claim 1 is characterized in that the neural network includes a feature extraction module and a prediction module, and the training process of the neural network is:

Select a portion of the sample history data from each of the sample history data to form a target training batch, and perform the following steps for the target sample history data in the target training batch:

Inputting the target sample historical data into a sample expansion module to generate target sample expansion data;

The feature extraction module extracts the features of the target sample historical data and the target sample expanded data respectively to obtain a first feature and a second feature;

Inputting the first feature and the second feature into the prediction module respectively, and obtaining a first prediction result and a second prediction result output by the prediction module;

Obtaining a batch training loss according to the first feature, the second feature, the first prediction result, the second prediction result, and the fault label corresponding to each sample historical data in the target training batch;

The parameters of the sample expansion module, the feature extraction module and the prediction module are updated according to the batch training loss, and the step of selecting part of the sample historical data from each of the sample historical data to form a target training batch is re-executed until the parameters converge.
The wind turbine fault prediction method based on multi-source heterogeneous data according to claim 5 is characterized in that the batch training loss is obtained according to the first feature, the second feature, the first prediction result, the second prediction result and the fault label corresponding to the target sample historical data corresponding to each sample historical data in the target training batch, including:

Obtaining a first loss according to the fault label corresponding to the target sample historical data and the first prediction result;

Inputting the first feature and the second feature into a discriminator respectively, obtaining a discrimination result output by the discriminator, and obtaining a second loss according to the discrimination result;

Obtaining a sample loss corresponding to the target sample historical data according to the first loss and the second loss;

Obtaining the probability distribution of the first fault prediction result corresponding to each sample historical data in the training batch as a first probability distribution, and obtaining the probability distribution of the second fault prediction result corresponding to each sample historical data in the training batch as a second probability distribution;

Obtaining a first batch loss according to the first probability distribution and the second probability distribution;

The sample losses corresponding to each sample historical data in the training batch are summed to obtain the second batch loss;

Obtaining the batch training loss according to the first batch loss and the second batch loss;

The updating of the parameters of the sample expansion module, the feature extraction module and the prediction module according to the batch training loss includes:

The parameters of the sample expansion module, the feature extraction module, the prediction module and the discriminator are updated according to the batch training loss.
The method for predicting wind turbine faults based on multi-source heterogeneous data according to claim 5 is characterized in that the feature extraction module extracts the features of the target sample historical data and the target sample expanded data respectively to obtain the first feature and the second feature, including:

Respectively performing dimensionality reduction processing on the target sample historical data and the target sample expanded data to obtain first dimensionality reduction data and second dimensionality reduction data;

The first dimension reduction data and the second dimension reduction data are respectively input into the feature extraction module to obtain the first feature and the second feature output by the feature extraction module.
A wind turbine fault prediction device based on multi-source heterogeneous data, characterized by comprising:

A data acquisition module, the data acquisition module is used to acquire historical data of a target wind turbine, the historical data of the target wind turbine at least including deformation data of each blade of the target wind turbine within a preset time period and vibration data of the target wind turbine, the preset time period being a period of preset duration before a current moment;

A prediction module, the prediction module is used to input the historical data of the target wind turbine into a trained neural network, and obtain a fault prediction result of the target wind turbine output by the neural network;

Wherein, the neural network is trained based on a training data set, and the training data set includes sample historical data and sample expansion data, the sample historical data is historical data of wind turbines actually collected, and the sample expansion data is data generated by performing sample expansion processing on the sample historical data.
A terminal, characterized in that the terminal includes: a processor, a computer-readable storage medium communicatively connected to the processor, the computer-readable storage medium is suitable for storing multiple instructions, and the processor is suitable for calling the instructions in the computer-readable storage medium to execute the steps of the wind turbine fault prediction method based on multi-source heterogeneous data as described in any one of claims 1 to 7.
A computer-readable storage medium, characterized in that the computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the wind turbine fault prediction method based on multi-source heterogeneous data as described in any one of claims 1-7.