Method and system for generating operation pressure of gear box of doubly-fed wind turbine generator

Abstract

The invention relates to a doubly-fed wind turbine generator gearbox operation pressure generation method and system. The method comprises the following steps: acquiring real-time operation pressure sequence data of a doubly-fed wind turbine generator gearbox; feature selection and downsampling are carried out based on the real-time operation pressure sequence data, multi-granularity pressure sequence data are obtained, and the multi-granularity pressure sequence data comprise fine-granularity pressure data and coarse-granularity pressure data; and inputting the multi-granularity pressure sequence data into a pre-trained pressure data generation model for processing so as to analyze the trend of the pressure sequence data with different granularities, and finally generating running pressure data of multiple step lengths in the future. Compared with the prior art, the method has the advantages of effectively reducing noise interference, being high in precision and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of time series data generation, and in particular to a method and system for generating operating pressure of a gearbox of a doubly-fed wind turbine generator set. Background Art

[0002] Modern energy has clear plans to vigorously develop wind power energy. At the same time, the power industry continues to promote the intelligent transformation and digital construction of the wind power energy industry through new-generation information technology, and builds a wind turbine data generation model that is fully mapped in the digital virtual space and has risk deduction capabilities. This makes risk assessment, status detection, fault diagnosis and other businesses more efficient and accurate, and accelerates the digital upgrade of the wind power industry.

[0003] At present, wind turbine technology mainly includes the following types: doubly-fed wind turbines, direct-drive permanent magnet synchronous wind turbines, semi-direct-drive synchronous wind turbines, and centralized transmission wind turbines. Among them, doubly-fed wind turbines are widely used in the wind power industry, mainly due to their adjustable performance, good grid friendliness, and mature and reliable technical characteristics. However, in the transmission chain of doubly-fed wind turbines, gearboxes, as key components, have high manufacturing costs and are prone to failure, resulting in high operation and maintenance costs. Among them, the core features such as the operating pressure and operating temperature of the gearbox are the most important. These core features play a decisive role in the operating stability of the entire wind turbine. However, the current research on the operating pressure of the gearbox of a doubly-fed wind turbine is still in the traditional dynamic model. The established pressure model is usually a simplified model, resulting in many features in the pressure data not being fully extracted, hindering a deeper study of the operating pressure of the wind turbine gearbox.

[0004] In recent years, artificial intelligence technology represented by deep learning has developed rapidly. Its powerful high-dimensional complex data feature extraction capability has provided a new breakthrough for breaking the technical bottleneck of wind turbine operation data generation. However, there are relatively few studies on the generation method of the operating pressure of the gearbox of the doubly-fed wind turbine in existing research, which has become a key problem to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for generating operating pressure of a gearbox of a doubly-fed wind turbine generator set, which improves the accuracy of generating pressure data.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for generating operating pressure of a gearbox of a doubly-fed wind turbine generator system comprises the following steps:

[0008] Obtain real-time operating pressure sequence data of the gearbox of the doubly-fed wind turbine;

[0009] Based on the real-time operating pressure sequence data, feature selection and downsampling are performed to obtain multi-granularity pressure sequence data, wherein the multi-granularity pressure sequence data includes fine-grained pressure data and coarse-grained pressure data;

[0010] The multi-granularity pressure sequence data is input into a pre-trained pressure data generation model for processing to analyze the trends of the pressure sequence data with different granularities, and finally generate the operating pressure data of multiple steps in the future.

[0011] Furthermore, the step of obtaining multi-granularity pressure sequence data includes:

[0012] Performing feature selection on the real-time operating pressure sequence data, and using the selected features as fine-grained pressure data;

[0013] The fine-grained pressure data is downsampled in the time dimension to obtain coarse-grained pressure data, wherein the time dimension downsampling adopts an average pooling operation, and the expression of the average pooling operation is:

[0014]

[0015] In the formula, y i is the i-th value after average pooling downsampling, k is the pooling window size, that is, k original values ​​are merged into one downsampled value, x j is the original value at the jth time point in the pressure data sequence.

[0016] Further, the selected features include the average gearbox A1 port pressure, the average gearbox A2 port pressure, the average gearbox A3 port pressure, the average gearbox A4 port pressure, the average gearbox main pump 1_1 outlet pressure, the average gearbox main pump 1_2 outlet pressure, the average gearbox main pump 1_3 outlet pressure, the average gearbox main pump 1_4 outlet pressure, the average gearbox cooling pump outlet pressure, the average gearbox bypass pump outlet pressure, the average gearbox water pump 1 outlet pressure, the average gearbox water pump 1 inlet pressure, the average gearbox water pump 2 outlet pressure and the average gearbox water pump 2 inlet pressure.

[0017] Furthermore, the pressure data generation model includes a trend-residual decomposition module, a multi-granularity trend-residual mixing module, a feature independence module and a Former generator module.

[0018] The trend-residual decomposition module is used to decompose the multi-granularity pressure series data to obtain trend terms and residual terms at multiple granularities;

[0019] The multi-granularity trend-residual mixing module is used to mix the trend term and the residual term under the multi-granularity to obtain multi-granularity trend-residual mixed pressure data;

[0020] The feature independence module is used to perform feature cutting independence on the multi-granularity trend-residual mixed pressure data to obtain pressure independent feature sets at different granularities;

[0021] The Former generator module is used to generate operating pressure data for multiple future steps based on the pressure independent feature sets at different granularities.

[0022] Further, the trend-residual decomposition module includes a first part for obtaining trend terms at multiple granularities and a second part for obtaining residual terms at multiple granularities.

[0023] The first part includes multiple channels, a fusion layer, a channel conversion layer, and a pooling convolution layer. The execution steps of the first part include:

[0024] Input the multi-granularity pressure sequence data into the first channel and the last channel, use the Repeat method to perform multiple replication and filling, and combine the multi-granularity pressure sequence data with the input into the fusion layer for fusion to obtain fused data;

[0025] The fused data is input into a channel conversion layer for multiple channel conversions, and then input into a pooling convolution layer for pooling convolution to obtain trend items under multiple granularities, wherein the trend items under multiple granularities include fine-grained trend items and coarse-grained trend items;

[0026] The execution steps of the second part include:

[0027] The difference between the multi-granularity pressure series data and the trend item under multi-granularity is calculated, and the difference is used as the residual item under multi-granularity, wherein the residual item under multi-granularity includes a fine-granularity residual item and a coarse-granularity residual item.

[0028] Furthermore, the execution steps of the multi-granularity trend-residual hybrid module include:

[0029] Mapping the last dimension of the residual term under the multiple granularities to a fixed dimension using a linear mapping;

[0030] The trend item under the multi-granularity is mixed with the residual item under the multi-granularity after linear mapping to obtain multi-granularity trend-residual mixed pressure data, wherein in the mixing step, the fine-grained trend item is mixed with the coarse-grained residual item after linear mapping, and the coarse-grained trend item is mixed with the fine-grained residual item after linear mapping, and the expression of the mixing operation is:

[0031]

[0032] In the formula, is a fine-grained mixing term, is a fine-grained trend term, is the coarse-grained residual term. represents a coarse-grained mixing term, is the coarse-grained trend term, Represents the fine-grained residual term.

[0033] Furthermore, the execution steps of the feature independent module include:

[0034] Based on the multi-granularity trend-residual mixed pressure data, the Reshape function is used to perform channel reconstruction and the Dropout function is used to perform regularization in turn, so as to perform feature cutting independence and obtain pressure independent feature sets at different granularities.

[0035] Furthermore, the Former generator module includes a generator part and a feature fusion part, the generator part adopts a Transformer model, and the execution steps of the Former generator module include:

[0036] The pressure-independent feature sets at different granularities are input into the Transformer model for processing, so as to generate initial operating pressure data for multiple future steps for each feature in each pressure-independent feature set, and then the initial operating pressure data for multiple future steps of the same feature in each pressure-independent feature set are fused according to the feature fusion part to obtain the final operating pressure data for multiple future steps.

[0037] Furthermore, during the training process of the pressure data generation model, the mean square error loss function is used to solve the loss, and the loss is back-propagated to iteratively optimize the pressure data generation model until the mean square error loss function converges to obtain a trained pressure data generation model, wherein the expression of the mean square error loss function is:

[0038]

[0039] In the formula, MSE Loss is the mean square error loss, y i represents the true value of the i-th sample, represents the generated value of the i-th sample, and N represents the number of samples.

[0040] The present invention also provides a doubly-fed wind turbine gearbox operating pressure generating system, comprising:

[0041] Single-granularity sequence acquisition module: used to obtain real-time operating pressure sequence data of the gearbox of the doubly-fed wind turbine;

[0042] A multi-granularity sequence acquisition module is used to perform feature selection and downsampling based on the real-time operating pressure sequence data to obtain multi-granularity pressure sequence data, wherein the multi-granularity pressure sequence data includes fine-granularity pressure data and coarse-granularity pressure data;

[0043] Generation module: used to input the multi-granularity pressure sequence data into a pre-trained pressure data generation model for processing, and generate operating pressure data for multiple steps in the future.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] (1) In view of the problem that the existing data acquisition has a relatively single granularity and lacks insights into the changing trends at longer or shorter time scales, resulting in the lack of comprehensiveness and accuracy in the generated results, the present invention processes the single-granularity sequence data into multi-granularity sequence data to provide information at different time scales, and discovers its hidden patterns and trends through the pressure data generation model, thereby obtaining deeper analytical insights and improving the accuracy of pressure data generation.

[0046] (2) The present invention considers the correlation between the operating pressure data of the gearbox of the doubly-fed wind turbine and the multidimensionality of the operating pressure data. According to the information entropy contained in the characteristic variables in the operating pressure data, the characteristic variables with a strong correlation with the operating pressure of the gearbox are selected, which can effectively reduce noise interference and improve the accuracy and robustness of the pressure data generation model. According to the multidimensionality of the operating pressure, 14 pressure characteristic variables are selected, which cover the core area of ​​the gearbox, and the generated gearbox pressure data has a high accuracy rate.

[0047] (3) The multi-granularity trend-residual mixing module of the present invention mixes the fine-grained trend term and the coarse-grained residual term, and mixes the coarse-grained trend term and the fine-grained residual term to obtain multi-granularity mixed pressure data. This multi-granularity mixing method is adopted: the coarse-grained residual term makes up for the fine-grained trend term, and the fine-grained residual term fills the coarse-grained trend term, which can effectively reduce the interference of noise and improve the stability and generalization of the generated data.

[0048] (4) The feature independence module of the present invention separates the 14 features in the trend-residual mixed pressure data at two granularities. When the trend-residual mixed pressure data is generated, the features affect each other, and multiple features may bring negative noise information. Generating each feature independently can effectively reduce the interference between features and improve the accuracy of the generated data.

[0049] (5) The operating pressure of the doubly-fed wind turbine gearbox generated by the present invention is highly accurate and can provide powerful data and technical support for the subsequent operation and maintenance of the wind turbine. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0051] Figure 2 A schematic diagram of the pressure data generation model structure of the present invention;

[0052] Figure 3 It is a schematic diagram of the structure of the trend-residual decomposition module of the present invention;

[0053] Figure 4 It is a schematic diagram of the structure of the multi-granularity trend-residual hybrid module and the feature independent module of the present invention. DETAILED DESCRIPTION

[0054] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0055] Example 1

[0056] This embodiment provides a method for generating the operating pressure of the gearbox of a doubly-fed wind turbine. The method first obtains the original pressure data of the gearbox of the doubly-fed wind turbine, performs feature selection on the characteristic variables in the original pressure data, and obtains 14 features with a strong correlation with the operating pressure of the gearbox. The selected pressure data is called fine-grained pressure data; the fine-grained pressure data is downsampled in the time dimension to obtain a coarse-grained pressure data; the pressure data of the two granularities are input into the pressure data generation model to generate pressure data for the next 12 steps. Figure 2 As shown, the pressure data generation model includes a trend-residual decomposition module, a multi-granularity trend-residual mixing module, a feature independent module and a Former generator module. The pressure data generation model first decomposes the trend term and the residual term of the multi-granularity data according to the included modules and then mixes them with each other to obtain multi-granularity trend-residual mixed data, and then cuts the 14 features in the multi-granularity trend-residual mixed pressure data independently, reducing the noise influence between features and improving the operating pressure generation accuracy of the double-fed wind turbine gearbox. Specifically, as Figure 1 As shown, the method comprises the following steps:

[0057] S1. Convert single-granularity sequence data into multi-granularity sequence data.

[0058] In reality, wind turbine gearbox operation data is often collected at fixed intervals, which results in a relatively single granularity of the data obtained. This data reflects information on a specific time scale and lacks insight into the changing trends on longer or shorter time scales. This may lead to a lack of comprehensiveness and accuracy in the generated results.

[0059] Converting this single-granularity sequence data into multi-granularity sequence data by downsampling in the time dimension can provide information at different time scales, help discover hidden patterns and trends, and thus obtain deeper analytical insights; at the same time, it can be flexibly selected and aggregated according to specific needs to better adapt to the analysis needs in different situations.

[0060] The above-mentioned conversion of the single-granularity operating pressure data of the gearbox of the doubly-fed wind turbine generator set into the multi-granularity operating pressure data includes two steps:

[0061] S11. Feature selection of raw pressure data.

[0062] 14 features that are strongly correlated with the gearbox operating pressure are selected: average gearbox A1 port pressure, average gearbox A2 port pressure, average gearbox A3 port pressure, average gearbox A4 port pressure, average gearbox main pump 1_1 outlet pressure, average gearbox main pump 1_2 outlet pressure, average gearbox main pump 1_3 outlet pressure, average gearbox main pump 1_4 outlet pressure, average gearbox cooling pump outlet pressure, average gearbox bypass pump outlet pressure, average gearbox water pump 1 outlet pressure, average gearbox water pump 1 inlet pressure, average gearbox water pump 2 outlet pressure, average gearbox water pump 2 inlet pressure.

[0063] S12. Downsampling the data after feature selection in the time dimension to obtain multi-granularity pressure data.

[0064] The time dimension downsampling adopts average pooling, which is expressed as

[0065]

[0066] x j Represents the original value of the jth time point in the pressure data sequence, y i represents the i-th value after average pooling downsampling, and k represents the pooling window size, that is, k original values ​​are merged into one downsampled value.

[0067] Downsampling the time dimension of fine-grained pressure data based on average pooling can effectively reduce the noise and outliers in the fine-grained pressure data while retaining the overall trends and characteristics of the fine-grained pressure data.

[0068] S2. Decomposition and mixing of multi-granularity sequence data based on trend-residual decomposition module and multi-granularity trend-residual mixing module.

[0069] In this embodiment, based on the above conversion of single-granularity sequence data into multi-granularity sequence data, the multi-granularity data is decomposed and mixed by a trend-residual decomposition module and a multi-granularity trend-residual mixing module. This step is divided into two stages, as follows:

[0070] The first stage is to use the trend-residual decomposition module to perform trend-residual decomposition, such as Figure 3 This stage consists of two parts. The first part is to obtain the trend item of the input pressure data, including copying and filling, merging and splicing, pooling convolution and channel reconstruction operations on the input data. The second part is to obtain the difference between the input pressure data and the trend item of the pressure data, and use the difference as the residual item of the pressure data.

[0071] The pressure data input in the first part has three channels. The first channel, channel 0, and the last channel, channel 2, are copied and filled multiple times, specifically using the Repeat method; then the input pressure data and the two channel data after copying and filling are fused; finally, the fused data is subjected to two channel conversions and one pooling convolution to obtain the input pressure data trend item.

[0072] The replication, padding and pooling convolutions not only expand the number of channels of the input pressure data, but also extract the trend features of the input pressure data and improve the accuracy of the trend items.

[0073] The second stage is to use a multi-granularity trend-residual mixing module to perform multi-granularity trend-residual mixing, such as Figure 4 As shown. This module consists of two parts: linear mapping and trend-residual mixing. The linear mapping part contains two linear layers and a Gelu activation function. The linear mapping part maps the last dimension of the residual term to a fixed dimension to facilitate subsequent splicing operations, where:

[0074] The first linear layer has 13 input channels and 25 output channels.

[0075] Gelu activation function: Use GELU library function instead;

[0076] First Linear layer: input channels are 25, output channels are 25.

[0077] The trend-residual mixing part mixes trend items and residual items of different granularities. Specifically, it mixes fine-grained trend items with coarse-grained residual items, and coarse-grained trend items with fine-grained residual items to obtain multi-granularity trend-residual mixed pressure data. The expressions of the mixing operation include:

[0078]

[0079] in is a fine-grained mixing term, is a fine-grained trend term, is the coarse-grained residual term. represents a coarse-grained mixing term, is the coarse-grained trend term, Represents the fine-grained residual term.

[0080] The multi-granularity trend-residual mixing method can effectively reduce the interference of noise and improve the accuracy of generated data.

[0081] S3. Feature independence of multi-granularity sequence data based on feature independence module.

[0082] In this embodiment, after the above multi-granularity trend-residual mixing, the mixed data features need to be cut independently to obtain pressure feature sets of two granularities. This step includes two parts: channel reconstruction and regularization. The channel reconstruction part first obtains the channel size of the input data, and then reconstructs the channel of the data through the Reshape function; the regularization part is implemented by the Dropout function.

[0083] Specifically, the 14 features in the trend-residual mixed pressure data at two granularities are cut and separated. When the trend-residual mixed pressure data is generated, the features affect each other, and multiple features may bring negative noise information. Generating each feature independently can effectively reduce the interference between features and improve the accuracy of the generated data.

[0084] S4. Pressure data generation model training.

[0085] Obtain the operating pressure training data of the gearbox of the doubly-fed wind turbine and perform feature selection;

[0086] Divide the training data after feature selection into batches, and input each batch of training data into the pressure data generation model in turn to generate the future 12-step pressure data of this batch of training data;

[0087] The mean square error (MSE) loss function is constructed based on the labeled data and generated data of the same batch to solve the loss, and the loss is back-propagated to iteratively optimize the pressure data generation model until the loss function converges, completing the training of the doubly-fed wind turbine gearbox operating pressure data generation model.

[0088] The expression of the mean square error is as follows:

[0089]

[0090] Where N represents the number of samples, y i represents the true value of the i-th sample, Represents the generated value of the i-th sample. The calculation process is as follows: for each sample, calculate the difference between the generated value and the true value, square each difference, sum all the squared values, and finally divide by the number of samples to get the average mean square error.

[0091] MSE as a loss function can effectively evaluate the performance of the pressure data generation model and help optimize the generation ability of the pressure data generation model.

[0092] S5. Pressure data generation based on the Former generator module.

[0093] In this example, the Former generator module is finally used to generate pressure data for feature sets at multiple granularities. This module consists of two parts: the generator and the feature fusion. The generator uses the Transformer model; the feature fusion part fuses 14 features at different granularities.

[0094] This embodiment uses the SCADA data collected from a wind farm to conduct experimental analysis based on the above method to prove the effectiveness of the above method. Specifically:

[0095] 1) Feature selection and data processing

[0096] This embodiment obtains the operating data of 7 wind turbines from the wind farm during the period from July 15, 2023 to July 31, 2023.

[0097] Feature selection: Select the characteristic variables with strong correlation with the operating pressure of the wind turbine gearbox, including the average gearbox A1 port pressure, the average gearbox A2 port pressure, the average gearbox A3 port pressure, the average gearbox A4 port pressure, the average gearbox main pump 1_1 outlet pressure, the average gearbox main pump 1_2 outlet pressure, the average gearbox main pump 1_3 outlet pressure, the average gearbox main pump 1_4 outlet pressure, the average gearbox cooling pump outlet pressure, the average gearbox bypass pump outlet pressure, the average gearbox water pump 1 outlet pressure, the average gearbox water pump 1 inlet pressure, the average gearbox water pump 2 outlet pressure and the average gearbox water pump 2 inlet pressure. The selected parameter results are shown in Table 1.

[0098] Table 1. Results of feature selection of operating pressure data of gearbox of doubly-fed wind turbine

[0099]

[0100] Data processing: Based on feature selection, the data needs to be merged and divided. First, the data of the seven wind turbines are merged, and then the data is divided according to the training set accounting for 70% and the test set accounting for 20%. After the division, the training set contains 3225 data samples and the test set contains 1815 data samples.

[0101] 2) Indicator settings

[0102] In the indicator setting, the mean squared error (MSE) is selected as the evaluation indicator. MSE is used to measure the gap between the generated value and the true value. It is a non-negative value. The smaller the value, the more accurate the generated data. The expression of the mean squared error is as follows:

[0103]

[0104] Where N represents the number of samples, y i represents the true value of the i-th sample, Represents the generated value of the i-th sample. The calculation process is as follows: for each sample, calculate the difference between the generated value and the true value, square each difference, sum all the squared values, and finally divide by the number of samples to get the mean square error.

[0105] 3) Model validation

[0106] The processed wind turbine gearbox pressure data was subjected to different comparative example experiments, and the experimental results are shown in Table 2:

[0107] Table 2 Experimental results of different comparative examples of the doubly-fed wind turbine gearbox operating pressure data generation model

[0108]

[0109] Among them, M1 represents the pressure data generation model that directly uses the Transformer generator to generate the gearbox pressure data; M2 represents the pressure data generation model that generates the gearbox pressure data by using the Transformer generator after making the features of the gearbox pressure data independent; M3 represents the pressure data generation model that performs trend-residual decomposition on the gearbox pressure data and makes it feature independent before generating it by Transformer; M4 represents the pressure data generation model that decomposes the multi-granularity sequence data and then performs multi-granularity trend-residual mixing, then makes it feature independent, and finally generates the data by Transformer.

[0110] By comparing the MSE of the four examples, it can be seen that the MSE index of example M4 is lower than that of M1, M2 and M3. The MSE of M4 is 0.834, which is reduced by 0.508, 0.019 and 0.014 respectively compared with M1, M2 and M3. It can be seen that the superiority of the operating pressure generation method of the doubly fed wind turbine gearbox proposed in this paper is superior.

[0111] Example 2

[0112] This embodiment provides a system for generating operating pressure of a gearbox of a doubly-fed wind turbine generator set, the system comprising:

[0113] Single-granularity sequence acquisition module: used to obtain real-time operating pressure sequence data of the gearbox of the doubly-fed wind turbine;

[0114] A multi-granularity sequence acquisition module is used to perform feature selection and downsampling based on the real-time operating pressure sequence data to obtain multi-granularity pressure sequence data, wherein the multi-granularity pressure sequence data includes fine-granularity pressure data and coarse-granularity pressure data;

[0115] Generation module: used to input the multi-granularity pressure sequence data into a pre-trained pressure data generation model for processing, and generate operating pressure data for multiple steps in the future.

[0116] The rest is the same as in Example 1.

[0117] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0118] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0119] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0122] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0123] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine generator set, characterized in that: The following steps are involved: Obtain real-time operating pressure sequence data of the gearbox of the doubly-fed wind turbine; Based on the real-time operating pressure sequence data, feature selection and downsampling are performed to obtain multi-granularity pressure sequence data, wherein the multi-granularity pressure sequence data includes fine-grained pressure data and coarse-grained pressure data; The multi-granularity pressure sequence data is input into a pre-trained pressure data generation model for processing to analyze the trends of the pressure sequence data with different granularities, and finally generate the operating pressure data of multiple steps in the future.

2. A method for generating operating pressure of a doubly-fed wind turbine gearbox according to claim 1, characterized in that: The step of obtaining multi-granularity pressure sequence data comprises: Performing feature selection on the real-time operating pressure sequence data, and using the selected features as fine-grained pressure data; The fine-grained pressure data is downsampled in the time dimension to obtain coarse-grained pressure data, wherein the time dimension downsampling adopts an average pooling operation, and the expression of the average pooling operation is: In the formula, y i is the i-th value after average pooling downsampling, k is the pooling window size, that is, k original values ​​are merged into one downsampled value, x j is the original value at the jth time point in the pressure data sequence.

3. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine according to claim 2, characterized in that: The selected features include the average gearbox A1 port pressure, the average gearbox A2 port pressure, the average gearbox A3 port pressure, the average gearbox A4 port pressure, the average gearbox main pump 1_1 outlet pressure, the average gearbox main pump 1_2 outlet pressure, the average gearbox main pump 1_3 outlet pressure, the average gearbox main pump 1_4 outlet pressure, the average gearbox cooling pump outlet pressure, the average gearbox bypass pump outlet pressure, the average gearbox water pump 1 outlet pressure, the average gearbox water pump 1 inlet pressure, the average gearbox water pump 2 outlet pressure and the average gearbox water pump 2 inlet pressure.

4. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine according to claim 1, characterized in that: The pressure data generation model includes a trend-residual decomposition module, a multi-granularity trend-residual mixing module, a feature independence module and a Former generator module. The trend-residual decomposition module is used to decompose the multi-granularity pressure series data to obtain trend terms and residual terms at multiple granularities; The multi-granularity trend-residual mixing module is used to mix the trend term and the residual term under the multi-granularity to obtain multi-granularity trend-residual mixed pressure data; The feature independence module is used to perform feature cutting independence on the multi-granularity trend-residual mixed pressure data to obtain pressure independent feature sets at different granularities; The Former generator module is used to generate operating pressure data for multiple future steps based on the pressure independent feature sets at different granularities.

5. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine according to claim 4, characterized in that: The trend-residual decomposition module includes a first part for obtaining trend terms at multiple granularities and a second part for obtaining residual terms at multiple granularities. The first part includes multiple channels, a fusion layer, a channel conversion layer, and a pooling convolution layer. The execution steps of the first part include: Input the multi-granularity pressure sequence data into the first channel and the last channel, use the Repeat method to perform multiple replication and filling, and combine the multi-granularity pressure sequence data with the input into the fusion layer for fusion to obtain fused data; The fused data is input into a channel conversion layer for multiple channel conversions, and then input into a pooling convolution layer for pooling convolution to obtain trend items under multiple granularities, wherein the trend items under multiple granularities include fine-grained trend items and coarse-grained trend items; The execution steps of the second part include: The difference between the multi-granularity pressure series data and the trend item under multi-granularity is calculated, and the difference is used as the residual item under multi-granularity, wherein the residual item under multi-granularity includes a fine-granularity residual item and a coarse-granularity residual item.

6. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine according to claim 5, characterized in that: The execution steps of the multi-granularity trend-residual hybrid module include: Mapping the last dimension of the residual term under the multiple granularities to a fixed dimension using a linear mapping; The trend item under the multi-granularity is mixed with the residual item under the multi-granularity after linear mapping to obtain multi-granularity trend-residual mixed pressure data, wherein in the mixing step, the fine-grained trend item is mixed with the coarse-grained residual item after linear mapping, and the coarse-grained trend item is mixed with the fine-grained residual item after linear mapping, and the expression of the mixing operation is: In the formula, is a fine-grained mixing term, is a fine-grained trend term, is the coarse-grained residual term. represents a coarse-grained mixing term, is the coarse-grained trend term, Represents the fine-grained residual term.

7. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine generator set according to claim 4, characterized in that: The execution steps of the feature independent module include: Based on the multi-granularity trend-residual mixed pressure data, the Reshape function is used to perform channel reconstruction and the Dropout function is used to perform regularization in turn, so as to perform feature cutting independence and obtain pressure independent feature sets at different granularities.

8. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine generator set according to claim 4, characterized in that: The Former generator module includes a generator part and a feature fusion part. The generator part adopts the Transformer model. The execution steps of the Former generator module include: The pressure-independent feature sets at different granularities are input into the Transformer model for processing, so as to generate initial operating pressure data for multiple future steps for each feature in each pressure-independent feature set, and then the initial operating pressure data for multiple future steps of the same feature in each pressure-independent feature set are fused according to the feature fusion part to obtain the final operating pressure data for multiple future steps.

9. A method for generating operating pressure of a gearbox of a doubly-fed wind turbine according to claim 1, characterized in that: During the training process of the pressure data generation model, the mean square error loss function is used to solve the loss, and the loss is back-propagated to iteratively optimize the pressure data generation model until the mean square error loss function converges to obtain a trained pressure data generation model, wherein the expression of the mean square error loss function is: In the formula, MSE Loss is the mean square error loss, y i represents the true value of the i-th sample, represents the generated value of the i-th sample, and N represents the number of samples.

10. A doubly-fed wind turbine gearbox operating pressure generation system, characterized in that: include: Single-granularity sequence acquisition module: used to obtain real-time operating pressure sequence data of the gearbox of the doubly-fed wind turbine; A multi-granularity sequence acquisition module is used to perform feature selection and downsampling based on the real-time operating pressure sequence data to obtain multi-granularity pressure sequence data, wherein the multi-granularity pressure sequence data includes fine-granularity pressure data and coarse-granularity pressure data; Generation module: used to input the multi-granularity pressure sequence data into a pre-trained pressure data generation model for processing, and generate operating pressure data for multiple steps in the future.

Images