Wind turbine generator running state identification method and related device

Abstract

The invention discloses a wind turbine generator operation state identification method and a related device, and the method comprises the steps: converting all types of time domain operation data of a wind turbine generator in all time periods into corresponding frequency domain data groups according to the sequence of the time periods; for any type, permutation and combination are carried out on the frequency domain data sets corresponding to the time periods of the type according to the sequence of the time periods, and a corresponding frequency domain data matrix is obtained; and inputting the frequency domain data matrix corresponding to each type into a pre-trained AttachNet neural network to obtain a unit operation state identification result output by the AttachNet neural network. Therefore, the data is transferred from the time domain to the frequency domain, then accurate feature information is mined from the multi-dimensional frequency domain data based on the AttachNet neural network, the corresponding unit operation state is accurately identified, and the identification result is relatively accurate.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method for identifying the operating status of a wind turbine generator set and a related device. Background Art

[0002] In the process of large-scale wind power generation, the superior performance of the units also brings greater safety challenges. Under the dual pressure of the continuous reduction of the market cost of electricity, the safety challenges of large-scale units are becoming more and more severe.

[0003] Under the current technical route, on the design side, the existing simulation calculation work focuses more on the end-to-end load output of the unit, and lacks analysis work and means for the characteristic states and characteristic quantities that cannot be directly reflected in the load. At present, as the large-scale units are becoming more and more popular, the difference between the simulation model and the actual physical state of the unit is becoming more and more obvious. Therefore, how to accurately evaluate the operating status of the unit has become a technical problem that needs to be solved urgently by people in this field. Summary of the invention

[0004] In view of the above problems, the present invention provides a method for identifying the operating status of a wind turbine set and a related device that overcomes the above problems or at least partially solves the above problems.

[0005] In a first aspect, a method for identifying an operating state of a wind turbine generator system includes:

[0006] According to the order of time periods, each type of time domain operation data of the wind turbine in each time period is converted into a corresponding frequency domain data group, wherein one type of time domain operation data involves multiple time periods, and one type of time domain operation data of one time period corresponds to one frequency domain data group, and the types include: generator high-speed shaft speed and nacelle left and right vibration acceleration;

[0007] For any type, the frequency domain data groups corresponding to each time period of the type are arranged and combined in the order of time periods to obtain a corresponding frequency domain data matrix, wherein one frequency domain data group serves as a row vector of the frequency domain data matrix;

[0008] The frequency domain data matrices corresponding to each of the types are input into a pre-trained AttachNet neural network to obtain the unit operation status recognition result output by the AttachNet neural network, wherein the frequency domain data matrices corresponding to different types are input into different additional branches of the AttachNet neural network.

[0009] Optionally, in some optional implementations, converting the time domain operation data of each type of wind turbine generator set in each time period into a corresponding frequency domain data group in the order of the time period includes:

[0010] For any type of time domain operation data, the time domain operation data of different time periods are converted from the time domain to the frequency domain in the order of the time period, and sampled in the form of row vectors to obtain the corresponding frequency domain data group, wherein the frequency domain value range of the frequency domain data group is 0 to 5 Hz, and the resolution of the frequency domain data group is 0.04 Hz.

[0011] Optionally, in some optional implementations, for any type, the frequency domain data groups corresponding to each time period of the type are arranged and combined in the order of time periods to obtain a corresponding frequency domain data matrix, including:

[0012] For any type, the frequency domain data groups corresponding to each time period of the type are arranged from top to bottom in the order of time periods to obtain the corresponding frequency domain data matrix, wherein the frequency domain data group with the earliest time period is used as the first row vector of the frequency domain data matrix, and the frequency domain data group with the latest time period is used as the last row vector of the frequency domain data matrix.

[0013] Optionally, in certain optional implementations, the frequency domain data matrix corresponding to each type is input into a pre-trained AttachNet neural network to obtain a unit operation status recognition result output by the AttachNet neural network, including:

[0014] Inputting the frequency domain data matrix corresponding to each of the types into the additional branch corresponding to the AttachNet neural network to obtain 1024 channel features extracted by each of the additional branches based on the corresponding frequency domain data matrix, wherein the feature extraction networks of different additional branches are identical in structure;

[0015] The attention module of the AttachNet neural network identifies and obtains corresponding feature vectors according to the features of each of the 1024 channels;

[0016] The feature vector is classified based on Softmax to obtain the corresponding unit operation status recognition result, wherein the loss function of the Softmax is a cross entropy loss function.

[0017] Optionally, in certain optional embodiments, the inputting the frequency domain data matrix corresponding to each of the types into the additional branch corresponding to the AttachNet neural network to obtain 1024 channel features extracted by each of the additional branches based on the corresponding frequency domain data matrix includes:

[0018] Inputting the frequency domain data matrix corresponding to each of the types into the additional branch corresponding to the AttachNet neural network;

[0019] The convolutional layers of each additional branch perform convolution, regularization and ReLU activation function operations on their respective frequency domain data matrices to extract corresponding 1024 channel features.

[0020] Optionally, in some optional embodiments, the attention module of the AttachNet neural network identifies and obtains corresponding feature vectors according to each of the 1024 channel features, including:

[0021] The spatial attention module in the attention module connects the 1024 channel features in the order of the corresponding additional branches to obtain a connection feature;

[0022] The fusion module in the attention module performs convolution, regularization, ReLU activation function operation, average pooling operation and maximum pooling operation on the connection features to obtain a corresponding feature vector.

[0023] In a second aspect, a wind turbine operating status identification device includes: a data conversion unit, a matrix combination unit and a status identification unit;

[0024] The data conversion unit is used to convert various types of time domain operation data of the wind turbine in each time period into corresponding frequency domain data groups in the order of time periods, wherein one type of time domain operation data involves multiple time periods, and one type of time domain operation data of one time period corresponds to one frequency domain data group, and the types include: generator high-speed shaft speed and nacelle left and right vibration acceleration;

[0025] The matrix combination unit is used for arranging and combining the frequency domain data groups corresponding to each time period of the type in the order of time periods for any type to obtain a corresponding frequency domain data matrix, wherein one frequency domain data group serves as a row vector of the frequency domain data matrix;

[0026] The state recognition unit is used to input the frequency domain data matrices corresponding to each of the types into a pre-trained AttachNet neural network to obtain the unit operation state recognition result output by the AttachNet neural network, wherein the frequency domain data matrices corresponding to different types are input into different additional branches of the AttachNet neural network.

[0027] Optionally, in certain optional implementations, the data conversion unit includes: a data conversion module;

[0028] The data conversion module is used to convert the time domain operation data of different time periods from the time domain to the frequency domain in the order of time periods for any type of time domain operation data, and sample the corresponding frequency domain data group in the form of row vectors, wherein the frequency domain value range of the frequency domain data group is 0 to 5 Hz, and the resolution of the frequency domain data group is 0.04 Hz.

[0029] In a third aspect, a computer-readable storage medium stores a program, and when the program is executed by a processor, the method for identifying the operating status of a wind turbine generator set described in any one of the above items is implemented.

[0030] In a fourth aspect, an electronic device is provided, comprising at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other via the bus; and the processor is used to call program instructions in the memory to execute any of the above-mentioned methods for identifying the operating status of a wind turbine set.

[0031] By means of the above technical scheme, the present invention provides a method and related device for identifying the operating state of a wind turbine set, which can convert the time domain operating data of each type of wind turbine set in each time period into a corresponding frequency domain data group in the order of time periods, wherein one type of time domain operating data involves multiple time periods, and one type of time domain operating data of one time period corresponds to one frequency domain data group, and the types include: generator high-speed shaft speed and nacelle left and right vibration acceleration; for any type, according to the order of time periods, the frequency domain data groups corresponding to each time period of the type are arranged and combined to obtain a corresponding frequency domain data matrix, wherein a frequency domain data group is used as a row vector of the frequency domain data matrix; the frequency domain data matrices corresponding to each type are input into the pre-trained AttachNet neural network to obtain the unit operating state identification result output by the AttachNet neural network, wherein the frequency domain data matrices corresponding to different types are input into different additional branches of the AttachNet neural network. It can be seen from this that the data is converted from the time domain to the frequency domain, and then accurate feature information is mined from the multi-dimensional frequency domain data based on the AttachNet neural network, and the corresponding unit operating state is accurately identified, and the identification result is relatively accurate.

[0032] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0034] Figure 1 A flow chart of a method for identifying the operating status of a wind turbine set provided by the present invention is shown;

[0035] Figure 2 A schematic structural diagram of a wind turbine operating status identification device provided by the present invention is shown;

[0036] Figure 3 A structural schematic diagram of an electronic device provided by the present invention is shown. DETAILED DESCRIPTION

[0037] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0038] like Figure 1 As shown, the present invention provides a method for identifying the operating status of a wind turbine generator set, comprising: S100, S200 and S300;

[0039] S100, converting each type of time domain operation data of the wind turbine generator set in each time period into a corresponding frequency domain data group in the order of time periods, wherein one type of time domain operation data involves multiple time periods, and one type of time domain operation data of one time period corresponds to one frequency domain data group, and the types include: generator high-speed shaft speed and nacelle left and right vibration acceleration;

[0040] For example, in some optional embodiments, the S100 includes:

[0041] For any type of time domain operation data, the time domain operation data of different time periods are converted from the time domain to the frequency domain in the order of the time period, and sampled in the form of row vectors to obtain the corresponding frequency domain data group, wherein the frequency domain value range of the frequency domain data group is 0 to 5 Hz, and the resolution of the frequency domain data group is 0.04 Hz.

[0042] Optionally, during the operation of the wind turbine, corresponding time domain operation data will be continuously generated. The present invention can divide the time domain operation data into data of multiple time periods according to a certain time step. For example, there is a period of time domain operation data from August 1 to August 2. The present invention can divide the period of time domain operation data by hour to obtain the time domain operation data from 00:00 to 01:00 on August 1, the time domain operation data from 01:00 to 02:00 on August 1, and the time domain operation data from 02:00 to 03:00 on August 1, and so on. At this time, 1 hour is taken as a time period, and the present invention does not limit this.

[0043] Optionally, the generator high-speed shaft speed and the left-right vibration acceleration of the cabin mentioned in the present invention belong to the well-known technical concepts in the art, and the present invention does not make too much description on this. It should be noted that the time domain operation data mentioned in the present invention is not limited to the generator high-speed shaft speed and the left-right vibration acceleration of the cabin, and can also cover other data according to actual needs. For these data, the present invention can collect them through sensors when the wind turbine is running, and the present invention does not limit this.

[0044] Optionally, the present invention can record the current frequency domain results (frequency domain range is 0-5Hz, resolution is 0.04Hz) in the form of row vectors at fixed time intervals, and record 125 consecutive results as a group. Taking the generator high-speed shaft speed as an example, the frequency domain data group corresponding to the generator high-speed shaft speed in one time period can be: the time domain signal of the generator high-speed shaft speed is converted into the frequency domain, and the frequency amplitude corresponding to the frequency range of 0-5Hz is obtained after conversion, and then the frequency domain amplitude corresponding to each resolution point is recorded with a resolution of 0.04Hz, that is, the amplitudes corresponding to the frequency points of 0.04Hz, 0.08Hz, 0.12Hz...5Hz are recorded one by one in the form of row vectors, and the present invention does not limit this.

[0045] Optionally, if it is in the model training stage, the present invention can also process the frequency domain data group (which can also be understood as: data set), screen out high-noise data sets and high-redundancy data sets, and divide them into: normal data set group and abnormal data set group (such as structural abnormality data, control abnormality data, environmental abnormality data, etc.) according to whether the unit has any operating abnormality, a total of 2000 groups, and finally divide the 2000 groups of data sets into training sets and verification sets in a ratio of 8:2, which are used for model training and model verification respectively. The present invention does not impose any restrictions on this.

[0046] S200, for any type, arranging and combining the frequency domain data groups corresponding to each time period of the type in the order of time periods to obtain a corresponding frequency domain data matrix, wherein one frequency domain data group serves as a row vector of the frequency domain data matrix;

[0047] For example, in some optional implementations, the S200 includes:

[0048] For any type, the frequency domain data groups corresponding to each time period of the type are arranged from top to bottom in the order of time periods to obtain the corresponding frequency domain data matrix, wherein the frequency domain data group with the earliest time period is used as the first row vector of the frequency domain data matrix, and the frequency domain data group with the latest time period is used as the last row vector of the frequency domain data matrix.

[0049] Optionally, as mentioned above, the time domain data of the present invention can be of multiple types, so each type of time domain data has corresponding frequency domain data after being converted into the frequency domain. The present invention can perform the process of S200 according to different types. That is, for each type of data, the corresponding frequency domain data group is arranged from top to bottom to obtain a corresponding frequency domain data matrix.

[0050] It should be noted that: as mentioned above, each frequency domain data group includes multiple amplitudes corresponding to different frequencies. That is, each frequency domain data group can be understood as a row vector, the elements of which are composed of multiple amplitudes of different frequencies arranged in sequence according to the frequency order, and the present invention does not limit this.

[0051] S300, inputting the frequency domain data matrices corresponding to each of the types into a pre-trained AttachNet neural network to obtain a unit operation status recognition result output by the AttachNet neural network, wherein the frequency domain data matrices corresponding to different types are input into different additional branches of the AttachNet neural network.

[0052] Optionally, the AttachNet neural network described in the present invention can be pre-trained. As mentioned above, the present invention can divide the training set and the verification set, and then train according to the training method of the AttachNet neural network. It should be noted that: the present invention involves multiple additional branches. For the different types of frequency domain data corresponding to the additional branches from left to right in the training stage of the AttachNet neural network, the present invention does not make specific requirements on the order of the additional branches. It only requires that after the neural network is trained, the order of input data from the additional branches from left to right should remain the same as that during training. For example, during training, the additional branches from left to right are trained in the order of rotation speed, vibration acceleration and load, and the data should also be input into the AttachNet neural network in this order when applied.

[0053] For example, in some optional embodiments, the S300 includes: step 1.1, step 1.2 and step 1.3;

[0054] Step 1.1, inputting the frequency domain data matrix corresponding to each type into the additional branch corresponding to the AttachNet neural network to obtain 1024 channel features extracted by each additional branch based on the corresponding frequency domain data matrix, wherein the feature extraction networks of different additional branches are the same in structure;

[0055] For example, in some optional embodiments, the step 1.1 includes: step 2.1 and step 2.2;

[0056] Step 2.1, input the frequency domain data matrix corresponding to each type into the additional branch corresponding to the AttachNet neural network;

[0057] Step 2.2: The convolutional layers of each of the additional branches perform convolution, regularization, and ReLU activation function operations on their respective frequency domain data matrices to extract corresponding 1024 channel features.

[0058] Optionally, the 1024-channel feature is a well-known technical concept in the art, and the present invention will not describe it in detail. Please refer to the relevant explanations in the art for details.

[0059] Step 1.2, the attention module of the AttachNet neural network identifies and obtains corresponding feature vectors according to the features of each of the 1024 channels;

[0060] Optionally, in some optional embodiments, the step 1.2 includes: step 3.1 and step 3.2;

[0061] Step 3.1, the spatial attention module in the attention module connects the 1024 channel features in the order of the corresponding additional branches to obtain the connection features;

[0062] Step 3.2: The fusion module in the attention module performs convolution, regularization, ReLU activation function operation, average pooling operation and maximum pooling operation on the connection features to obtain the corresponding feature vector.

[0063] Step 1.3: classify the feature vector based on Softmax to obtain the corresponding unit operation status recognition result, wherein the loss function of the Softmax is a cross entropy loss function.

[0064] Optionally, the AttachNet neural network of the present invention can be divided into two parts: feature extraction and feature fusion of input data. For the feature extraction part, the present invention can be input as a 125×125 frequency domain data matrix, whose row vector is the frequency domain information of the unit, and is constructed into 125 rows in chronological order. Different types of input data will be sent to different additional branches. The feature extraction networks in each additional branch are exactly the same in structure. In order to fully retain the frequency domain information in the shallow network, the feature extraction module is a residual structure. ResNet-50 is used as the backbone network as a whole, and feature extraction is realized through multiple convolution blocks. Convolution, regularization, and ReLU activation function operations are completed in sequence in the convolution layer in each convolution block. Finally, the feature extraction network of each branch extracts the corresponding 1024 channel features of its input data.

[0065] In order to capture the intra-layer temporal information contained in the feature map, the feature information extracted by each additional branch will be sent to its corresponding pixel-level attention module (including spatial attention module and fusion module), whose structure is as follows:

[0066] F output =F input ×Att p

[0067] Att p =sigmoid(cat(Avgpool(T input ), Maxpool(F input )))

[0068] Among them, Att p is the spatial attention module, cat is the connection operation, and the number of channels in the spatial attention module is 2, which is the same as the input feature size.

[0069] Specifically, the features of each branch are connected in sequence (a fixed operation of the neural network, specifically stacking the data of different channels in a certain order. For example, connecting two four-channel data can get one eight-channel data. The order here refers to the arrangement of different types of input data in the above text. In the connection process, the order of different input data in the stacking process will affect the result.) and sent to the fusion module, which contains two convolution blocks and a channel attention module, and its structure is as follows:

[0070] Convolutional Block 1:

[0071]

[0072] Among them, Conv n×n , c is the n×n×c convolution operation, Bn is the regularization, and ReLU is the activation function.

[0073] Convolutional Block 2:

[0074]

[0075] Channel Attention Module:

[0076] F output =F input ×Att c

[0077] Att c =sigmoid(MLP(Avgpool(F input ))+MLP(Maxpool(F input )))

[0078] MLP=f(W T x+b)

[0079] Among them, F input 、F output is the input and output features, Att c is channel attention, MLP is a multi-layer neural network with x as input, W and b represent the weight and bias to be learned respectively, Avgpool and Maxpoo1 are average pooling and maximum pooling operations respectively.

[0080] Finally, based on the features extracted by the network, the judgment results are classified through Softmax, and the loss function uses the cross entropy loss function loss bce , the formula is:

[0081]

[0082] Among them, G(x) is the ground truth corresponding to the input, T(x) is the label value calculated by the network, and the neural network parameters are iteratively updated by back propagation through the calculation of the loss function.

[0083] The AttachNet neural network is trained end-to-end. m Optimizer, learning rate is set to 0.001, beta = (0.9, 0.999), eps = 1e-8, weight_decay = 0. Input Batch size is set to 8.

[0084] Optionally, the present invention can also improve the accuracy of the model by online learning. Online learning is similar to the above steps, except that online learning needs to obtain the actual operation data of the site and then fine-tune the model to capture more accurate feature information.

[0085] In summary, the AttachNet neural network designed in the present invention analyzes the operating status of the unit by capturing the changing patterns of the frequency domain information corresponding to the signals of each dimension under different operating conditions over time, and by introducing attention modules of different dimensions, the AttachNet neural network can accurately extract key dynamic features. As the data scale continues to expand, the AttachNet neural network can learn feature information based on big data, and has stronger recognition ability and higher recognition accuracy.

[0086] like Figure 2 As shown, the present invention provides a wind turbine operating status identification device, comprising: a data conversion unit 100, a matrix combination unit 200 and a status identification unit 300;

[0087] The data conversion unit 100 is used to convert various types of time domain operation data of the wind turbine in each time period into corresponding frequency domain data groups in the order of time periods, wherein one type of time domain operation data involves multiple time periods, and one type of time domain operation data of one time period corresponds to one frequency domain data group, and the types include: generator high-speed shaft speed and nacelle left and right vibration acceleration;

[0088] The matrix combination unit 200 is used for arranging and combining the frequency domain data groups corresponding to each time period of the type in the order of time periods for any type to obtain a corresponding frequency domain data matrix, wherein one frequency domain data group serves as a row vector of the frequency domain data matrix;

[0089] The state recognition unit 300 is used to input the frequency domain data matrices corresponding to each type into a pre-trained AttachNet neural network to obtain the unit operation state recognition result output by the AttachNet neural network, wherein the frequency domain data matrices corresponding to different types are input into different additional branches of the AttachNet neural network.

[0090] Optionally, in some optional implementations, the data conversion unit 100 includes: a data conversion module;

[0091] The data conversion module is used to convert the time domain operation data of different time periods from the time domain to the frequency domain in the order of time periods for any type of time domain operation data, and sample the corresponding frequency domain data group in the form of row vectors, wherein the frequency domain value range of the frequency domain data group is 0 to 5 Hz, and the resolution of the frequency domain data group is 0.04 Hz.

[0092] Optionally, in some optional implementations, the matrix combination unit 200 includes: a matrix combination module;

[0093] The matrix combination module is used to arrange the frequency domain data groups corresponding to each time period of the type from top to bottom in the order of time periods for any type, so as to obtain the corresponding frequency domain data matrix, wherein the frequency domain data group with the earliest time period is used as the first row vector of the frequency domain data matrix, and the frequency domain data group with the latest time period is used as the last row vector of the frequency domain data matrix.

[0094] Optionally, in some optional implementations, the state recognition unit 300 includes: a matrix input module, an attention module and a state recognition module;

[0095] The matrix input module is used to input the frequency domain data matrix corresponding to each type into the additional branch corresponding to the AttachNet neural network to obtain 1024 channel features extracted by each additional branch based on the corresponding frequency domain data matrix, wherein the feature extraction networks of different additional branches are the same in structure;

[0096] The attention module is used to identify and obtain corresponding feature vectors according to the 1024 channel features;

[0097] The state recognition module is used to classify the feature vector based on Softmax to obtain the corresponding unit operation state recognition result, wherein the loss function of the Softmax is a cross entropy loss function.

[0098] Optionally, in some optional implementations, the matrix input module includes: a data input module and a convolutional layer module;

[0099] The data input module is used to input the frequency domain data matrix corresponding to each type into the additional branch corresponding to the AttachNet neural network;

[0100] The convolution layer module is used to control the convolution layers of each of the additional branches to perform convolution, regularization and ReLU activation function operations on the respective frequency domain data matrices to extract corresponding 1024 channel features.

[0101] Optionally, in certain optional embodiments, the attention module comprises: a spatial attention module and a fusion module;

[0102] The spatial attention module is used to connect the 1024 channel features in the order of the corresponding additional branches to obtain a connection feature;

[0103] The fusion module is used to perform convolution, regularization, ReLU activation function operation, average pooling operation and maximum pooling operation on the connection features to obtain corresponding feature vectors.

[0104] The present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the method for identifying the operating status of a wind turbine set described in any one of the above items is implemented.

[0105] like Figure 3 As shown, the present invention provides an electronic device 70, which includes at least one processor 701, and at least one memory 702 and a bus 703 connected to the processor 701; wherein the processor 701 and the memory 702 communicate with each other through the bus 703; the processor 701 is used to call the program instructions in the memory 702 to execute any of the above-mentioned wind turbine operating status identification methods.

[0106] In the present invention, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0107] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0108] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in the present invention may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown in the present invention, but will conform to the widest scope consistent with the principles and novel features disclosed in the present invention.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for identifying the operating status of a wind turbine generator set, characterized in that: include: According to the order of time periods, each type of time domain operation data of the wind turbine in each time period is converted into a corresponding frequency domain data group, wherein one type of time domain operation data involves multiple time periods, and one type of time domain operation data of one time period corresponds to one frequency domain data group, and the types include: generator high-speed shaft speed and nacelle left and right vibration acceleration; For any type, the frequency domain data groups corresponding to each time period of the type are arranged and combined in the order of time periods to obtain a corresponding frequency domain data matrix, wherein one frequency domain data group serves as a row vector of the frequency domain data matrix; The frequency domain data matrices corresponding to each of the types are input into a pre-trained AttachNet neural network to obtain the unit operation status recognition result output by the AttachNet neural network, wherein the frequency domain data matrices corresponding to different types are input into different additional branches of the AttachNet neural network.

2. The method according to claim 1, characterized in that The method converts the time domain operation data of each type of wind turbine generator set in each time period into a corresponding frequency domain data group in the order of the time period, including: For any type of time domain operation data, the time domain operation data of different time periods are converted from the time domain to the frequency domain in the order of the time period, and sampled in the form of row vectors to obtain the corresponding frequency domain data group, wherein the frequency domain value range of the frequency domain data group is 0 to 5 Hz, and the resolution of the frequency domain data group is 0.04 Hz.

3. The method according to claim 1, characterized in that For any type, the frequency domain data groups corresponding to each time period of the type are arranged and combined in the order of time periods to obtain a corresponding frequency domain data matrix, including: For any type, the frequency domain data groups corresponding to each time period of the type are arranged from top to bottom in the order of time periods to obtain the corresponding frequency domain data matrix, wherein the frequency domain data group with the earliest time period is used as the first row vector of the frequency domain data matrix, and the frequency domain data group with the latest time period is used as the last row vector of the frequency domain data matrix.

4. The method according to claim 1, characterized in that The frequency domain data matrix corresponding to each type is input into a pre-trained AttachNet neural network to obtain the unit operation status recognition result output by the AttachNet neural network, including: Inputting the frequency domain data matrix corresponding to each of the types into the additional branch corresponding to the AttachNet neural network to obtain 1024 channel features extracted by each of the additional branches based on the corresponding frequency domain data matrix, wherein the feature extraction networks of different additional branches are identical in structure; The attention module of the AttachNet neural network identifies and obtains corresponding feature vectors according to the features of each of the 1024 channels; The feature vector is classified based on Softmax to obtain the corresponding unit operation status recognition result, wherein the loss function of the Softmax is a cross entropy loss function.

5. The method according to claim 4, characterized in that The step of inputting the frequency domain data matrix corresponding to each of the types into the additional branch corresponding to the AttachNet neural network to obtain 1024 channel features extracted by each of the additional branches based on the corresponding frequency domain data matrix comprises: Inputting the frequency domain data matrix corresponding to each of the types into the additional branch corresponding to the AttachNet neural network; The convolutional layers of each additional branch perform convolution, regularization and ReLU activation function operations on their respective frequency domain data matrices to extract corresponding 1024 channel features.

6. The method according to claim 4, characterized in that The attention module of the AttachNet neural network identifies and obtains corresponding feature vectors based on the features of each of the 1024 channels, including: The spatial attention module in the attention module connects the 1024 channel features in the order of the corresponding additional branches to obtain a connection feature; The fusion module in the attention module performs convolution, regularization, ReLU activation function operation, average pooling operation and maximum pooling operation on the connection features to obtain a corresponding feature vector.

7. A device for identifying the operating status of a wind turbine generator set, characterized in that: include: Data conversion unit, matrix combination unit and state recognition unit; The data conversion unit is used to convert various types of time domain operation data of the wind turbine in each time period into corresponding frequency domain data groups in the order of time periods, wherein one type of time domain operation data involves multiple time periods, and one type of time domain operation data of one time period corresponds to one frequency domain data group, and the types include: generator high-speed shaft speed and nacelle left and right vibration acceleration; The matrix combination unit is used for arranging and combining the frequency domain data groups corresponding to each time period of the type in the order of time periods for any type to obtain a corresponding frequency domain data matrix, wherein one frequency domain data group serves as a row vector of the frequency domain data matrix; The state recognition unit is used to input the frequency domain data matrices corresponding to each of the types into a pre-trained AttachNet neural network to obtain the unit operation state recognition result output by the AttachNet neural network, wherein the frequency domain data matrices corresponding to different types are input into different additional branches of the AttachNet neural network.

8. The device according to claim 7, characterized in that The data conversion unit includes: a data conversion module; The data conversion module is used to convert the time domain operation data of different time periods from the time domain to the frequency domain in the order of time periods for any type of time domain operation data, and sample the corresponding frequency domain data group in the form of row vectors, wherein the frequency domain value range of the frequency domain data group is 0 to 5 Hz, and the resolution of the frequency domain data group is 0.04 Hz.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the method for identifying the operating status of a wind turbine set according to any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: The electronic device includes at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the wind turbine operating status identification method as described in any one of claims 1 to 6.

Images