Wind turbine generator fault evaluation model training method and fault diagnosis method

Abstract

The application provides a wind turbine generator fault evaluation model training method, comprising: performing feature extraction on obtained original vibration data and original temperature data of a generator, inputting the extracted features into a to-be-trained model to obtain a fusion index sequence, determining a fault type and a fault mode of the wind turbine generator based on the fusion index sequence and a preset mapping relationship and a contribution degree corresponding to each feature; and adjusting model parameters based on residual features in the original fusion index sequence and a predicted fusion index sequence to obtain a wind turbine generator fault evaluation model. Through the above method, the wind turbine generator fault evaluation model can be more effectively trained, the data processing efficiency of the wind turbine generator fault evaluation model can be effectively improved, batch data can be quickly processed, and the possible fault type and fault mode of the wind turbine generator can be timely warned.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of data analysis, in particular to a wind turbine generator fault evaluation model training method and a wind turbine generator fault diagnosis method. BACKGROUND

[0002] With the increasingly serious energy crisis, the development and utilization of wind energy, as a substitute for traditional energy and having the characteristics of renewability and cleanliness, gradually attracts the attention of people all over the world. On the other hand, with the continuous progress of development technology and the continuous reduction of development cost in the past half century, the wind power industry has shown an unprecedented prosperity in recent years. While the wind power industry is growing rapidly and the unit capacity of the unit is gradually increasing, the operation safety of the unit is also attracting more and more attention. Among them, the generator, as one of the core components of the wind turbine transmission chain, is mainly responsible for completing the conversion of mechanical energy to electrical energy, and its health status directly affects the performance of the entire unit. Moreover, wind farms are mostly built in remote areas away from cities or near-sea areas, where transportation is not convenient, and the generator is located at a high altitude, so it is very difficult to maintain. In addition, the generator of the wind turbine has complex operating conditions and high failure frequency. Once a fault occurs, the maintenance period is long and the maintenance cost is high, so the state evaluation and fault diagnosis method of the generator is of great significance to ensure the safe and reliable operation of the wind turbine, reduce the operation and maintenance cost, and efficiently utilize wind energy.

[0003] At present, the data source for the state evaluation and fault diagnosis of the wind turbine generator mainly comes from the slowly varying variable SCADA data and the high-frequency CMS data. The two kinds of data have the problem of not being fused and shared with each other. The SCADA data has comprehensive monitoring types, but the sampling frequency is low, and it is difficult to analyze the specific information of the generator fault. The high-frequency CMS data contains rich operating information of the generator, and the mechanical failure mechanism can be analyzed through various signal processing methods, but the wind turbine is in a complex environment and has complex operation, and the high-frequency CMS data is more susceptible to interference, which reduces its monitoring effect. When using only SCADA data or high-frequency CMS data for early warning of the wind turbine generator, although the condition of a certain component or a certain measuring point can be reflected, the failure mode of the entire generator cannot be captured, and even the conclusions of the same measuring point are inconsistent, resulting in a high false alarm rate.

[0004] From two data sources, the current fault diagnosis of the wind turbine generator can be realized by the following schemes: (1) the traditional threshold fault warning method is that when some features of SCADA data or high-frequency CMS exceed the set threshold, the alarm information of the corresponding measuring point will be generated, but at this time the fault has often occurred, which cannot realize timely warning, and cannot accurately judge the fault type. (2) the mechanism diagnosis method based on high-frequency CMS data, the health status of the generator is obtained by manually analyzing the features or spectrum after signal processing of the high-frequency CMS data, which depends on personnel experience and human resources, is not intelligent enough, and cannot effectively deal with the rapid processing of batch data. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a wind turbine generator fault evaluation model training method, system, computer device, computer readable storage medium and wind turbine generator fault diagnosis method, which can solve the problems that the existing wind turbine generator fault diagnosis method cannot realize timely warning and cannot effectively deal with the rapid processing of batch data.

[0006] One aspect of an embodiment of the present application provides a wind turbine generator fault evaluation model training method, comprising:

[0007] Obtain a plurality of sets of original data and corresponding original fault conclusion data of a wind turbine generator, divide the plurality of sets of original data into a training set and a prediction set, one set of original data comprising original vibration data and original temperature data, the original data in the training set and the prediction set being different sets of data;

[0008] Extract corresponding first features from a plurality of original vibration data in the training set and corresponding second features from a plurality of original temperature data in the training set, extract corresponding third features from a plurality of original vibration data in the prediction set and corresponding fourth features from a plurality of original temperature data in the prediction set;

[0009] Input a plurality of first features and corresponding a plurality of second features into a pre-set to-be-trained model, and fuse to obtain an original fusion index sequence, input a plurality of third features and corresponding a plurality of fourth features into the to-be-trained model, and fuse to obtain a prediction fusion index sequence;

[0010] Obtain a plurality of wind turbine generator state evaluation standard values based on the original fusion index sequence, the prediction fusion index sequence and a pre-set mapping relationship;

[0011] calculate contribution degrees of each type of data in the predicted fusion index sequence based on the plurality of wind turbine generator state evaluation standard values, and determine a fault type and a corresponding fault mode of the wind turbine generator based on the obtained plurality of contribution degrees; and

[0012] adjust model parameters of the to-be-trained model based on residual features in the original fusion index sequence and the predicted fusion index sequence respectively, and obtain a wind turbine generator fault evaluation model according to the adjusted model parameters, the wind turbine generator fault evaluation model being used for fault diagnosis on to-be-evaluated data of the wind turbine generator.

[0013] Optionally, the first features include generator vibration intensity features, bearing mechanism fault factors and shaft mechanism fault factors; and the second features include bearing temperature features, generator cooling air temperature features and environmental temperature features.

[0014] The extracting of the corresponding first features from the plurality of original vibration data in the training set and the extracting of the corresponding second features from the plurality of original temperature data in the training set include:

[0015] The corresponding generator vibration intensity features are extracted from the plurality of original vibration data in the training set.

[0016] Each original vibration data in the training set is transformed to obtain corresponding frequency domain features.

[0017] The corresponding bearing mechanism fault factors and shaft mechanism fault factors are calculated based on the plurality of frequency domain features, wherein the bearing mechanism fault factors include a bearing inner ring mechanism fault factor, a bearing outer ring mechanism fault factor, a bearing cage mechanism fault factor and a bearing rolling element mechanism fault factor, and the shaft mechanism fault factors include an unbalance mechanism fault factor, a misalignment mechanism fault factor, a shaft bending mechanism fault factor and a foundation looseness mechanism fault factor.

[0018] The corresponding bearing temperature features, generator cooling air temperature features and environmental temperature features are extracted from the plurality of original temperature data in the training set based on time data corresponding to the plurality of original vibration data.

[0019] A corresponding bearing relative temperature effective value is constructed based on the bearing temperature features, the generator cooling air temperature features and the environmental temperature features.

[0020] Optionally, the inputting of the plurality of first features and the corresponding plurality of second features into a preset to-be-trained model to obtain an original fusion index sequence includes:

[0021] combining the generator vibration intensity features, the bearing inner ring mechanism fault factors, the bearing outer ring mechanism fault factors, the bearing retainer mechanism fault factors, the bearing rolling element mechanism fault factors, the unbalance mechanism fault factors, the misalignment mechanism fault factors, the shaft bending mechanism fault factors, the foundation loosening mechanism fault factors, the bearing temperature features, the generator cooling air temperature features, the ambient temperature features, and the bearing relative temperature effective values to generate a plurality of original feature vectors;

[0022] inputting the plurality of original feature vectors into the to-be-trained model for reconstruction to obtain an original reconstruction residual feature matrix; and

[0023] based on the original reconstruction residual feature matrix, calculating the original fusion indicator sequence.

[0024] Optionally, based on the original fusion indicator sequence, the predicted fusion indicator sequence, and a preset mapping relationship, a plurality of wind turbine generator state evaluation standard values are obtained, including:

[0025] evaluating the original fusion indicator sequence to obtain a generator state evaluation threshold value; and

[0026] based on the preset mapping relationship between the predicted fusion indicator sequence and the generator state evaluation threshold value, a plurality of wind turbine generator state evaluation standard values are calculated.

[0027] Optionally, based on a plurality of wind turbine generator state evaluation standard values, the contribution degree of each type of data in the predicted fusion indicator sequence is calculated, and the fault type and the corresponding fault mode of the wind turbine generator are determined based on the obtained plurality of contribution degrees, including:

[0028] based on a plurality of wind turbine generator state evaluation standard values, determining the target state corresponding to each group of data in the prediction set;

[0029] when the target state is different from a preset state, based on the predicted fusion indicator sequence, the contribution degree of a plurality of generator vibration intensity features and the contribution degree of a plurality of bearing relative temperature effective values are calculated;

[0030] based on the contribution degree of a plurality of generator vibration intensity features and the contribution degree of a plurality of bearing relative temperature effective values, a plurality of contribution degree difference factors are calculated;

[0031] binning a plurality of contribution degree difference factors to determine the fault type of the wind turbine generator;

[0032] acquire at least one type of target feature from the predicted fusion index sequence based on the fault type;

[0033] calculate a contribution degree of each type of target feature, and determine a fault mode of the wind turbine generator based on the contribution degree of each type of target feature.

[0034] Optionally, the calculating the contribution degree of each type of target feature, and determining the fault mode of the wind turbine generator based on the contribution degree of each type of target feature comprises:

[0035] calculating the contribution degree of each type of target feature, and calculating a mean value of the contribution degree corresponding to each type of target feature based on the contribution degree of each type of target feature; and

[0036] defining a target feature with the largest mean value of the contribution degree to determine the fault mode of the wind turbine generator.

[0037] An aspect of an embodiment of the present application further provides a wind turbine generator fault evaluation model training system, comprising:

[0038] an acquisition module configured to acquire multiple groups of original data and corresponding original fault conclusion data of a wind turbine generator, divide the multiple groups of original data into a training set and a prediction set, wherein one group of original data comprises original vibration data and original temperature data, and the original data in the training set and the prediction set are different groups of data;

[0039] an extraction module configured to extract corresponding first features from multiple original vibration data in the training set, extract corresponding second features from multiple original temperature data in the training set, extract corresponding third features from multiple original vibration data in the prediction set, and extract corresponding fourth features from multiple original temperature data in the prediction set;

[0040] a fusion module configured to input multiple first features and corresponding multiple second features into a preset to-be-trained model to obtain an original fusion index sequence, and input multiple third features and corresponding multiple fourth features into the to-be-trained model to obtain a predicted fusion index sequence;

[0041] a calculation module configured to obtain multiple wind turbine generator state evaluation standard values based on the original fusion index sequence, the predicted fusion index sequence, and a preset mapping relationship;

[0042] a determination module configured to calculate a contribution degree of each type of data in the predicted fusion index sequence based on the multiple wind turbine generator state evaluation standard values, and determine a fault type and a corresponding fault mode of the wind turbine generator based on the obtained multiple contribution degrees; and

[0043] An adjusting module is configured to adjust model parameters of a to-be-trained model based on residual features in the original fusion index sequence and the predicted fusion index sequence respectively, and obtain a wind turbine generator fault evaluation model according to the adjusted model parameters, the wind turbine generator fault evaluation model being configured to perform fault diagnosis on to-be-evaluated data of the wind turbine generator.

[0044] An aspect of an embodiment of the present application further provides a wind turbine generator fault diagnosis method, comprising:

[0045] Obtaining to-be-evaluated data of the wind turbine generator, extracting a corresponding third feature from the to-be-evaluated data and extracting a corresponding fourth feature from the to-be-evaluated data;

[0046] Inputting the third feature and the corresponding fourth feature into the wind turbine generator fault evaluation model as described above to obtain a target fusion index sequence through fusion;

[0047] Obtaining a target wind turbine generator state evaluation standard value based on the target fusion index sequence and a preset mapping relationship; and

[0048] Based on the target wind turbine generator state evaluation standard value, calculating target contribution degrees of various types of data in the target fusion index sequence, and determining a target fault type and a corresponding target fault mode of the wind turbine generator based on the obtained multiple target contribution degrees.

[0049] An aspect of an embodiment of the present application further provides a computer device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements steps of the wind turbine generator fault evaluation model training method as described above when executing the computer program.

[0050] An aspect of an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executable on at least one processor to make the at least one processor execute steps of the wind turbine generator fault evaluation model training method as described above.

[0051] The wind turbine generator fault evaluation model training method, system, computer device, computer readable storage medium and wind turbine generator fault diagnosis method provided by the embodiment of the present application can more effectively train the wind turbine generator fault evaluation model, effectively improve the data processing efficiency of the wind turbine generator fault evaluation model, and can timely warn the possible fault type and fault mode of the wind turbine generator, effectively cope with the rapid processing of batch data, and has high generator fault evaluation efficiency.

[0052] The present application is described in detail below with reference to the accompanying drawings and specific embodiments, but is not limited to the present application. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 The environment application schematic diagram of the wind turbine generator fault evaluation model training method according to the embodiment of the present application is schematically shown;

[0054] Fig. 2 schematically shows an example flowchart of the wind turbine generator fault evaluation model training method according to the present application;

[0055] Fig. 3 schematically shows a step flowchart of the wind turbine generator fault evaluation model training method according to the present application;

[0056] Figure 4 The step flowchart of the wind turbine generator fault evaluation model training method according to the present application is schematically shown;

[0057] Figure 5 The step flowchart of the wind turbine generator fault evaluation model training method according to the present application is schematically shown;

[0058] Figure 6 The step flowchart of the wind turbine generator fault evaluation model training method according to the present application is schematically shown;

[0059] Figure 7The diagram schematically illustrates a wind turbine generator fault assessment model training system according to Embodiment 2 of the present invention.

[0060] Figure 8 This schematically illustrates a flowchart of the steps for implementing a wind turbine generator fault diagnosis method according to the present invention;

[0061] Figure 9 The schematic diagram illustrates the hardware structure of a computer device suitable for implementing a wind turbine generator fault assessment model training method according to Embodiment 4 of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0063] It should be noted that the descriptions involving "first," "second," etc., in the embodiments of the present invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0064] In the description of this invention, it should be understood that the numerical labels before the steps do not indicate the order in which the steps are performed, but are only used to facilitate the description of this invention and to distinguish each step, and therefore should not be construed as a limitation of this invention.

[0065] Example 1

[0066] Please see Figure 1 This document illustrates a flowchart of the steps in training a wind turbine generator fault assessment model according to an embodiment of the present invention. It is understood that the flowchart in this embodiment is not intended to limit the order of execution steps. The following description uses a computer device as the execution subject as an example:

[0067] like Figure 1 As shown, the wind turbine generator fault assessment model training method may include steps S100 to S110, wherein:

[0068] Step S100, a plurality of groups of original data of a wind turbine generator and corresponding original fault conclusion data are acquired, the plurality of groups of original data are divided into a training set and a prediction set, one group of original data includes original vibration data and original temperature data, and the original data in the training set and the prediction set are different groups of data.

[0069] For example, in order to train the model more effectively, the monitoring data of the wind turbine generator under the preset working condition is screened out, the data under the preset working condition can include SCADA temperature data (temperature monitoring data, i.e. original temperature data) and high-frequency CMS data (high-frequency state monitoring data, i.e. original vibration data) when the unit is full of 80% or more. The plurality of groups of original data of the wind turbine generator collected are divided into a training set and a prediction set, each group of original data in the training set is different groups of data, and each group of original data in the prediction set is different groups of data, wherein the training set further includes a validation set.

[0070] Step S102, corresponding first features are extracted from a plurality of original vibration data in the training set, and corresponding second features are extracted from a plurality of original temperature data in the training set, corresponding third features are extracted from a plurality of original vibration data in the prediction set, and corresponding fourth features are extracted from a plurality of original temperature data in the prediction set.

[0071] The first features include generator vibration intensity features RMS, bearing mechanism fault factors and shaft mechanism fault factors; the second features include bearing temperature features generator_bearing_temp, generator cooling air temperature features generator_coolingair_temp and environmental temperature features env_temp. Please refer to Figure 2-1 In an exemplary embodiment,

[0072] The step S102 of extracting corresponding first features from the plurality of original vibration data in the training set and extracting corresponding second features from the plurality of original temperature data in the training set further comprises the following operations: a step S200A of extracting corresponding generator vibration intensity features from the plurality of original vibration data in the training set; a step S202A of transforming each original vibration data in the training set to obtain corresponding frequency domain features; a step S204A of calculating corresponding bearing mechanism fault factors and shaft mechanism fault factors based on the plurality of frequency domain features, wherein the bearing mechanism fault factors comprise a bearing inner ring mechanism fault factor, a bearing outer ring mechanism fault factor, a bearing cage mechanism fault factor, and a bearing rolling element mechanism fault factor, and the shaft mechanism fault factors comprise an imbalance mechanism fault factor, a misalignment mechanism fault factor, a shaft bending mechanism fault factor, and a foundation looseness mechanism fault factor; a step S206A of extracting corresponding bearing temperature features, generator cooling wind temperature features, and environmental temperature features from the plurality of original temperature data in the training set based on time data corresponding to the plurality of original vibration data; and a step S208A of constructing corresponding bearing relative temperature effective values based on the bearing temperature features, the generator cooling wind temperature features, and the environmental temperature features. In this embodiment, the generator vibration intensity features RMS are extracted from the original vibration data. The frequency domain features include first frequency domain features and second frequency domain features. The original vibration data is subjected to fast Fourier transform to obtain fft_spectrum (first frequency domain features), and the original vibration data is subjected to Hilbert envelope spectrum transform to obtain env_spectrum (second frequency domain features). After the frequency domain features are obtained, the generator bearing mechanism fault factors are constructed. The bearing mechanism fault factors include a bearing inner ring mechanism fault factor inner_factor, a bearing outer ring mechanism fault factor outer_factor, a bearing cage mechanism fault factor cage_factor, and a bearing rolling element mechanism fault factor roll_factor. The maximum values in the bearing inner ring front 4 times fault frequency interval are searched in the fft_spectrum and summed, that is, the maximum values within one frequency resolution left and right of the bearing inner ring fault frequency are searched and summed to obtain fft_energy (first energy feature), and the maximum values in the bearing inner ring front 4 times fault frequency interval are searched in the env_spectrum and summed to obtain env_energy (second energy feature). Based on the first energy feature and the second energy feature, the bearing inner ring mechanism fault factor inner_factor is calculated using the L2 norm function of regularized energy, which is specifically realized by the following L2 norm function:

[0073]

[0074] Similarly, the bearing outer ring mechanism failure factor outer_factor, the bearing cage mechanism failure factor cage_factor, and the bearing rolling element mechanism failure factor roll_factor can be calculated.

[0075] The generator bearing mechanism failure factor is constructed, the maximum value in the interval of the front 4 times of the failure frequency of the bearing inner ring in fft_spectrum is searched, and the sum is recorded as fft_energy, the maximum value in the interval of the front 4 times of the failure frequency of the bearing inner ring in env_spectrum is searched, and the sum is recorded as env_energy. The method for defining the bearing inner ring mechanism failure factor uses the L2 norm of the regularized energy, which is specifically: Similarly, there are outer_factor, cage_factor, and roll_factor.

[0076] The shaft mechanism failure factor is constructed, the maximum value in the interval of the front 10 times of the characteristic frequency of the rotating frequency in fft_spectrum is searched, that is, the maximum value in the interval of the characteristic frequency and one frequency resolution on both sides is recorded as 1X~10X,

[0077] The unbalance mechanism failure factor is defined as: unbalance_factor=1X / 2X;

[0078] The misalignment mechanism failure factor is defined as: misalignment=(1X+2X) / 1X;

[0079] The shaft bending mechanism failure factor is defined as: bent_shaft_factor=2X-1X;

[0080] The looseness mechanism failure factor is defined as: looseness_factor=(4X+...+10X) / (1X+2X+3X);

[0081] The generator bearing temperature feature corresponding to the time dimension of the original vibration data is extracted from the original temperature data and recorded as generator_bearing_temp, the generator cooling air temperature feature is recorded as generator_coolingair_temp, and the environment temperature is recorded as env_temp, which ensures the consistency of the data of the two sources in the sampling time, and the relative temperature effective value TRMS is constructed to avoid the error caused by a single temperature variable,

[0082] Please refer to Figure 2-2In the example embodiment, the third features include generator vibration intensity features, bearing mechanism failure factors and rotating shaft mechanism failure factors; the fourth features include bearing temperature features, generator cooling air temperature features and ambient temperature features; the step S102 of extracting corresponding third features from the plurality of original vibration data in the prediction set and extracting corresponding fourth features from the plurality of original temperature data in the prediction set can also be obtained from the following operations: a step S200B of extracting corresponding generator vibration intensity features from the plurality of original vibration data in the prediction set; a step S202B of transforming each original vibration data in the prediction set to obtain corresponding frequency domain features; a step S204B of calculating corresponding bearing mechanism failure factors and rotating shaft mechanism failure factors based on the plurality of frequency domain features, wherein the bearing mechanism failure factors include bearing inner ring mechanism failure factors, bearing outer ring mechanism failure factors, bearing cage mechanism failure factors and bearing rolling element mechanism failure factors, and the rotating shaft mechanism failure factors include unbalance mechanism failure factors, misalignment mechanism failure factors, rotating shaft bending mechanism failure factors and base looseness mechanism failure factors; a step S206B of extracting corresponding bearing temperature features, generator cooling air temperature features and ambient temperature features from the plurality of original temperature data in the prediction set based on time data corresponding to the plurality of original vibration data; and a step S208B of constructing corresponding bearing relative temperature effective values based on the bearing temperature features, the generator cooling air temperature features and the ambient temperature features. In this embodiment, the step flow and method of extracting third features and fourth features from the prediction set are basically the same as the step flow and method of extracting first features and second features from the training set, and will not be described here.

[0083] In step S104, the plurality of first features and the corresponding plurality of second features are input into a pre-set to-be-trained model to fuse to obtain an original fusion index sequence, and the plurality of third features and the corresponding plurality of fourth features are input into the to-be-trained model to fuse to obtain a prediction fusion index sequence.

[0084] The to-be-trained model can be a deep auto-encoder model, i.e., a DAE model. The DAE model structure is as follows:

[0085] The input layer dimension input_shape=(None, 1, 13);

[0086] The encoding layer is composed of three Dense layers and one Dropout layer, the number of neurons is 128, 64, 32 in turn, and the activation function is relu function (Linear rectification function) which has better sparseness and can avoid gradient disappearance. The Dropout layer is added after the first Dense layer with a random inactivation rate of 20% to prevent overfitting of the model to be trained.

[0087] The potential expression layer is composed of one Dense layer, the number of neurons is 16, and the activation function is sigmoid function (S-type function).

[0088] The decoding layer is composed of three Dense layers, the number of neurons is 32, 64, 128 in turn, and the activation function is relu function.

[0089] The output layer is composed of one Dense layer, the number of neurons is 13.

[0090] The model optimizer selects Adam algorithm (Adaptive Moment Estimation, self-adaptive moment estimation algorithm), and the loss function selects Mae function (Mean Absolute Error, mean absolute error function).

[0091] Please refer to Figure 3-1 In the exemplary embodiment, the step S104 of inputting the plurality of first features and the corresponding plurality of second features into the preset to-be-trained model to obtain the original fusion index sequence can further include steps S300A-S304A, wherein: step S300A, combining the corresponding generator vibration intensity features, bearing inner ring mechanism fault factors, bearing outer ring mechanism fault factors, bearing retainer mechanism fault factors, bearing rolling element mechanism fault factors, unbalance mechanism fault factors, misalignment mechanism fault factors, shaft bending mechanism fault factors, foundation loosening mechanism fault factors, bearing temperature features, generator cooling wind temperature features, environmental temperature features and bearing relative temperature effective values to generate a plurality of original feature vectors; step S302A, inputting the plurality of original feature vectors into the to-be-trained model for reconstruction to obtain an original reconstruction residual feature matrix; and step S304A, calculating the original fusion index sequence based on the original reconstruction residual feature matrix. In this embodiment, a plurality of first features and a plurality of second features are combined to obtain a plurality of original feature vectors feature_vector, wherein the original feature vector feature_vector includes 9 types of original vibration data and 4 types of original temperature data, and the feature sequence of the original feature vector is as follows:

[0092]

[0093] The plurality of feature vectors form a feature matrix, denoted as x after standardization. After the DAE model is trained, a reconstruction matrix is obtained The original reconstruction residual feature matrix M is defined as The RMS and TRMS columns of features are extracted from M to form the original matrix Mf, and the Mahalanobis distance of Mf is calculated to obtain the fusion index sequence fusion_factor of the original vibration data and the original temperature data. The inverse covariance matrix invMf and the sample mean meanMf of Mf are retained during the calculation of the Mahalanobis distance.

[0094] Please refer to Figure 3-2 In the exemplary embodiment, the step S104 of inputting the plurality of third features and the corresponding plurality of fourth features into the to-be-trained model to obtain a predicted fusion index sequence by fusion can further include steps S300B-S304B, wherein: step S300B, combining the corresponding generator vibration intensity features, bearing inner ring mechanism fault factors, bearing outer ring mechanism fault factors, bearing retainer mechanism fault factors, bearing rolling element mechanism fault factors, unbalance mechanism fault factors, misalignment mechanism fault factors, shaft bending mechanism fault factors, foundation loosening mechanism fault factors, bearing temperature features, generator cooling wind temperature features, environmental temperature features, and bearing relative temperature effective values to generate a plurality of predicted feature vectors; step S302B, inputting the plurality of predicted feature vectors into the to-be-trained model for reconstruction to obtain a predicted reconstruction residual feature matrix; and step S304B, calculating the predicted fusion index sequence based on the predicted reconstruction residual feature matrix. In this embodiment, the difference between the step of calculating the predicted fusion index sequence and the method of calculating the original fusion index sequence is that when calculating the Mahalanobis distance of the prediction set data, the invMf and meanMf of Mf in the matrix x data are used to obtain the corresponding predicted fusion index sequence fusion_factor_pred of the prediction set; the remaining steps and methods are basically the same, and will not be repeated here.

[0095] Step S106, based on the original fusion index sequence, the predicted fusion index sequence, and a preset mapping relationship, a plurality of wind turbine generator state evaluation standard values are obtained. Please refer to Figure 4In the example embodiment, the step S106 of obtaining the plurality of wind turbine generator state evaluation standard values based on the original fusion index sequence, the predicted fusion index sequence and the preset mapping relationship can also be obtained by the following operations, specifically as follows: a step S400 of evaluating the original fusion index sequence to obtain a generator state evaluation threshold value; and a step S402 of calculating the plurality of wind turbine generator state evaluation standard values based on the preset mapping relationship between the predicted fusion index sequence and the generator state evaluation threshold value. In this embodiment, since the distribution of the original fusion index sequence fusion_factor is unknown, the Chebyshev theorem is used to obtain the generator state evaluation threshold value th according to the following formula: wherein ε=5σ, i.e. at least 24 / 25

or 96%

[0096] SOH=sigmoid(fusion_factor_pred / th).

[0097] A step S108 of calculating the contribution degree of each type of data in the predicted fusion index sequence based on the plurality of wind turbine generator state evaluation standard values, and determining the fault type and the corresponding fault mode of the wind turbine generator based on the obtained plurality of contribution degrees.

[0098] Please refer to Figure 5In the example embodiment, the step S108 of calculating the contribution degree of each type of data in the prediction fusion index sequence based on the plurality of wind turbine generator state evaluation standard values and determining the fault type and corresponding fault mode of the wind turbine generator based on the obtained plurality of contribution degrees can further include steps S500-S510, wherein: step S500, determining the target state corresponding to each group of data in the prediction set based on the plurality of wind turbine generator state evaluation standard values; step S502, when the target state is different from the preset state, calculating the contribution degree corresponding to the plurality of generator vibration intensity features and the contribution degree corresponding to the plurality of bearing relative temperature effective values based on the prediction fusion index sequence; step S504, calculating the corresponding plurality of contribution degree difference factors based on the contribution degree corresponding to the plurality of generator vibration intensity features and the contribution degree corresponding to the plurality of bearing relative temperature effective values; step S506, binning the plurality of contribution degree difference factors to determine the fault type of the wind turbine generator; step S508, based on the fault type, obtaining at least one type of target feature from the prediction fusion index sequence; and step S510, calculating the contribution degree of each type of target feature and determining the fault mode of the wind turbine generator based on the contribution degree of each type of target feature. In this embodiment, the SOH is divided into four states, specifically [0-0.2 is the normal state, 0.2-0.4 is the attention state, 0.4-0.6 is the pre-warning state, and 0.6-1 is the alarm state], and the mapping function is sigmoid = 1 / (1+e -(fusion_factor_pred / th)+c ), where c is sigmoid = 0.2, and fusion_factor_pred / th = 1. For the generator SOH judgment of batch prediction set data, the binning method is used for evaluation to enhance the robustness of the entire algorithm. In the four intervals [0-0.2 is the normal state, 0.2-0.4 is the attention state, 0.4-0.6 is the pre-warning state, and 0.6-1 is the alarm state], if the proportion of the number of batch prediction data in the attention state and above is more than 30%, it is classified as the state corresponding to the interval, i.e., the target state is determined. The preset state is the normal state.

[0099] When the generator state is in the attention state and above, the contribution degree of Mf_pred is calculated.

[0100] For the contribution degree of RMS and TRMS features in Mf_pred, where importance i,j is the contribution degree of the jth feature of the ith group of data; fusion_factor_pred i is the fusion index of the ith group of data; fusion_factor_pred i,jThe fusion index of the jth feature is removed for the ith group of data; softmax is a mapping function that maps the contribution degree to between 0 and 1.

[0101] The contribution degree difference factor is defined by the following function: Where i is the ith group of data.

[0102] Similarly, the diff_importance sequence is binned and counted, that is, [-1-0.33 temperature rise failure, -0.33-0.33 mechanical failure, 0.33-1 mechanical wear], and the maximum value of the number of batch prediction data is assigned to a certain interval, that is, the fault type is determined. The data corresponding to the fault type is selected by the generator state evaluation threshold, and further fault diagnosis is performed.

[0103] See Figure 6 In the exemplary embodiment, the step S510 of calculating the contribution degree of each target feature and determining the fault mode of the wind turbine generator based on the contribution degree of each target feature can further include: step S600, calculating the contribution degree of each target feature, and based on the contribution degree of each target feature, calculating the contribution degree mean value corresponding to each target feature; and step S602, defining the target feature with the maximum contribution degree mean value to determine the fault mode of the wind turbine generator. In this embodiment, when the wind turbine generator belongs to a certain type of failure, the contribution degree of the corresponding target feature is analyzed to further perform fault diagnosis, as follows:

[0104] When the number of data in the prediction set has the maximum contribution degree difference factor in the interval [-1-0.33], the generator failure is defined as a temperature rise failure, and then

[0105] [generator_bearing_temp, cooling_air_temp, env_temp] are used for contribution degree calculation, and finally the mean value of each type of contribution degree of the selected data is calculated, and the feature with the highest mean value is defined as the specific fault mode, which includes: generator bearing lubrication failure, generator cooling air abnormality, high ambient temperature.

[0106] When the number of data in the prediction set has the maximum contribution degree difference factor in the interval [-0.33-0.33], the generator failure is defined as a mechanical failure, and then

[0107] Contribution degree calculation is performed, and finally the average value of each type of contribution degree of the screened data is calculated. The feature with the highest average value is defined as the specific failure mode, which includes: generator bearing inner ring failure, generator bearing outer ring failure, generator bearing cage failure, generator bearing rolling element (roller) failure, generator rotor imbalance, generator rotor misalignment, generator shaft bending failure, and generator foundation loosening failure.

[0108] When the contribution degree difference factor of the data in the prediction set is the largest in the interval [0.33-1], the generator failure is defined as a mechanical failure, and then

[0109] Contribution degree calculation is performed, and finally the average value of each type of contribution degree of the screened data is calculated. The feature with the highest average value is defined as the specific failure mode, which includes: generator bearing inner ring failure, generator bearing outer ring failure, generator bearing cage failure, generator bearing rolling element (roller) failure, generator rotor imbalance, generator rotor misalignment, generator shaft bending failure, and generator foundation loosening failure.

[0110] In other embodiments, from the perspective of failure mechanism, it is considered that early wear of the device has little effect on temperature rise. After obtaining the contribution degree average value, the maximum contribution degree average value is taken as the device state evaluation weight value, and the device SOH is re-evaluated, i.e. when this phenomenon occurs, the SOH is updated as SOH*importance, to re-determine the failure type and failure mode.

[0111] In step S110, the model parameters of the to-be-trained model are adjusted based on the residual features in the original fusion index sequence and the predicted fusion index sequence, respectively. The wind turbine generator fault evaluation model is obtained according to the adjusted model parameters, and the wind turbine generator fault evaluation model is used for fault diagnosis on the to-be-evaluated data of the wind turbine generator.

[0112] In an exemplary embodiment, if the residual features in the original fusion index sequence and the residual features in the predicted fusion index sequence both show a downward trend as a whole, it is considered that the model is normally trained and fitted.

[0113] The following exemplary describes the wind turbine generator fault evaluation model training method based on the original fault conclusion data and corresponding original data of the generator imbalance fault diagnosis. 420 groups of original vibration data and corresponding original temperature data are selected from a certain wind farm, of which the first 380 groups of data are all normal data, of which 300 groups are taken as a training set, and 80 groups are taken as a validation set. 25 groups of generator imbalance fault data are randomly added to the last 50 groups as a test set.

[0114] In engineering applications, the first 380 groups of data are regarded as normal operation data of the wind turbine under the condition of meeting the working condition, and the last 50 groups are regarded as unit evaluation data.

[0115] The 380 groups of data of the normal operation of the unit are subjected to feature extraction and data reconstruction to obtain a generator state evaluation threshold value th, inverse covariance matrix invMf and sample mean meanMf in the index matrix Mf, and a trained DAE model.

[0116] The feature extraction and data reconstruction are performed on the evaluation data to obtain the SOH curve corresponding to the evaluation data, and the proportion of each state of the data in the prediction set is statistically boxed. Among them, the normal state [0-0.2] accounts for 46%, the attention state [0.2-0.4] accounts for 34%, the early warning state [0.4-0.6] accounts for 16%, and the alarm state [0.6-1] accounts for 4%, the target state is the attention state, and the target feature exceeding the generator state evaluation threshold value is further screened out.

[0117] Further, the RMS and TRMS in Mf_pred are subjected to contribution difference factor analysis to obtain a contribution difference factor diff_importance, a plurality of contribution difference factors are subjected to box statistics to determine a target interval. Among them: the mechanical fault [-0.33-0.33] interval accounts for 0.592593; the temperature rise fault [-1-0.33] interval accounts for 0.296296; the mechanical wear [0.33-1] interval accounts for 0.111111, and the fault is defined as a mechanical fault.

[0118] The target residual feature corresponding to the mechanical fault is selected from the corresponding Mf_pred for contribution degree analysis, the contribution degree is subjected to mean value statistics, and the feature name corresponding to the maximum value of the contribution degree mean value corresponds to the generator unbalance fault. And the frequency spectrum analysis of the evaluation data obtains a frequency spectrum diagram, and it can be known from the frequency spectrum diagram that there is obvious 1 times frequency, which belongs to obvious unbalance fault.

[0119] The wind turbine generator fault evaluation model training method and the application of the generator fault evaluation model of the application have at least the following beneficial effects:

[0120] (1) Multi-dimensional construction model, improve the practicability of the model. The two data sources (original vibration data and original temperature data) are fused by using a deep learning framework, the generator state evaluation index is constructed by using SCADA data and high-frequency CMS data, the index is used to distinguish the state (whether healthy) of the generator by using a statistical method, and the contribution degree analysis is further used to distinguish whether the failure mode of the generator is a temperature rise fault or a mechanical fault; meanwhile, from the perspective of failure mode analysis, the mechanism knowledge and the data driven method are effectively combined, and on the basis of the index early warning, the mechanical fault and the temperature rise fault positioning are further realized through the contribution degree analysis.

[0121] (2) The generator comprehensive diagnosis method fusing different signal sources, mechanism and data driven technology can reduce the false positive rate of generator state evaluation and fault diagnosis by using only one type of data without the data support of fault samples, improve the overall fault early warning and fault diagnosis capability of the generator, and has strong practical application value.

[0122] (3) The original temperature data and the original vibration data are fused from the perspective of failure mechanism, the failure mode is analyzed from the perspective of mechanism, the data driven technology is combined, the generator state evaluation, the fault degree judgment and the fault mode positioning can be accurately realized, the manual intervention setting threshold process is less in the intermediate process, the data is screened through the contribution degree analysis, and the reliability of the diagnosis is ensured.

[0123] Example two

[0124] Please continue to refer to Figure 7 , a block diagram of a wind turbine generator fault evaluation model training system is schematically shown. In this embodiment, the wind turbine generator fault evaluation model training system can include or be divided into one or more program modules, one or more program modules are stored in a storage medium and executed by one or more processors to complete the present application, and the above-mentioned wind turbine generator fault evaluation model training method can be realized. The program module referred to in the embodiment of the present application refers to a series of computer program instruction segments that can complete a specific function, and is more suitable for describing the execution process of the wind turbine generator fault evaluation model training system in the storage medium. The following description will specifically introduce the functions of each program module in this embodiment.

[0125] As Figure 7 shown, the wind turbine generator fault evaluation model training system can include an acquisition module 700, an extraction module 702, a fusion module 704, a calculation module 706, a determination module 708 and an adjustment module 710, wherein:

[0126] The acquisition module 700 is configured to acquire a plurality of groups of original data of a wind turbine generator and corresponding original fault conclusion data, divide the plurality of groups of original data into a training set and a prediction set, one group of original data comprising original vibration data and original temperature data, and the original data in the training set and the prediction set being different groups of data;

[0127] The extraction module 702 is configured to extract corresponding first features from a plurality of original vibration data in the training set and second features from a plurality of original temperature data in the training set, and extract third features from a plurality of original vibration data in the prediction set and fourth features from a plurality of original temperature data in the prediction set.

[0128] The fusion module 704 is configured to input a plurality of first features and corresponding second features into a preset to-be-trained model to obtain an original fusion index sequence, and input a plurality of third features and corresponding fourth features into the to-be-trained model to obtain a prediction fusion index sequence.

[0129] The calculation module 706 is configured to obtain a plurality of wind turbine generator state evaluation standard values based on the original fusion index sequence, the prediction fusion index sequence, and a preset mapping relationship.

[0130] The determination module 708 is configured to calculate the contribution degrees of various types of data in the prediction fusion index sequence based on the plurality of wind turbine generator state evaluation standard values, and determine a fault type and a corresponding fault mode of the wind turbine generator based on the obtained plurality of contribution degrees.

[0131] The adjustment module 710 is configured to adjust model parameters of a to-be-trained model based on residual features in the original fusion index sequence and the prediction fusion index sequence, respectively, obtain a wind turbine generator fault evaluation model according to the adjusted model parameters, and use the wind turbine generator fault evaluation model to perform fault diagnosis on to-be-evaluated data of the wind turbine generator.

[0132] Embodiment Three

[0133] Referring to Figure 8 , a step flowchart of a wind turbine generator fault diagnosis method according to an embodiment of the present application is shown. It can be understood that the flowchart in the present method embodiment is not used to limit the order of execution steps. The following is an exemplary description taking a computer device as an execution subject, as follows:

[0134] As Figure 8 indicated, the wind turbine generator fault diagnosis method can include steps S800-S806, wherein:

[0135] Step S800, obtaining the to-be-evaluated data of the wind turbine generator, extracting a corresponding third feature from the to-be-evaluated data and extracting a corresponding fourth feature from the to-be-evaluated data;

[0136] Step S802, inputting the third feature and the corresponding fourth feature into the wind turbine generator fault evaluation model to obtain a target fusion index sequence through fusion;

[0137] Step S804, based on the target fusion index sequence and a preset mapping relationship, evaluating to obtain a target wind turbine generator state evaluation standard value; and

[0138] Step S806, based on the target wind turbine generator state evaluation standard value, calculating a target contribution degree of each type of data in the target fusion index sequence, and determining a target fault type and a corresponding target fault mode of the wind turbine generator based on the obtained multiple target contribution degrees.

[0139] In the embodiment, the operation steps of obtaining the wind turbine generator state evaluation standard value and determining the fault type and the corresponding fault mode of the wind turbine generator based on the multiple contribution degrees are consistent with the operation steps of the above-mentioned embodiments.

[0140] In the present application, the above-mentioned method can quickly realize the generator state evaluation, fault degree judgment and fault mode positioning of the generator, and the data is screened through the contribution degree analysis of various features, which ensures the reliability and effectiveness of the diagnosis, ensures the fault warning effect, and improves the fault diagnosis efficiency.

[0141] Embodiment Four

[0142] Referring to Figure 9 is a hardware architecture schematic diagram of a computer device 10000 suitable for realizing the wind turbine generator fault evaluation model training method of embodiment four of the present application. In the embodiment, the computer device 10000 is a device capable of automatically performing numerical calculation and / or information processing according to pre-set or stored instructions. The computer device 10000 can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a rack server, a blade server, a tower server or a cabinet server (including a standalone server or a server cluster composed of multiple servers), a gateway, etc. As shown in the figure, the computer device 10000 at least includes, but is not limited to, a memory 10010, a processor 10020 and a network interface 10030 which can be connected to each other through a system bus. Among them: Figure 9

[0143] ​The memory 10010 in this embodiment includes at least one type of computer readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 10010 can be an internal storage unit of the computer device 10000, such as a hard disk or memory of the computer device 10000. In other embodiments, the memory 10010 can also be an external storage device of the computer device 10000, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 10000. Of course, the memory 10010 can also include both the internal storage unit and the external storage device of the computer device 10000. In this embodiment, the memory 10010 is generally used to store the operating system and various application software installed on the computer device 10000, such as the program code of the wind turbine generator fault assessment model training system in the above embodiments, etc. In addition, the memory 10010 can also be used to temporarily store various data that has been output or will be output.

[0144] The processor 10020 in some embodiments can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 10020 is generally used to control the overall operation of the computer device 10000, such as performing control and processing related to data interaction or communication of the computer device 10000, etc. In this embodiment, the processor 10020 is used to run the program code or process data stored in the memory 10010, such as running the wind turbine generator fault assessment model training system to implement the wind turbine generator fault assessment model training method of the above embodiments.

[0145] The network interface 10030 can include a wireless network interface or a wired network interface, which is generally used to establish a communication connection between the computer device 10000 and other electronic devices. For example, the network interface 10030 is used to connect the computer device 10000 with an external terminal through a network, establish a data transmission channel and a communication connection between the computer device 10000 and the external terminal, and the like. The network can be an Intranet, the Internet, a Global System for Mobile Communications (GSM), a Wideband Code Division Multiple Access (WCDMA), a 4G network, a 5G network, Bluetooth, Wi-Fi, and the like wireless or wired network.

[0146] It should be noted that, Figure 9 Only the computer device 10000 with components 10010-10030 is shown, but it should be understood that all the shown components are not required to be implemented, and more or fewer components can be alternatively implemented.

[0147] In this embodiment, the wind turbine generator fault assessment model training system stored in the memory 10010 can also be divided into one or more program modules, which are stored in the memory 10010 and executed by one or more processors (in this embodiment, the processor 10020) to complete the present application.

[0148] For example, Figure 7 The program module schematic diagram of the wind turbine generator fault assessment model training system embodiment two is shown, in which the wind turbine generator fault assessment model training system can be divided into an acquisition module 700, an extraction module 702, a fusion module 704, a calculation module 706, a determination module 708, and an adjustment module 710. The program module referred to by the present application refers to a series of computer program instruction segments that can complete a specific function, and is more suitable than a program to describe the execution process of the wind turbine generator fault assessment model training system in the computer device 10000. The specific functions of the program modules 700-710 have been described in detail in embodiment two, and will not be described here.

[0149] Embodiment five

[0150] The embodiment also provides a computer readable storage medium, which stores a computer program. The computer program is executed by at least one processor to implement the steps of the wind turbine generator fault evaluation model training method in the embodiment.

[0151] In the embodiment, the computer readable storage medium includes a flash memory, a hard disk, a multimedia card, a card type memory (for example, an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read only memory (ROM), an electrically erasable programmable read only memory (EEPROM), a programmable read only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, a server, an App application market, etc. In some embodiments, the computer readable storage medium can be an internal storage unit of a computer device, for example, a hard disk or a memory of the computer device. In other embodiments, the computer readable storage medium can also be an external storage device of the computer device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Of course, the computer readable storage medium can also include both the internal storage unit and the external storage device of the computer device. In the embodiment, the computer readable storage medium is usually used to store an operating system and various application software installed on the computer device, for example, program codes of the stuttering detection method in the embodiment, etc. In addition, the computer readable storage medium can also be used to temporarily store various data that have been output or will be output.

[0152] Obviously, those skilled in the art should understand that each module or each step of the above-mentioned embodiment of the application can be realized by a general computing device, which can be concentrated on a single computing device or distributed on a network composed of multiple computing devices, and alternatively, each module or each step can be realized by program codes executable by a computing device, so that each module or each step can be stored in a storage device and executed by a computing device, and in some cases, the steps shown or described can be executed in different sequences, or each module or each step can be manufactured into an individual integrated circuit module or multiple modules or steps can be manufactured into a single integrated circuit module. Therefore, the embodiment of the application is not limited to any particular combination of hardware and software.

[0153] The above-mentioned serial numbers of the embodiments of the application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0154] The above merely describes the preferred embodiments of the present application, and is not intended to limit the patent scope of the present application, and any equivalent structure or equivalent process conversion, or direct or indirect application in other related technical fields, which are made by using the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A training method for a fault assessment model of a wind turbine generator, characterized in that, include: Multiple sets of raw data from a wind turbine generator are acquired, and the multiple sets of raw data are divided into a training set and a prediction set. Each set of raw data includes raw vibration data and raw temperature data. The raw data in the training set and the prediction set are different sets of data. Extract the corresponding first feature from multiple raw vibration data in the training set and the corresponding second feature from multiple raw temperature data in the training set; extract the corresponding third feature from multiple raw vibration data in the prediction set and the corresponding fourth feature from multiple raw temperature data in the prediction set. Multiple first features and corresponding multiple second features are input into a preset training model and fused to obtain an original fusion index sequence. Multiple third features and corresponding multiple fourth features are input into the training model and fused to obtain a predicted fusion index sequence. The training model is a deep autoencoder model. Based on the original fusion index sequence, the predicted fusion index sequence, and the preset mapping relationship, multiple wind turbine generator state evaluation standard values ​​are obtained. Based on multiple wind turbine generator condition assessment standard values, the contribution of various types of data in the predictive fusion index sequence is calculated, and the fault type and corresponding fault mode of the wind turbine generator are determined based on the obtained multiple contribution values. and Based on the residual features in the original fusion index sequence and the predicted fusion index sequence, the model parameters of the model to be trained are adjusted, and a wind turbine generator fault assessment model is obtained according to the adjusted model parameters. The wind turbine generator fault assessment model is used to diagnose faults in the wind turbine generator to be assessed data. The process of obtaining multiple wind turbine generator state assessment standard values ​​based on the original fusion index sequence, the predicted fusion index sequence, and a preset mapping relationship includes: The original fusion index sequence is evaluated to obtain the generator state assessment threshold value; and Based on the preset mapping relationship between the predicted fusion index sequence and the generator condition assessment threshold value, multiple wind turbine generator condition assessment standard values ​​are calculated.

2. The wind turbine generator fault assessment model training method according to claim 1, characterized in that, The first feature includes generator vibration intensity characteristics, bearing mechanism failure factors, and shaft mechanism failure factors; the second feature includes bearing temperature characteristics, generator cooling air temperature characteristics, and ambient temperature characteristics. The step of extracting the corresponding first feature from multiple raw vibration data in the training set and the corresponding second feature from multiple raw temperature data in the training set includes: The corresponding vibration intensity features of the generator are extracted from multiple raw vibration data in the training set; Transform each original vibration data in the training set to obtain the corresponding frequency domain features; Based on multiple frequency domain features, the corresponding bearing mechanism failure factors and shaft mechanism failure factors are calculated. The bearing mechanism failure factors include: bearing inner ring mechanism failure factors, bearing outer ring mechanism failure factors, bearing cage mechanism failure factors, and bearing rolling element mechanism failure factors. The shaft mechanism failure factors include unbalance mechanism failure factors, misalignment mechanism failure factors, shaft bending mechanism failure factors, and foundation loosening mechanism failure factors. Based on the time data corresponding to the multiple original vibration data, the bearing temperature characteristics, generator cooling air temperature characteristics, and ambient temperature characteristics are extracted from the multiple original temperature data in the training set; and Based on the bearing temperature characteristics, the generator cooling air temperature characteristics, and the ambient temperature characteristics, the corresponding effective value of the bearing relative temperature is constructed.

3. The wind turbine generator fault assessment model training method according to claim 2, characterized in that, The step of inputting multiple first features and corresponding multiple second features into a preset training model and fusing them to obtain an original fusion index sequence includes: By combining the corresponding generator vibration intensity characteristics, bearing inner ring mechanism failure factors, bearing outer ring mechanism failure factors, bearing cage mechanism failure factors, bearing rolling element mechanism failure factors, imbalance mechanism failure factors, misalignment mechanism failure factors, shaft bending mechanism failure factors, foundation loosening mechanism failure factors, bearing temperature characteristics, generator cooling air temperature characteristics, ambient temperature characteristics, and bearing relative temperature effective value, multiple original feature vectors are generated. The original feature vectors are input into the model to be trained for reconstruction, resulting in the original reconstructed residual feature matrix; and Based on the original reconstructed residual feature matrix, the original fusion index sequence is calculated.

4. The wind turbine generator fault assessment model training method according to claim 2 or 3, characterized in that, The step of calculating the contribution of various data in the predicted fusion index sequence based on multiple wind turbine generator condition assessment standard values, and determining the fault type and corresponding fault mode of the wind turbine generator based on the obtained multiple contribution values, includes: Based on multiple wind turbine generator state assessment standard values, the target state corresponding to each set of data in the prediction set is determined; When the target state differs from the preset state, the contribution of multiple generator vibration intensity characteristics and multiple bearing relative temperature effective values ​​are calculated based on the predicted fusion index sequence. Based on the contribution of multiple generator vibration intensity characteristics and multiple bearing relative temperature effective values, multiple contribution difference factors are calculated. By statistically analyzing multiple contribution difference factors in each box, the fault type of the wind turbine generator is determined. Based on the fault type, at least one type of target feature is obtained from the predictive fusion index sequence; The contribution of each type of target feature is calculated, and the failure mode of the wind turbine generator is determined based on the contribution of each type of target feature.

5. The wind turbine generator fault assessment model training method according to claim 4, characterized in that, The calculation of the contribution of each type of target feature, and the determination of the fault mode of the wind turbine generator based on the contribution of each type of target feature, includes: Calculate the contribution of each type of target feature, and based on the contribution of each type of target feature, calculate the average contribution of each type of target feature; and The target feature with the highest average contribution is defined to determine the failure mode of the wind turbine generator.

6. A training system for a wind turbine generator fault assessment model, characterized in that, include: The acquisition module is used to acquire multiple sets of raw data of the wind turbine generator and corresponding raw fault conclusion data. The multiple sets of raw data are divided into a training set and a prediction set. One set of raw data includes raw vibration data and raw temperature data. The raw data in the training set and the prediction set are different sets of data. An extraction module is used to extract a corresponding first feature from multiple raw vibration data in the training set, extract a corresponding second feature from multiple raw temperature data in the training set, extract a corresponding third feature from multiple raw vibration data in the prediction set, and extract a corresponding fourth feature from multiple raw temperature data in the prediction set. The fusion module is used to input multiple first features and corresponding multiple second features into a preset model to be trained, and fuse them to obtain an original fusion index sequence; and to input multiple third features and corresponding multiple fourth features into the model to be trained, and fuse them to obtain a predicted fusion index sequence; the model to be trained is a deep autoencoder model. The calculation module is used to obtain multiple wind turbine generator state evaluation standard values ​​based on the original fusion index sequence, the predicted fusion index sequence and the preset mapping relationship. The determination module is used to calculate the contribution of various types of data in the predictive fusion index sequence based on multiple wind turbine generator state assessment standard values, and determine the fault type and corresponding fault mode of the wind turbine generator based on the obtained multiple contribution values. and The adjustment module is used to adjust the model parameters of the model to be trained based on the residual features in the original fusion index sequence and the predicted fusion index sequence, respectively, and to obtain a wind turbine generator fault assessment model based on the adjusted model parameters. The wind turbine generator fault assessment model is used to diagnose faults in the wind turbine generator to be assessed data. The process of obtaining multiple wind turbine generator state assessment standard values ​​based on the original fusion index sequence, the predicted fusion index sequence, and a preset mapping relationship includes: The original fusion index sequence is evaluated to obtain the generator state assessment threshold value; and Based on the preset mapping relationship between the predicted fusion index sequence and the generator condition assessment threshold value, multiple wind turbine generator condition assessment standard values ​​are calculated.

7. A computer device, the computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the wind turbine generator fault assessment model training method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that can be executed by at least one processor to cause the at least one processor to perform the steps of the wind turbine generator fault assessment model training method as described in any one of claims 1 to 5.

Images