Wind turbine generator health state intelligent diagnosis method and system

Abstract

The invention discloses a wind turbine generator health state intelligent diagnosis method and system, and belongs to the technical field of wind turbine generator operation and maintenance, and the method comprises the steps: dividing key parts of a generator set, collecting the operation data of the key parts, and obtaining the macroscopic operation data of a region where the generator set is located, and other related data of the generator set; the macroscopic operation data comprises wind speed and wind direction; the other related data comprises maintenance history and fault records; the operation data comprises vibration, temperature, pressure, current and power; preprocessing the collected operation data, wherein the preprocessing comprises the steps of cleaning the operation data, removing noise data and abnormal values and normalizing the operation data; through comprehensive data acquisition, data preprocessing, multi-dimensional feature extraction, deep learning model construction and fault type division, the diagnosis capability of the health state of the wind turbine generator is improved, the operation and maintenance cost is reduced, and the operation efficiency and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of operation and maintenance of wind turbines, and particularly to an intelligent diagnosis method and system for the health status of wind turbines. Background Art

[0002] With the rapid development of the wind power industry, the scale and quantity of wind turbines are continuously increasing. The health status of wind turbines is directly related to the power generation efficiency and operation cost of wind farms. Traditional diagnostic methods for the health status of wind turbines mainly rely on manual inspections and rule-based threshold judgments, which have problems such as low efficiency, poor accuracy, and difficulty in timely detecting potential faults.

[0003] Regarding the research in this aspect, the application document with the application number CN201811351945.2 provides an intelligent monitoring and diagnosis method for the fault status of wind turbines. This technical solution includes establishing a non-linear model of wind turbines using partial least squares method; constructing a fault prediction model by combining extreme learning machine, chaotic mapping, and firefly algorithm; establishing a DBN-ELM fault diagnosis model through deep belief learning combined with extreme learning machine; and monitoring the status of the unit by calculating the residuals between the non-linear mathematical model and the prediction model to determine whether the wind turbine has a fault, and starting the fault diagnosis model to diagnose and locate the fault. This technical solution reduces the complexity of fault monitoring and improves the diagnostic accuracy of faults.

[0004] Another application document with the application number CN202211477900.6 provides an intelligent diagnosis method for evaluating the operation status of wind turbines. This technical solution includes collecting the operation data of each wind turbine and performing data preprocessing, deeply analyzing its operation status, constructing a wind turbine status evaluation algorithm to evaluate each wind turbine, selecting representative wind turbines as benchmark units, analyzing the health degree of the sorted and summarized wind turbine evaluation results, formulating an inspection plan, eliminating defects in a timely manner, deeply mining the wind turbine evaluation results and inspection data, and judging whether there are faults and potential hazards in the wind turbines. This technical solution realizes fast, accurate, and safe diagnosis of the operation status of wind turbines, and reduces time and economic costs.

[0005] However, the above technical solutions mainly rely on the SCADA system and simple vibration monitoring devices. Among them, the status parameters monitored by the SCADA system are mainly signals such as current, voltage, and power, and there is less monitoring of the vibration, temperature, etc. of key components. It is difficult to comprehensively reflect the health status of the unit, effectively extract the characteristic information under complex fault modes, and has poor adaptability to the characteristic changes under different working conditions. For example, traditional methods based on single or several characteristic quantities are difficult to accurately reflect the true operation status of wind turbines, and are prone to insufficient recognition ability for early weak fault characteristics, resulting in untimely fault warnings and inability to provide effective guidance for preventive maintenance. Summary of the Invention

[0006] In view of the above problems existing in the existing wind turbine operation and maintenance technology field, the present invention is proposed.

[0007] Therefore, one object of the present invention is to provide an intelligent diagnosis method and system for the health status of wind turbines, which improves the diagnostic ability of the health status of wind turbines, reduces the operation and maintenance costs, and improves the operation efficiency and reliability through methods such as comprehensive data collection, data preprocessing, multi-dimensional feature extraction, deep learning model construction, fault type classification, statistic analysis, linear regression, and support vector machine regression.

[0008] To solve the above technical problems, the present invention provides the following technical solutions: On the one hand, the present invention provides an intelligent diagnosis method for the health status of wind turbines, including the following steps: dividing key components of the generating set, collecting operation data of the key components, and at the same time obtaining the macro operation data of the area where the generating set is located, as well as other relevant data of the generating set; the macro operation data includes wind speed and wind direction; the other relevant data includes maintenance history and fault records; the operation data includes vibration, temperature, pressure, current, and power; preprocessing the collected operation data, the preprocessing includes cleaning the operation data, removing noise data and outliers, and normalizing; sorting the operation data in a time series, analyzing the regular changes of the operation data collected in different time periods, and comparing the regular changes with the operation data corresponding to the fault records, marking the operation data corresponding to the fault records as reference data, and analyzing the difference between the collected operation data and the reference data; extracting features according to the regular changes, including obtaining time domain features, frequency domain features, and time-frequency domain features in the operation data through wavelet transform or time-frequency analysis, and fusing different features to construct a comprehensive feature vector, and at the same time generating a diagnostic model through a deep learning algorithm; classifying the fault records according to the reference data into fault types, including classifying into , where represents the th fault type divided; and training and validating the diagnostic model in combination with the divided fault types; wherein, in the diagnostic model, the operation data after the preprocessing and feature extraction is used as the input layer sample, and the divided fault types are used as the output layer samples; obtaining the lowest operation data corresponding to the fault records among the divided fault types, presetting a safety threshold according to the lowest operation data, and at the same time determining the health status of the generating set according to the safety threshold.

[0009] As a preferred embodiment of the present invention, the following steps are included: training and validating the diagnostic model, and the training and validation methods include generating a data set from the operation data corresponding to each divided fault type in the reference data, dividing the data set into a training set and a test set, using the training set to train the diagnostic model, and using the test set to validate the diagnostic model.

[0010] As a preferred embodiment of the present invention, the following steps are included: differentiating the data set, and the differentiation method includes differentiating according to the time series, including obtaining the duration of the operation data collection corresponding to each divided fault type in the data set, dividing the duration into two time-equal periods, dividing the operation data in the earlier period into the training set, and dividing the operation data in the later period into the test set.

[0011] As a preferred embodiment of the present invention, the following steps are included: calculating the correlation impact of the change in the operation data of different training sets on the operation data of the test set based on the data set differentiation method, including calculating according to the mean square error method, as shown below: ; where is the mean square error; in the formula, represents the actual value, represents the predicted value, represents the number of samples; the actual value is each operation data corresponding to the training set, and the predicted value is the prediction of the operation data in the test set based on each operation data corresponding to the training set.

[0012] As a preferred embodiment of the present invention, the following steps are included: calculating statistics according to each operation data corresponding to the training set, analyzing the characteristic changes of the statistics of each divided fault type, and the statistics include the mean value, variance, and standard deviation. Among them, the mean value is the mean value of the operation data in each calculated fault type, and at the same time, the differences in the fault types caused are analyzed according to the mean value; the variance and standard deviation are the variance and standard deviation of the operation data in each calculated fault type, and the dispersion degree and fluctuation of the operation data in each fault type are analyzed according to the variance and standard deviation.

[0013] As a preferred embodiment of the present invention, the following steps are included: based on the statistics, obtaining the mean values of pressure, current, and power in each fault type, and obtaining the vibration value and temperature value corresponding to the mean value, generating a fault reference combination with the vibration value and temperature value corresponding to each fault type. When the operation data collected from the generator set in the future period, if the vibration value and temperature value are the same as the vibration value and temperature value in any one of the fault reference combinations, it is determined that the generator set has a health fault; otherwise, it is not determined.

[0014] As a preferred embodiment of the present invention, in the fault reference combination, the influence change on the vibration value and temperature value is calculated according to the change of the average values of pressure, current and power, and the calculation is carried out by means of linear regression, as follows: ; where represents the dependent variable, and the dependent variable is the vibration value and temperature value; in the formula, represents the independent variable, and the independent variable is the average values of pressure, current and power; represents the error term, is the regression coefficient.

[0015] As a preferred embodiment of the present invention, it further includes calculation by means of support vector machine regression, as follows: ; where represents the dependent variable, and the dependent variable is the vibration value and temperature value; represents the independent variable, and the independent variable is the average values of pressure, current and power; is the prediction function, and the prediction function is the prediction function of the vector machine model, represents the error term; a risk threshold is preset according to the calculated influence change. When the influence change on the vibration and temperature of the generator set in the future period is the same as the risk threshold, it is determined that the generator set has a health fault; otherwise, it is not determined.

[0016] On the other hand, the present invention provides a system applied to an intelligent diagnosis method for the health state of a wind turbine unit as described above, including: a data acquisition module, configured to divide key components of the generator set and acquire the operation data of the key components, and at the same time obtain the macro operation data of the area where the generator set is located and other relevant data of the generator set; the macro operation data includes wind speed and wind direction; the other relevant data includes maintenance history and fault records; the operation data includes vibration, temperature, pressure, current and power; a data preprocessing module, the data preprocessing module responding to the operation data and being configured to preprocess the operation data, the preprocessing including cleaning the operation data, removing noise data and outliers, and normalizing; a data fusion processing module, the data fusion processing module including an analysis unit, a feature extraction unit, a division unit and a determination unit; the analysis unit is configured to sort the operation data in a time series, analyze the regular changes of the operation data collected in different time periods, compare the regular changes with the operation data corresponding to the fault records, mark the operation data corresponding to the fault records as reference data, and analyze the difference between the collected operation data and the reference data; the feature extraction unit is configured to extract features according to the regular changes, including obtaining time-domain features, frequency-domain features and time-frequency domain features in the operation data by means of wavelet transform or time-frequency analysis, fusing different features, constructing a comprehensive feature vector, and generating a diagnosis model through a deep learning algorithm; the division unit is configured to divide the fault types of the fault records according to the reference data, including dividing into , where represents the th fault type divided; and training and validating the diagnosis model in combination with the divided fault types; the determination unit is configured to obtain the lowest operation data corresponding to the fault record among the divided fault types, preset a safety threshold according to the lowest operation data, and determine the health state of the generator set according to the safety threshold.

[0017] Beneficial effects: By collecting the operation data of key components of the generator set, as well as the macroscopic operation data, maintenance history, fault records, etc., the comprehensive integration of multi-source data is realized, and thus the operation state of the generator set can be more comprehensively reflected. At the same time, methods such as wavelet transform or time-frequency analysis are used to extract the time-domain, frequency-domain, and time-frequency domain features of the operation data, and these features are fused to construct a comprehensive feature vector. This multi-dimensional feature extraction method can capture fault features more accurately and improve the accuracy of diagnosis; the fault types of the fault records are classified based on reference data, and the diagnostic model is trained and verified in combination with the classified fault types, which improves the model's recognition ability for different fault types. And a safety threshold is preset according to the lowest operation data, which can timely detect potential faults, realize real-time monitoring and early warning of the health state of the generator set, reduce the fault shutdown time, and improve the operation reliability; by calculating the average value, variance, and standard deviation of the operation data in each fault type, the differences between fault types and the dispersion degree of the data are analyzed, providing richer statistical information for fault diagnosis. At the same time, the influence changes of the average values of pressure, current, and power on the vibration value and temperature value are calculated by methods such as linear regression and support vector machine regression, and a risk threshold is preset to realize early warning and risk assessment of health faults; by generating a fault reference combination, when the operation data collected in the future matches the data in the fault reference combination, it can quickly determine that the generator set has a health fault, providing timely and accurate decision-making support for the operation and maintenance personnel. And by real-time monitoring and dynamically updating the model, it can adapt to the operation changes of the generator set, maintain the accuracy and effectiveness of the diagnostic model, reduce the operation and maintenance cost, and improve the operation and maintenance efficiency. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them: Figure 1 It is a modular structure diagram of the intelligent diagnosis system for the health state of the wind turbine generator set according to the embodiment of the present invention; Figure 2 It is a schematic flowchart of the method according to the embodiment of the present invention.

[0019] Reference numerals in the drawings: 110 - data acquisition module; 120 - data preprocessing module; 130 - data fusion processing module; 1301 - analysis unit; 1302 - feature extraction unit; 1303 - classification unit; 1304 - determination unit. Detailed Embodiments

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the embodiments of the present invention in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention fall within the scope of protection of the present invention.

[0021] Since the existing technology mainly relies on SCADA systems and simple vibration monitoring devices, there is less monitoring of the vibration, temperature, etc. of key components, making it difficult to comprehensively reflect the health status of the unit, difficult to effectively extract the characteristic information under complex fault modes, and having poor adaptability to the characteristic changes under different working conditions, which easily leads to insufficient ability to identify the early weak fault characteristics, resulting in untimely fault warnings and inability to provide effective guidance for preventive maintenance.

[0022] Based on this, the present invention proposes an intelligent diagnosis method and system for the health status of wind turbines, which improves the health status diagnosis ability of wind turbines, reduces the operation and maintenance costs, and improves the operation efficiency and reliability through comprehensive data acquisition, feature extraction, accurate fault diagnosis, and data-driven model optimization.

[0023] The following further specifically describes this solution through embodiments in conjunction with the accompanying drawings.

[0024] Refer to Figures 1 to 2, which is an embodiment of the present invention. This embodiment provides an intelligent diagnosis method for the health status of a wind turbine generator set, including the following steps: Step S10: Divide the key components of the generator set, collect the operation data of the key components, and at the same time obtain the macro operation data of the area where the generator set is located, as well as other relevant data of the generator set; the macro operation data includes wind speed and wind direction; other relevant data includes maintenance history and fault records; the operation data includes vibration, temperature, pressure, current, and power; in this embodiment, the key components include the blades, generator, gearbox, pitch system, yaw system, etc. of the generator set, and by installing a variety of sensors, including vibration sensors, temperature sensors, pressure sensors, current sensors, power sensors, etc., to collect the operation data of the wind turbine generator set in real time; it should be emphasized that by collecting the operation data (such as vibration, temperature, pressure, current, power) of the key components of the generator set, as well as the macro operation data (such as wind speed, wind direction) and maintenance history, fault records, etc., the comprehensive collection of multi-source data is realized. This multi-dimensional data collection method can more comprehensively reflect the operation status of the generator set and provide a rich data basis for subsequent analysis and modeling; by combining different types of operation data with macro operation data and maintenance history, the health status of the generator set can be more comprehensively reflected, improving the accuracy and reliability of fault diagnosis; rich data sources help the model learn more types of fault patterns and enhance the generalization ability of the model; Step S20: Preprocess the collected operation data, and the preprocessing includes cleaning the operation data, removing noise data and outliers, and normalizing; in this embodiment, the quality and usability of the data are improved, providing a reliable data basis for subsequent analysis and modeling; high-quality data can improve the training effect of the model, reduce the risk of model overfitting, and improve the generalization ability of the model; by removing noise and outliers, the true operation status of the generator set can be more accurately reflected, enhancing the reliability of the diagnosis result; Step S30: Sort the operation data in time series, analyze the regular changes in the operation data collected in different time periods, compare the regular changes with the operation data corresponding to the fault records, mark the operation data corresponding to the fault records as reference data, and analyze the difference between the collected operation data and the reference data; in this embodiment, by sorting in time series and analyzing the regular changes in the operation data collected in different time periods, the changing trend of the data over time can be captured, providing information in the time dimension for fault diagnosis; by marking the operation data corresponding to the fault records as reference data and analyzing the difference between the operation data and the reference data, the fault characteristics can be more accurately identified, improving the accuracy of the diagnosis; Step S40: Extract features according to the regular changes, including obtaining the time-domain features, frequency-domain features, and time-frequency domain features in the operation data through wavelet transform or time-frequency analysis methods, fusing different features to construct a comprehensive feature vector, and at the same time generating a diagnosis model through a deep learning algorithm;In this embodiment, the time-domain features include mean, variance, and kurtosis; the frequency-domain features include spectral peak and power spectral density; the time-frequency domain features include wavelet energy spectrum; the deep learning algorithms include convolutional neural network (CNN), recurrent neural network (RNN) and its variants long short-term memory (LSTM) and gated recurrent unit (GRU); the time-domain, frequency-domain and time-frequency domain features of the operation data are extracted by methods such as wavelet transform or time-frequency analysis, and these features are fused to construct a comprehensive feature vector. This multi-dimensional feature extraction method can capture fault features more accurately and improve the accuracy of diagnosis; the diagnostic model is generated by deep learning algorithms, which can automatically learn the complex patterns and relationships in the data and has higher diagnostic accuracy and efficiency than traditional methods; Step S50: Divide the fault records into fault types based on the reference data, including dividing into; , where represents the th fault type divided; and train and validate the diagnostic model in combination with the divided fault types; among them, in the diagnostic model, the operation data after preprocessing and feature extraction is used as the input layer sample, and the divided fault types are used as the output layer samples; in this embodiment, training and validating the diagnostic model generated based on the operation data is a key step to ensure the accuracy and generalization ability of the model; dividing the fault records into fault types based on the reference data and training and validating the diagnostic model in combination with the divided fault types improve the model's ability to identify different fault types; Step S60: Obtain the lowest operation data corresponding to the fault record among the divided fault types, preset a safety threshold according to the lowest operation data, and at the same time determine the health status of the generator set according to the safety threshold; in this embodiment, the determination method includes if the operation data collected for the generator set in the future period is the same as the safety threshold, it is determined that the generator set has a health fault, and if the operation data is lower than the safety threshold, it is not determined that the generator set has a health fault.

[0025] In step S50, the diagnostic model is trained and validated. The training and validation methods include generating a data set from the operation data corresponding to each divided fault type in the reference data, dividing the data set into a training set and a test set, using the training set to train the diagnostic model, and using the test set to validate the diagnostic model. On this basis, the data set is divided. The division method includes dividing according to the time series, including obtaining the duration of the operation data collection corresponding to each divided fault type in the data set, dividing the duration into two time-equal periods, dividing the operation data in the earlier period into the training set, and dividing the operation data in the later period into the test set.

[0026] Furthermore, based on the method of dividing the data set, calculate the correlation impact of the change in the operation data of different training sets on the operation data of the test set, including calculating according to the mean square error method, as follows: ; where is the mean squared error; in the formula, represents the actual value, represents the predicted value, represents the number of samples; the actual value is each operating data corresponding to the training set, and the predicted value is the prediction of the operating data in the test set based on each operating data corresponding to the training set. In this embodiment, it should be noted that statistical quantities are calculated based on each operating data corresponding to the training set, and the characteristic changes of the statistical quantities of each divided fault type are analyzed. The statistical quantities include the mean, variance, and standard deviation. Among them, the mean is the mean of the operating data in each calculated fault type, and at the same time, the differences in the fault types caused are analyzed based on the mean; the variance and standard deviation are the variance and standard deviation of the operating data in each calculated fault type, and the dispersion degree and fluctuation of the operating data in each fault type are analyzed based on the variance and standard deviation; on the above basis, based on the statistical quantities, the mean values of pressure, current, and power are obtained in each fault type, and the vibration values and temperature values corresponding to the mean values are obtained. The vibration values and temperature values corresponding to each fault type are generated into a fault reference combination. When in the operating data collected from the generator set in the future period, if the vibration value and temperature value are the same as the vibration value and temperature value in any one of the fault reference combinations, it is determined that the generator set has a health fault; otherwise, it is not determined; in this embodiment, the mean, variance, and standard deviation of the operating data in each fault type are calculated, and the differences in the fault types and the dispersion degree of the data are analyzed, providing richer statistical information for fault diagnosis; and generating a fault reference combination, when the operating data collected in the future matches the data in the fault reference combination, it can quickly determine that the generator set has a health fault, providing timely and accurate decision support for the operation and maintenance personnel.

[0027] Further, in the fault reference combination, the influence changes of the vibration value and temperature value are calculated according to the changes in the mean values of pressure, current, and power, and the calculation is carried out in the way of linear regression, as shown below: ; where represents the dependent variable, and the dependent variable is the vibration value and temperature value; in the formula, represents the independent variable, and the independent variable is the mean values of pressure, current, and power; represents the error term, is the regression coefficient; in this embodiment, by estimating the regression coefficient, the influence degree of the independent variable on the dependent variable can be obtained; through the linear regression model, the influence changes of the mean values of pressure, current, and power on the vibration value and temperature value are analyzed, and a simple linear relationship model can be established to quickly evaluate the relationship between the parameters.

[0028] It should be emphasized in this embodiment that the calculation is also carried out in the way of support vector machine regression, as shown below: ; wherein, represents the dependent variable, which is the vibration value and the temperature value; represents the independent variable, which is the average value of pressure, current, and power; is the prediction function, which is the prediction function of the support vector machine model, represents the error term; according to the calculated impact change, a risk threshold is preset. When the impact change on the vibration and temperature of the generator set in the future period is the same as the risk threshold, it is determined that the generator set has a health fault; otherwise, it is not determined. In this embodiment, Support Vector Machine Regression (SVR) can obtain the non-linear impact of the independent variable on the dependent variable by training the SVR model; by calculating the impact change through the support vector machine regression model and presetting the risk threshold, early warning and risk assessment of health faults can be achieved, a complex non-linear relationship model can be established, the relationship between various parameters can be evaluated more accurately, and the fault diagnosis ability can be enhanced.

[0029] Based on the above, this application improves the health status diagnosis ability of the wind turbine generator set, reduces the operation and maintenance cost, and improves the operation efficiency and reliability through comprehensive data collection, feature extraction, accurate fault diagnosis, and data-driven model optimization.

[0030] Based on the above intelligent diagnosis method for the health status of wind turbines, this embodiment also proposes a working system applied to this method, as follows: A data acquisition module 110 is used to divide the key components of the generator set, collect the operation data of the key components, and at the same time obtain the macroscopic operation data of the region where the generator set is located, as well as other relevant data of the generator set; the macroscopic operation data includes wind speed and wind direction; other relevant data includes maintenance history and fault records; the operation data includes vibration, temperature, pressure, current, and power; A data preprocessing module 120, which responds to the operation data and is used to preprocess the operation data. The preprocessing includes cleaning the operation data, removing noise data and outliers, and normalizing; A data fusion processing module 130, which includes an analysis unit 1301, a feature extraction unit 1302, a division unit 1303, and a determination unit 1304; The analysis unit 1301 is used to sort the operation data in a time series, analyze the regular changes of the operation data collected in different time periods, compare the regular changes with the operation data corresponding to the fault records, mark the operation data corresponding to the fault records as reference data, and analyze the difference between the collected operation data and the reference data; The feature extraction unit 1302 is used to extract features according to the regular changes, including obtaining time-domain features, frequency-domain features, and time-frequency domain features in the operation data through wavelet transform or time-frequency analysis, fusing different features, constructing a comprehensive feature vector, and generating a diagnostic model through a deep learning algorithm; The division unit 1303 is used to divide the fault types of the fault records according to the reference data, including dividing into , where represents the th fault type divided; and training and validating the diagnostic model in combination with the divided fault types; The determination unit 1304 is used to obtain the lowest operation data corresponding to the fault record among the divided fault types, preset a safety threshold according to the lowest operation data, and at the same time determine the health status of the generator set according to the safety threshold.

[0031] In summary, through comprehensive data acquisition, data preprocessing, multi-dimensional feature extraction, deep learning model construction, fault type division, statistic analysis, linear regression, support vector machine regression and other methods, the present invention improves the diagnostic ability of the health status of wind turbines, reduces the operation and maintenance costs, and improves the operation efficiency and reliability.

[0032] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An intelligent diagnosis method for the health state of a wind turbine, characterized in that, Including the following steps: Dividing the key components of the generator set, collecting the operation data of the key components, and simultaneously obtaining the macro-operation data of the region where the generator set is located, as well as other relevant data of the generator set; The macroscopic operation data includes wind speed and wind direction; the other relevant data includes maintenance history and fault records; the operation data includes vibration, temperature, pressure, current, and power; the collected operation data is preprocessed, and the preprocessing includes cleaning the operation data, removing noise data and outliers, and normalizing; sorting the operation data in a time series, analyzing the regular changes in the operation data collected in different time periods, comparing the regular changes with the operation data corresponding to the fault records, marking the operation data corresponding to the fault records as reference data, and analyzing the difference between the collected operation data and the reference data; extracting features according to the regular changes, including obtaining time-domain features, frequency-domain features, and time-frequency domain features in the operation data through wavelet transform or time-frequency analysis, fusing different features, constructing a comprehensive feature vector, and generating a diagnostic model through a deep learning algorithm; dividing the fault types of the fault records based on the reference data, including dividing into , where represents the th fault type divided; and training and validating the diagnostic model in combination with the divided fault types; wherein, in the diagnostic model, the operation data after the preprocessing and feature extraction is used as the input layer sample, and the divided fault types are used as the output layer samples; obtaining the lowest operation data corresponding to the fault record among the divided fault types, presetting a safety threshold according to the lowest operation data, and determining the health status of the generator set according to the safety threshold.

2. The intelligent diagnosis method for the health state of a wind turbine unit according to claim 1, characterized in that Training and validating the diagnostic model, and the training and validation methods include generating datasets from the operation data corresponding to each divided fault type in the reference data, dividing the datasets into a training set and a test set, using the training set to train the diagnostic model, and using the test set to validate the diagnostic model.

3. The intelligent diagnosis method for the health state of a wind turbine unit according to claim 2, wherein, Differentiating the datasets, and the differentiation method includes differentiating according to the time series, including obtaining the duration of the operation data collection corresponding to each divided fault type in the datasets, dividing the duration into two time-equal periods, dividing the operation data in the earlier period into the training set, and dividing the operation data in the later period into the test set.

4. The intelligent diagnosis method for the health state of a wind turbine unit according to claim 3, wherein Based on the way of differentiating the dataset, calculate the associated impact of the changes in the running data of different training sets on the running data of the test set, including calculating according to the mean square error, as shown below: ; where is the mean square error; in the formula, represents the actual value, represents the predicted value, represents the number of samples; the actual value is each running data corresponding to the training set, and the predicted value is the prediction of the running data in the test set based on each running data corresponding to the training set.

5. The intelligent diagnosis method for the health state of a wind turbine unit according to claim 4, characterized in that, Calculating statistics according to each operation data corresponding to the training set, analyzing the characteristic changes of the statistics of each divided fault type, where the statistics include the mean, variance, and standard deviation. Among them, the mean is the mean of the operation data in each calculated fault type, and at the same time, analyzing the differences in the fault types caused by the mean; the variance and standard deviation are the variance and standard deviation of the operation data in each calculated fault type, and analyzing the dispersion degree and fluctuation of the operation data in each fault type according to the variance and standard deviation.

6. The intelligent diagnosis method for the health state of a wind turbine unit according to claim 5, wherein, Based on the statistics, obtaining the mean values of pressure, current, and power in each of the fault types, and obtaining the vibration values and temperature values corresponding to the mean values, generating a fault reference combination with the vibration values and temperature values corresponding to each fault type. When the operation data collected from the generator set in the future period, if the vibration value and temperature value are the same as the vibration value and temperature value in any one of the fault reference combinations, it is determined that the generator set has a health fault; otherwise, it is not determined.

7. The intelligent diagnosis method for the health state of a wind turbine unit according to claim 6, characterized in that, In the fault reference combination, calculate the influence change on the vibration value and temperature value according to the change of the average values of pressure, current and power, and calculate by means of linear regression as follows: ; where represents the dependent variable, and the dependent variable is the vibration value and temperature value; in the formula, represents the independent variable, and the independent variable is the average values of pressure, current and power; represents the error term, is the regression coefficient.

8. The intelligent diagnosis method for the health state of a wind turbine unit according to claim 7, wherein, It also includes calculation by means of support vector machine regression as follows: ; where represents the dependent variable, and the dependent variable is the vibration value and the temperature value; represents the independent variable, and the independent variable is the average value of pressure, current and power; is the prediction function, and the prediction function is the prediction function of the vector machine model, represents the error term; according to the preset risk threshold calculated for the influence change, when the influence change of the vibration and temperature of the generator set in the future period is the same as the risk threshold, it is determined that the generator set has a health fault; otherwise, it is not determined.

9. A system applied to the intelligent diagnosis method for the health state of a wind turbine unit as described in claim 1, characterized in that, Including: A data acquisition module for dividing the key components of the generator set, collecting the operation data of the key components, and simultaneously obtaining the macro-operation data of the region where the generator set is located, as well as other relevant data of the generator set; The macro-operation data includes wind speed and wind direction; the other relevant data includes maintenance history and fault records; the operation data includes vibration, temperature, pressure, current, and power; A data preprocessing module, which responds to the operation data and is used to preprocess the operation data. The preprocessing includes cleaning the operation data, removing noise data and outliers, and normalizing; Data fusion processing module, the data fusion processing module includes an analysis unit, a feature extraction unit, a division unit and a determination unit; the analysis unit is used to sort the operation data in time series, analyze the regular changes of the operation data collected in different time periods, and compare the regular changes with the operation data corresponding to the fault record, mark the operation data corresponding to the fault record as reference data, and analyze the difference between the collected operation data and the reference data; the feature extraction unit is used to extract features according to the regular changes, including obtaining time domain features, frequency domain features and time-frequency domain features in the operation data through wavelet transform or time-frequency analysis, fusing different features, constructing a comprehensive feature vector, and generating a diagnostic model through a deep learning algorithm; the division unit is used to divide the fault types of the fault record according to the reference data, including dividing into , where represents the th fault type divided; and training and verifying the diagnostic model in combination with the divided fault types; the determination unit is used to obtain the lowest operation data corresponding to the fault record among the divided fault types, preset a safety threshold according to the lowest operation data, and determine the health status of the generator set according to the safety threshold.

Images