A fault warning method and device for a pitch control system of an offshore wind turbine

Abstract

The present invention proposes a fault warning method and device for an offshore wind turbine pitch system. The method comprises constructing a health data set and a first fault data set; inserting a plurality of balance data into the first fault data set to obtain a second fault data set; performing a fault mechanism analysis on the offshore wind turbine pitch system and extracting fault characteristic values ​​corresponding to the fault mechanism to obtain a third fault data set; calculating the Mahalanobis distance between the data sets, constructing a health score sequence and a first sliding window, analyzing the abnormal values ​​of the health score sequence within the first sliding window, and obtaining a health assessment result of the offshore wind turbine pitch system; and providing a fault warning for deterioration of the operating state of the offshore wind turbine pitch system. The present invention analyzes the fault mechanism to obtain the probability of abnormal data occurrence, and then predicts and evaluates the health of the offshore wind turbine pitch system. By obtaining reliable health assessment results, the reliability of the fault warning of the offshore wind turbine pitch system is improved.

Description

Technical Field

[0001] The present invention relates to the field of health assessment of offshore wind turbines, and in particular to a fault early warning method and device for a pitch control system of an offshore wind turbine. Background Art

[0002] The offshore wind turbine pitch system consists of a motor drive, pitch motor, bearings, blades, limit switches, and redundant encoders. The offshore wind turbine pitch system will frequently adjust its actions according to different operating conditions. Frequent adjustments make the system load variable and more prone to failure. Therefore, it is necessary to assess the health status of the offshore wind turbine pitch system, predict potential risks, and issue timely warnings to ensure the safe operation of the offshore wind turbine pitch system.

[0003] At present, the traditional fault warning method for the offshore wind turbine pitch system usually uses the set values ​​in the wind turbine set setting table as the reference alarm line. The fault warning has a lag. When the alarm signal is issued, the fault has already occurred at a deeper level. The fault causes shutdown and affects the operational safety of the offshore wind turbine pitch system. When the fault warning model of the machine learning algorithm is used, it lacks analysis of the deep-seated mechanism of the fault and has weak interpretability. A single parameter is used to select the fault characteristics and fault type, which is not comprehensive and accurate when predicting the health level, resulting in poor reliability of the fault warning results of the offshore wind turbine pitch system. Summary of the Invention

[0004] In order to solve the above problems, the present invention proposes a fault warning method and device for the offshore wind turbine pitch system, which analyzes the fault mechanism to obtain the probability of abnormal data occurrence, and then predicts and evaluates the health of the offshore wind turbine pitch system. By obtaining reliable health assessment results, the fault warning reliability of the offshore wind turbine pitch system is improved.

[0005] To achieve the above objectives, an embodiment of the present invention provides a fault warning method for an offshore wind turbine pitch system, comprising: obtaining a number of historical offshore wind turbine pitch system SCADA data to construct a health data set and a first fault data set; generating a number of balance data based on a preset first algorithm, and inserting the number of balance data into the first fault data set to obtain a second fault data set; performing a fault mechanism analysis on the offshore wind turbine pitch system, and extracting fault feature values ​​corresponding to the fault mechanism from the second fault data set based on a preset second algorithm to obtain a third fault data set; calculating the Mahalanobis distance between the data sets based on the health data set and the third fault data set to obtain a fault Mahalanobis distance data set; obtaining a health score of the offshore wind turbine pitch system based on the fault Mahalanobis distance data set; constructing a health score sequence and a first sliding window based on a preset time interval and the health score of the offshore wind turbine pitch system, analyzing the abnormal values ​​of the health score sequence within the first sliding window to obtain a sequence abnormality rate; obtaining a health assessment result of the offshore wind turbine pitch system based on the sequence abnormality rate, and issuing a fault warning for deterioration of the operating state of the offshore wind turbine pitch system based on the health assessment result of the offshore wind turbine pitch system.

[0006] The embodiment of the present invention proposes a fault warning method for an offshore wind turbine pitch system. By dividing the historical offshore wind turbine pitch system SCADA data into a health data set and a first fault data set, and then inserting the balance data into the first fault data set, the problem of small amount of fault data can be supplemented, and a reliable data basis can be provided for subsequent data analysis and health assessment. Then, the fault mechanism of the offshore wind turbine pitch system is analyzed, and the fault characteristics corresponding to the fault mechanism are extracted. The cause of the fault, the process of the fault, the main form of the fault and the change law of the fault characteristics can be analyzed, thereby improving the interpretability of subsequent abnormal analysis and health assessment. The offshore wind turbine pitch system is analyzed by combining the health data set and calculating the Mahalanobis distance. The health score is analyzed, which can consider the correlation between different characteristics in the process of health assessment of offshore wind turbine pitch system, and provide reliable support for the health assessment of the system. Finally, the offshore wind turbine pitch system is analyzed for abnormalities and the health of the offshore wind turbine pitch system is predicted and evaluated by constructing a sliding window with a preset time interval. The health assessment results of the offshore wind turbine pitch system are obtained according to the sequence abnormality rate to assist in making fault warnings for the deterioration of the operating status of the offshore wind turbine pitch system. The probability of fault occurrence is predicted in advance through abnormal analysis and reliable health assessment results of the offshore wind turbine pitch system are obtained to make quick response decisions and avoid the occurrence of major faults, thereby improving the reliability of fault warnings of the offshore wind turbine pitch system.

[0007] Furthermore, the method of obtaining several historical offshore wind turbine pitch system SCADA data and constructing a healthy data set and a first fault data set includes: obtaining several historical offshore wind turbine pitch system SCADA data; based on the several historical offshore wind turbine pitch system SCADA data, eliminating all data points that are less than a preset cut-in wind speed and greater than a preset cut-out wind speed to obtain a first cleaned data set; calculating the mean and standard deviation of the first cleaned data set, eliminating all data points that meet preset criteria to obtain a second cleaned data set; and constructing a healthy data set and a first fault data set based on a preset time series and the second cleaned data set.

[0008] Through the above scheme, the historical SCADA data of the offshore wind turbine pitch control system is cleaned, and all data points with a wind speed less than the preset cut-in wind speed and greater than the preset cut-out wind speed when the system is in an inoperative state are eliminated. Then all data points that meet the preset criteria are eliminated, and the data is cleaned a second time. The data is divided into a healthy data set and a first fault data set, reducing the data redundancy caused by irrelevant data in data analysis, thereby improving the reliability of subsequent abnormal data analysis.

[0009] Furthermore, based on the preset first algorithm, a number of balanced data are generated, and the number of balanced data are inserted into the first fault data set to obtain a second fault data set, including: selecting a first sample point based on the first fault data set; finding the nearest neighbor sample point corresponding to the first sample point to obtain a number of second sample points corresponding to the first sample point; based on the number of second sample points corresponding to the first sample point, randomly selecting a second sample point as a third sample point, and subtracting it from the first sample point to obtain a sample difference value; connecting the first sample point and the corresponding third sample point, and randomly selecting a sample point on the connecting line based on the sample difference value to generate balanced data corresponding to the first sample point; repeating the above steps to obtain a number of balanced data, and inserting the number of balanced data into the first fault data set to obtain a second fault data set.

[0010] Through the above scheme, in order to address the problem of small amount of fault data, the fault samples are enhanced by inserting a balanced data set into the first fault data set, and synthetic samples of the minority class are generated to balance the data set, so as to obtain sufficient data volume to ensure the reliability of subsequent anomaly analysis, thereby improving the fault warning reliability of the offshore wind turbine variable pitch system.

[0011] Furthermore, the fault mechanism analysis of the offshore wind turbine pitch system is performed, and based on a preset second algorithm, the fault characteristic values ​​corresponding to the fault mechanism are extracted from the second fault data set to obtain a third fault data set, including: when a fault occurs in the offshore wind turbine pitch system, abnormal change parameters are monitored, and the fault mechanism is analyzed according to the abnormal change parameters to obtain fault characteristic points; based on the preset second algorithm, feature analysis is performed with the fault characteristic points as principal components to extract the fault characteristic values ​​corresponding to the fault mechanism to obtain the third fault data set.

[0012] Through the above scheme, when a fault occurs in the pitch control system of an offshore wind turbine, the abnormal change parameters are monitored, the fault mechanism corresponding to the abnormal change parameters is analyzed, the cause of the fault is analyzed in depth, and then the corresponding fault feature points after the fault mechanism analysis are extracted. Through principal component analysis, the data dimension is reduced while retaining the original data information, the core component information is retained, and the interpretability of subsequent abnormal analysis and health assessment is improved, thereby improving the fault warning reliability of the pitch control system of the offshore wind turbine.

[0013] Furthermore, based on the healthy data set and the third fault data set, the Mahalanobis distance between the data sets is calculated to obtain the fault Mahalanobis distance data set, including: calculating the difference between the third fault data set and the healthy data set to obtain the data set difference; and obtaining the fault Mahalanobis distance data set based on the inverse matrix of the covariance matrix and the data set difference.

[0014] Through the above scheme, the Mahalanobis distance between data sets is calculated to analyze the correlation between different features of the data sets, and the calculation weight of the highly correlated data sets is reduced by the inverse matrix of the covariance matrix, providing reliable support for the health assessment of the system. In addition, by calculating the difference between the data sets, it is also possible to effectively distinguish between healthy data and fault data to obtain reliable health assessment results of the offshore wind turbine pitch system, thereby improving the fault warning reliability of the offshore wind turbine pitch system.

[0015] Furthermore, based on the fault Mahalanobis distance dataset, a health score of the offshore wind turbine variable pitch system is obtained, including: calculating the dataset mean and standard deviation of the fault Mahalanobis distance dataset to obtain a normalized result; and based on the normalized result, obtaining the health score of the offshore wind turbine variable pitch system.

[0016] Through the above scheme, the fault Mahalanobis distance dataset is normalized to reduce the impact of maximum values ​​on the data normalization results, so as to obtain a more reliable health score of the offshore wind turbine pitch control system. By constructing the dimension of the health score, a reference standard is provided for the analysis of the fault degree of subsequent systems, thereby improving the fault warning reliability of the offshore wind turbine pitch control system.

[0017] Furthermore, the Mahalanobis distance between the fault-free operation data set of the offshore wind turbine variable pitch system in the non-limited power state and the healthy data set is calculated, and the health score corresponding to the maximum Mahalanobis distance is used as the health threshold; based on the preset time interval, a health score sequence and a first sliding window are constructed; based on the health threshold, outliers in the health score sequence are marked through the first sliding window to obtain an outlier sequence within the first sliding window; and based on the outlier sequence within the first sliding window, a sequence anomaly rate is calculated.

[0018] Through the above scheme, a health threshold is constructed according to the health score of the offshore wind turbine pitch system during healthy operation, the judgment standard of abnormal analysis is improved, and then a sliding window is constructed according to the preset time interval, the abnormal data in the window is marked, and the sequence anomaly rate in the sliding window is calculated. The sliding window technology is improved to reduce the false alarm rate of the data, improve the reliability of the health assessment results of the offshore wind turbine pitch system, and thus improve the fault warning reliability of the offshore wind turbine pitch system.

[0019] Furthermore, based on the outlier sequence in the first sliding window, the sequence anomaly rate is calculated, including: when the health score sequence in the first sliding window is greater than the health threshold, there is no outlier sequence and the sequence anomaly rate is 0; when the health score sequence in the first sliding window is less than the health threshold and the health score sequence in the first sliding window is restored to greater than the health threshold, there is an outlier sequence and the sequence anomaly rate is greater than 0 and less than 1; when the health score sequence in the first sliding window is less than the health threshold and the health score sequence in the first sliding window cannot be restored to greater than the health threshold, there is an outlier sequence and the sequence anomaly rate is 1. Furthermore, based on the sequence abnormality rate, a health assessment result of the offshore wind turbine pitch system is obtained to assist in issuing an early warning of the deterioration of the operating state of the offshore wind turbine pitch system, including: updating the preset time interval to construct a second sliding window; obtaining the number of times the sequence abnormality rate is 1 in the second sliding window; when the number of times the sequence abnormality rate is 1 in the second sliding window is less than or equal to the preset first threshold, the health assessment result of the offshore wind turbine pitch system is healthy; when the number of times the sequence abnormality rate is 1 in the second sliding window is greater than the preset first threshold and less than or equal to the preset second threshold, the health assessment result of the offshore wind turbine pitch system is a concern; when the number of times the sequence abnormality rate is 1 in the second sliding window is greater than or equal to the preset third threshold, the health assessment result of the offshore wind turbine pitch system is a serious abnormality; based on the health assessment result of the offshore wind turbine pitch system, an early warning of the deterioration of the operating state of the offshore wind turbine pitch system is issued.

[0020] Through the above scheme, by setting sliding windows with different time intervals and combining the health score threshold to analyze the sequence anomaly rate, it is possible to effectively distinguish between minor faults and major faults, and to conduct a health assessment of the offshore wind turbine pitch system based on the number of minor faults. The health status of the offshore wind turbine pitch system is divided into multiple levels, and the probability of fault occurrence is predicted in advance based on the sequence anomaly rate, thereby improving the reliability of the health assessment results of the offshore wind turbine pitch system, and assisting in making reliable fault warnings for the deterioration of the operating status of the offshore wind turbine pitch system.

[0021] An embodiment of the present invention also provides a fault warning device for an offshore wind turbine pitch system, comprising: a data set construction module, a data set balancing module, a fault mechanism analysis module, a Mahalanobis distance calculation module, a health score acquisition module, an outlier analysis module, a health assessment module and a fault warning module; the data set construction module is used to obtain a number of historical offshore wind turbine pitch system SCADA data, and construct a health data set and a first fault data set; the data set balancing module is used to generate a number of balancing data based on a preset first algorithm, and insert the number of balancing data into the first fault data set to obtain a second fault data set; the fault mechanism analysis module is used to perform fault mechanism analysis on the offshore wind turbine pitch system, and based on a preset second algorithm, extract the fault characteristic value corresponding to the fault mechanism in the second fault data set to obtain a third fault data set. data set; the Mahalanobis distance calculation module is used to calculate the Mahalanobis distance between the data sets based on the health data set and the third fault data set to obtain a fault Mahalanobis distance data set; the health score acquisition module is used to obtain the health score of the offshore wind turbine pitch system based on the fault Mahalanobis distance data set; the outlier analysis module is used to construct a health score sequence and a first sliding window based on a preset time interval and the health score of the offshore wind turbine pitch system, and analyze the outliers of the health score sequence in the first sliding window to obtain a sequence anomaly rate; the health assessment module is used to obtain a health assessment result of the offshore wind turbine pitch system based on the sequence anomaly rate; the fault warning module is used to make a fault warning of the deterioration of the operating state of the offshore wind turbine pitch system according to the health assessment result of the offshore wind turbine pitch system.

[0022] The embodiment of the present invention proposes a fault warning device for an offshore wind turbine pitch system, which divides the historical offshore wind turbine pitch system SCADA data into a health data set and a first fault data set through a data set construction module, and then inserts the balanced data into the first fault data set through a data set balancing module, which can make up for the problem of small amount of fault data and provide a reliable data basis for subsequent data analysis and health assessment. Then, the fault mechanism analysis module is used to analyze the fault mechanism of the offshore wind turbine pitch system, and the fault characteristics corresponding to the fault mechanism are extracted, which can analyze the cause of the fault, the process of the fault, the main form of the fault and the change law of the fault characteristics, thereby improving the interpretability of subsequent abnormal analysis and health assessment, and combining the health data set with the Mahalanobis distance calculation module and the health score acquisition module and using the method of calculating the Mahalanobis distance. By analyzing the health score of the offshore wind turbine pitch system, the correlation between different features can be considered in the process of health assessment of the offshore wind turbine pitch system, providing reliable support for the health assessment of the system. Finally, the outlier analysis module is used to construct a sliding window with a preset time interval to perform abnormal analysis on the offshore wind turbine pitch system and predict the health of the offshore wind turbine pitch system. The health assessment module obtains the health assessment result of the offshore wind turbine pitch system according to the sequence abnormality rate to assist the fault warning module in making fault warnings for the deterioration of the operating status of the offshore wind turbine pitch system. The abnormal analysis is used to predict the probability of fault occurrence in advance and obtain reliable health assessment results of the offshore wind turbine pitch system, so as to make quick response decisions and avoid the occurrence of major faults, thereby improving the reliability of fault warning of the offshore wind turbine pitch system. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of a fault warning method for an offshore wind turbine pitch control system according to an embodiment of the present invention Figure 1 ;

[0024] Figure 2 A schematic diagram of a process flow diagram of an offshore wind turbine pitch control system according to a fault warning method of an offshore wind turbine pitch control system provided by an embodiment of the present invention;

[0025] Figure 3 A schematic diagram of Mahalanobis distance comparison for a fault warning method for a pitch control system of an offshore wind turbine provided by one embodiment of the present invention;

[0026] Figure 4 A schematic diagram of a health score time series of a fault warning method for a pitch control system of an offshore wind turbine provided by an embodiment of the present invention;

[0027] Figure 5A schematic diagram of the health score time series of a pitch motor on April 13, according to a fault warning method for a pitch system of an offshore wind turbine provided by one embodiment of the present invention;

[0028] Figure 6 A statistical diagram of the abnormality rate of a pitch motor on April 13, according to a fault warning method for a pitch system of an offshore wind turbine provided by one embodiment of the present invention;

[0029] Figure 7 A schematic diagram of a fault warning method for an offshore wind turbine pitch control system according to an embodiment of the present invention Figure 2 ;

[0030] Figure 8 A schematic diagram of the module structure of a fault warning device for a pitch control system of an offshore wind turbine provided by one embodiment of the present invention;

[0031] Reference numerals:

[0032] 1. Limit switch; 2. Encoder; 3. Ring slide; 4. Pitch control system; 5. Main control system; 6. Pitch motor; 201. Data construction module; 202. Data balancing module; 203. Fault mechanism analysis module; 204. Mahalanobis distance calculation module; 205. Health score acquisition module; 206. Outlier analysis module; 207. Health assessment module; 208. Fault warning module. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0034] Example 1

[0035] See also Figure 1 , Figure 1 A schematic diagram of a fault warning method for an offshore wind turbine pitch control system according to an embodiment of the present invention Figure 1 .like Figure 1 As shown, the embodiment of the present invention provides a fault warning method for a pitch control system of an offshore wind turbine, including steps 101 to 108, each of which is specifically as follows:

[0036] Step 101, obtaining a number of historical offshore wind turbine pitch control system SCADA data, and constructing a healthy data set and a first fault data set;

[0037] As an example of this embodiment, several historical offshore wind turbine pitch control system SCADA data are obtained; based on the several historical offshore wind turbine pitch control system SCADA data, all data points that are less than a preset cut-in wind speed and greater than a preset cut-out wind speed are eliminated to obtain a first cleaned data set; the mean and standard deviation of the first cleaned data set are calculated, and all data points that meet preset criteria are eliminated to obtain a second cleaned data set; based on the preset time series and the second cleaned data set, a healthy data set and a first fault data set are constructed.

[0038] When offshore wind power is operating normally, it will flexibly adjust the pitch angle of its blades according to the current wind speed. Specifically, if the wind speed is lower than the minimum cut-in threshold, the wind turbine will automatically enter the standby state and the pitch system will not be activated. Once the wind speed reaches and maintains between the cut-in wind speed and the cut-out wind speed, the pitch system will immediately perform the corresponding blade angle adjustment operation according to the wind speed change. If the wind speed suddenly rises to a critical point exceeding the cut-out wind speed, the pitch system will immediately trigger the emergency pitching procedure to ensure that the blades are safely pitched. After that, the wind turbine will enter the shutdown mode. At the same time, the pitch system The system will also end its operating condition and be in an inoperative state. The existing fault warning method does not clean and eliminate the data in the inoperative state, resulting in redundant data processing and poor reliability. In order to solve the problems of the prior art, the embodiment of the present invention proposes a specific implementation method. First, data points with wind speeds below the cut-in wind speed and above the cut-out wind speed are eliminated. The pitch control system does not operate in this part of the data to obtain a first cleaned data set. Then, the remaining data samples are subjected to secondary data cleaning according to the Laida criterion (3δ criterion) to obtain a second cleaned data set. The specific method of secondary data cleaning is as follows:

[0039] Assume that the i-th dimension data of m data samples is:

[0040]

[0041] It can be simplified as:

[0042] x=(x1,x2,…x m )

[0043] If the sample data meets the following conditions (3δ criteria, equivalent to the preset criteria), it will be eliminated:

[0044] |x j -μ|≥3δ

[0045] In the above formula, μ is the mean of the first cleaned data set, and δ is the standard deviation of the first cleaned data set;

[0046] For the above filtered SCADA data, according to the power limiting mode and the fault information counted in the fault report, a healthy data set and a first fault data set are established according to the time series (which can be set to 10 minutes).

[0047] Step 102: Generate a plurality of balancing data based on a preset first algorithm, and insert the plurality of balancing data into the first fault data set to obtain a second fault data set;

[0048] As an example of this embodiment, based on the first fault data set, a first sample point is selected; the nearest neighbor sample point corresponding to the first sample point is found to obtain several second sample points corresponding to the first sample point; based on the several second sample points corresponding to the first sample point, a second sample point is randomly selected as a third sample point, and the second sample point is subtracted from the first sample point to obtain a sample difference value; the first sample point and the corresponding third sample point are connected by a line, and a sample point is randomly selected on the line based on the sample difference value to generate balanced data corresponding to the first sample point; the above steps are repeated to obtain several balanced data, and the several balanced data are inserted into the first fault data set to obtain a second fault data set.

[0049] Since the SCADA data obtained in step 101 is 10-minute data, the amount of fault data is small. In order to improve the accuracy of the model, it is necessary to enhance the data of the small fault samples to improve the classification performance between the unbalanced data. In order to deal with the problem of small amount of fault data, the fault samples are enhanced by inserting the balanced data set into the first fault data set, and synthetic samples of the minority class are generated to balance the data set, so as to obtain sufficient data volume to ensure the reliability of subsequent abnormal analysis, thereby improving the fault warning reliability of the offshore wind turbine pitch system. A specific implementation method is to use the SMOTE algorithm (Synthetic Minority Over-sampling Technique) to balance the data set by generating synthetic samples of the minority class. Its goal is to increase the classifier's learning ability for minority class samples, thereby improving the overall performance of the model. More specifically, the SMOTE algorithm logic is as follows: First, select a minority class sample point in the first fault data set: assume the minority class sample point X1; then find K nearest neighbors: use the K-nearest neighbor algorithm to find the K nearest neighbor sample points of X1, assuming X 11 , X 12 , X 13 , X 14 ; Then randomly select a neighbor and interpolate: randomly select a sample point from these K neighbors, assuming it is X 11 , calculate X1 and X 11The difference between X1 and X2 represents the distance between the two sample points in the feature space. Finally, a new sample is generated (equivalent to balanced data): 11 A point is randomly selected on the line between them. This point is the newly generated sample. The specific calculation formula is:

[0050] r1=X1+ran×diff

[0051] Among them, ran is a random number between [0, 1], representing the 11 Thus, the above steps are repeated to obtain a number of balanced data, and the number of balanced data are inserted into the first fault data set to obtain a second fault data set.

[0052] Step 103: performing a fault mechanism analysis on the offshore wind turbine pitch control system, extracting fault characteristic values ​​corresponding to the fault mechanism from the second fault data set based on a preset second algorithm, and obtaining a third fault data set;

[0053] As an example of this embodiment, when a fault occurs in the variable pitch system of an offshore wind turbine, abnormal change parameters are monitored, and the fault mechanism is analyzed based on the abnormal change parameters to obtain fault characteristic points; based on the preset second algorithm, feature analysis is performed with the fault characteristic points as the main components, and the fault characteristic values ​​corresponding to the fault mechanism are extracted to obtain a third fault data set.

[0054] After obtaining enough data through the SMOTE algorithm in step 102, combined with the analysis of the wind turbine pitch system operation mechanism, when a fault occurs in the offshore wind turbine pitch system, the abnormal change parameters are monitored, the fault mechanism corresponding to the abnormal change parameters is analyzed, the cause of the fault is analyzed in depth, and then the corresponding fault feature points after the fault mechanism analysis are extracted. For a specific implementation method, see Figure 2 , Figure 2 A schematic diagram of a process flow diagram of an offshore wind turbine pitch control system according to a fault warning method of an offshore wind turbine pitch control system provided by a certain embodiment of the present invention; Figure 2As shown, the pitch system of the offshore wind turbine includes: a limit switch 1, an encoder 2, a slip ring 3, a pitch control system 4, a main control system 5 and a pitch motor 6; in this embodiment, from the perspective of the signal transmission path, the main control system performs bidirectional signal interaction with the pitch control system through the slip ring. The pitch control system receives signals from the terminal data acquisition equipment such as the limit switch and encoder on the one hand, and outputs instructions to the pitch motor through the controller and receives signal feedback such as the temperature and current of the pitch motor on the other hand. Failure in any link of the data acquisition equipment (including the limit switch 1 and the encoder 2), the intermediate transmission path (including the slip ring 3), the control system (including the pitch control system 4 and the main control system 5), and the drive system (including the pitch motor 6) may cause a failure of the pitch system, thereby affecting the safe operation of the wind turbine. Therefore, a mechanism analysis is performed on the data acquisition equipment (including the limit switch 1 and the encoder 2), the intermediate transmission path (including the slip ring 3), the control system (including the pitch control system 4 and the main control system 5), and the drive system (including the pitch motor 6). Specifically:

[0055] (1) Data acquisition equipment failure: There are generally two situations when the data acquisition equipment fails. One is the failure of the equipment itself, which is mainly manifested in the jump, jitter or abnormal display of the collected data. The other is the data abnormality caused by external reasons, such as the failure of the mechanical transmission chain, the jamming of the pitch motor, the reduction box, the bearing or the blade encoder transmission gear, and the mechanical failure. The 91° limit switch will not be activated or the redundant 95° limit switch 1 will be triggered; or the redundant encoder 2 will have a difference greater than 5°. The abnormal collected data caused by external reasons can be further analyzed for abnormal conditions of the current, voltage, temperature and speed signals of the relevant equipment in the mechanical transmission chain;

[0056] (2) Intermediate transmission path failure: When the intermediate transmission path fails, it is mainly manifested as the signal transmission between the pitch control system and the main control system is lost, and it lasts for 600ms, then the fault is reported. The main reasons are that the slip ring connecting the pitch control system and the main control system is faulty, or the shielding layer and grounding of the communication cable are abnormal, or the wiring at both ends of the communication cable is loose, or the terminal resistance at both ends of the communication line is abnormal;

[0057] (3) Control system failure: There are two main reasons why the control system causes pitch system failure. First, the main control sets the blade angle frequently switching between clockwise and counterclockwise directions. The blade needs buffer time to switch from clockwise to counterclockwise movement. If the main control setting angle frequently switches the movement direction and there is no buffer time for the blade, the deviation between the actual angle and the main control setting angle will exceed the fault limit. Second, the main control setting position changes quickly. The maximum blade speed of the pitch system is only 6° / s. If the time to open the brake is normal, and the change speed of the main control setting position is 6° / s from the beginning of the action, the pitch motor will deviate too much from the main control setting angle after the brake is opened and accelerated, resulting in the triggering of the fault limit.

[0058] (4) Drive system failure: Drive system failure is the main cause of pitch system failure. The main failure forms include driver failure, pitch motor overload failure and backup power failure. Driver failure is when the driver fails and the pitch system fails to automatically reset, it will feed back the pitch driver failure status value to the main control. When this failure occurs, the pitch system will automatically move to a safe position and the blades will move to the limit switch position. Pitch motor overload failure is when the pitch motor is overloaded. The main manifestation is that the pitch motor current increases and the torque increases, which leads to a gradual increase in motor temperature and motor heat dissipation. The heater temperature gradually rises, and in severe cases, the pitch motor will be burned. When the pitch motor fails, the blades cannot move to the set position, and the blade angle and pitch speed are abnormal. At this time, the main control system will first report a pitch following fault, and further report a pitch motor overcurrent and pitch motor overtemperature fault; the backup power supply failures mainly include charger failure and backup power battery failure. When the charger fails, the charging bus voltage is normal, but the output voltage is abnormal. When the backup power battery fails, the main manifestations are battery capacity decline, battery voltage is too low, or battery temperature is abnormal. When the backup power voltage of the pitch system is lower than 270V, it is a Class I fault, which triggers the system to emergency feather.

[0059] Through the above fault mechanism analysis, the characteristic values ​​selected from the pitch system include blade angle, pitch speed, blade motor current, pitch motor temperature, battery box temperature, shaft control box temperature, hub temperature, drive heat dissipation temperature, motor torque, bus voltage, backup power supply voltage, active power and average wind speed. Taking the pitch motor failure as an example, according to the fault mechanism analysis, when the pitch motor fails, the blade angle, pitch speed, blade motor current, blade motor temperature, blade motor torque, radiator temperature and active power parameters will all change, while the changes with other parameters of the pitch system are not significant, so other input characteristic parameters can be deleted.

[0060] After analyzing the fault mechanism and extracting the corresponding fault feature points, principal component analysis is used to reduce the data dimension and retain the core component information while retaining the original data information, thereby improving the interpretability of subsequent abnormal analysis and health assessment, and thus improving the fault warning reliability of the offshore wind turbine pitch system. A specific implementation method is to use PCA (Principal Component Analysis) to extract fault feature values. For example, the blade angle is related to the pitch speed, and the motor torque is highly correlated with the motor current. Principal component analysis can further remove the above redundant information, reduce the data dimension, and construct the pitch system SCADA feature point table shown in Table 1:

[0061] Table 1 SCADA characteristic points of pitch control system

[0062]

[0063]

[0064] Through the principal component analysis method, the main features corresponding to the fault mechanism are extracted, irrelevant redundant information is removed, and the fault warning reliability of the offshore wind turbine pitch system is improved. The principal component analysis method is a relatively mature technology in the existing technology and will not be repeated here.

[0065] Step 104, based on the healthy data set and the third fault data set, calculating the Mahalanobis distance between the data sets to obtain a fault Mahalanobis distance data set;

[0066] As an example of this embodiment, the difference between the third fault data set and the healthy data set is calculated to obtain a data set difference; and the fault Mahalanobis distance data set is obtained based on the inverse matrix of the covariance matrix and the data set difference.

[0067] Mahalanobis distance can be used to measure the similarity between two sample sets. Taking a set of healthy data sets as a benchmark, the farther the Mahalanobis distance between the data in the sample and the healthy data set is, the worse the health is. By calculating the Mahalanobis distance between data sets, the correlation between different features of the data sets is analyzed, and then the calculation weight of the highly correlated data sets is reduced by the inverse matrix of the covariance matrix to provide reliable support for the health assessment of the system. At the same time, the Mahalanobis distance calculation is independent of scale and therefore has good adaptability. In addition, by calculating the difference between the data sets, it can also effectively distinguish between healthy data and fault data to obtain reliable health assessment results of the offshore wind turbine pitch system, thereby improving the fault warning reliability of the offshore wind turbine pitch system. A specific implementation method is as follows:

[0068]

[0069] Where: y, z represent two sample sets, S represents the covariance matrix, D m (y,z) represents the Mahalanobis distance between the two, and the value range of the Mahalanobis distance is [0, +∞);

[0070] The Mahalanobis distance between data sets is calculated by the above formula. In one explanation of the embodiment of the present invention, see Figure 3 , Figure 3 A schematic diagram of Mahalanobis distance comparison of a fault warning method for a pitch control system of an offshore wind turbine provided by an embodiment of the present invention; Figure 3 As shown in the figure, the normal data, fault data and healthy data of the pitch motor current are obtained, and the Mahalanobis distance between the normal data, fault data and healthy data of the pitch motor current is calculated. It can be seen from the figure that the Mahalanobis distance of the normal data is about 0.5, and the Mahalanobis distance of the fault data is between 1.5 and 3. The Mahalanobis distance can be used to effectively distinguish normal data from fault data.

[0071] Step 105: obtaining a health score of the offshore wind turbine pitch system based on the fault Mahalanobis distance dataset;

[0072] As an example of this embodiment, the dataset mean and standard deviation of the fault Mahalanobis distance dataset are calculated to obtain a normalized result; and based on the normalized result, a health score of the offshore wind turbine pitch system is obtained.

[0073] Normalize the fault Mahalanobis distance dataset to reduce the impact of the maximum value on the data normalization result, so as to obtain a more reliable health score of the offshore wind turbine pitch system. By constructing the dimension of the health score, a reference standard is provided for the analysis of the fault degree of the subsequent system, thereby improving the fault warning reliability of the offshore wind turbine pitch system. A specific implementation method is that since the Z-score method does not depend on the upper and lower limits of the data, the Z-score method (standard deviation normalization method) is used to normalize the Mahalanobis distance to a health score between [0, 1], which can greatly reduce the impact of the maximum value on the normalization result. Specifically, the real-time health score RHI of the offshore wind turbine pitch system can be expressed as:

[0074]

[0075] RHI=1-Z

[0076] Where: X represents the original data, μ ′represents the mean of the fault Mahalanobis distance data set, σ represents the standard deviation of the fault Mahalanobis distance data set, Z represents the normalized result, RHI represents the health score, and the value of RHI is between [0, 1]. The closer the RHI value is to 1, the healthier the variable pitch system is, and the closer it is to 0, the more potential faults exist in the variable pitch system.

[0077] Step 106: constructing a health score sequence and a first sliding window based on a preset time interval and the health score of the offshore wind turbine pitch control system, analyzing abnormal values ​​of the health score sequence within the first sliding window, and obtaining a sequence abnormality rate;

[0078] As an example of this embodiment, the Mahalanobis distance between the fault-free operating dataset of the offshore wind turbine pitch system in a non-limited power state and the healthy dataset is calculated, and the health score corresponding to the maximum Mahalanobis distance is used as the health threshold; based on the preset time interval, a health score sequence and a first sliding window are constructed; based on the health threshold, outliers in the health score sequence are marked using the first sliding window to obtain an outlier sequence within the first sliding window; and a sequence anomaly rate is calculated based on the outlier sequence within the first sliding window. More specifically, when the health score sequence within the first sliding window is greater than the health threshold, there is no outlier sequence and the sequence anomaly rate is 0; when the health score sequence within the first sliding window is less than the health threshold and the health score sequence recovers to a value greater than the health threshold within the first sliding window, there is an outlier sequence and the sequence anomaly rate is greater than 0 and less than 1; when the health score sequence within the first sliding window is less than the health threshold and the health score sequence cannot recover to a value greater than the health threshold within the first sliding window, there is an outlier sequence and the sequence anomaly rate is 1.

[0079] Based on the health score of the offshore wind turbine pitch system during healthy operation, a health threshold is established to improve the judgment standard of abnormal analysis. Then, a sliding window is constructed according to the preset time interval, abnormal data within the window is marked, and the sequence abnormality rate within the sliding window is calculated. The sliding window technology is used to reduce the false alarm rate of the data, improve the reliability of the health assessment results of the offshore wind turbine pitch system, and thus improve the reliability of the fault warning of the offshore wind turbine pitch system. For a specific implementation method, see Figure 4 , Figure 4 A schematic diagram of a health score timing sequence of a fault warning method for a pitch control system of an offshore wind turbine provided by an embodiment of the present invention; Figure 4 As shown in Figure 1, the method for determining the health threshold is as follows: calculate the Mahalanobis distance between the normal working data set and the healthy benchmark data set under the non-limited power state, take the maximum value, and convert the maximum Mahalanobis distance into a health score, that is, to obtain the health threshold ξ. If the health score is lower than the threshold, there is a risk of failure. Figure 4When the health score of the pitch motor is lower than the threshold of 0.85 (dashed line in the figure), the pitch system will face the risk of failure shutdown. By comparing with the actual fault repair records, the timing data below the dotted line is consistent with the failure shutdown time.

[0080] After building the health threshold, the sliding window technology is further used to determine the abnormality rate. The specific principle is as follows: First, set the time interval T of the sliding window, the start time T1, and the end time T n , then in the time interval T, the health score RHI sequence of the pitch system is:

[0081] RHI=[s1,s2…s n ]

[0082] By summarizing historical fault data, we can determine the range of the pitch system health threshold ξ. When the pitch system health score within time T is lower than the health threshold, the abnormal value at that moment is recorded as 1, indicating that the pitch system parameters are abnormal; otherwise, the abnormal value at that moment is recorded as 0, indicating that the pitch system parameters are normal. The abnormal value sequence within time T is:

[0083] σ i =0,σ i ≥ξ;i=1~n

[0084] σ i =1,σ i <ξ;i=1~n

[0085] F=[σ1,σ2…σ n ]

[0086] The anomaly rate φ of the outlier sequence in the time window T is:

[0087]

[0088] In this embodiment, a small sliding window (equivalent to the first sliding window) can be set to distinguish between minor faults and major faults. Specifically, the time interval T of the small window sliding is set to 30 minutes. When there is no fault within 30 minutes (equivalent to the health score sequence in the first sliding window being greater than the health threshold), the abnormality rate is 0; when a minor fault occurs within 30 minutes (equivalent to the health score sequence in the first sliding window being less than the health threshold and the health score sequence in the first sliding window recovering to be greater than the health threshold), the abnormality rate will appear as a square wave less than 1 in the sliding window. When the fault cannot be reset or self-healed within 30 minutes (equivalent to the health score sequence in the first sliding window being less than the health threshold and the health score sequence in the first sliding window being unable to recover to be greater than the health threshold), the waveform will gradually rise to a value of 1. Therefore, minor faults and major faults can be effectively distinguished by small window sliding.

[0089] For an explanation of an experiment dated April 13, see Figure 5 and Figure 6 , Figure 5 A schematic diagram of the health score time series of a pitch motor on April 13, according to a fault warning method for a pitch system of an offshore wind turbine provided by one embodiment of the present invention; Figure 6 A statistical diagram of the abnormality rate of a pitch motor of an offshore wind turbine pitch system according to a fault warning method provided by an embodiment of the present invention on April 13; Figure 5 and Figure 6 As shown, the wind turbine SCADA data on April 13th was selected. Figure 5 The data in the chart shows that the pitch motor had two minor data anomalies at 7:40 and 11:30, which were quickly recovered after the system self-reset. At 17:40, the fault intensified until it completely stopped. Correspondingly, Figure 6 At 7:40 and 11:30, the health score of the pitch motor was lower than the health threshold of 0.85, and one minor fault occurred each time. At 17:40, the health score continued to decline. During the sliding time window, the abnormality rate continued to increase until it reached 1, and the unit shut down. This shows that the health score sequence and the abnormal value sequence can confirm each other.

[0090] Step 107: obtaining a health assessment result of the offshore wind turbine pitch control system based on the sequence abnormality rate;

[0091] As an example of this embodiment, the preset time interval is updated to construct a second sliding window;

[0092] Obtain the number of times the sequence anomaly rate is 1 within the second sliding window; when the number of times the sequence anomaly rate is 1 within the second sliding window is less than or equal to the preset first threshold, the health assessment result of the offshore wind turbine pitch system is healthy; when the number of times the sequence anomaly rate is 1 within the second sliding window is greater than the preset first threshold and less than or equal to the preset second threshold, the health assessment result of the offshore wind turbine pitch system is a concern; when the number of times the sequence anomaly rate is 1 within the second sliding window is greater than or equal to the preset third threshold, the health assessment result of the offshore wind turbine pitch system is a serious abnormality.

[0093] After analyzing the sequence anomaly rate by setting a small sliding window in combination with a health score threshold and effectively distinguishing between light faults and severe faults, the health assessment of the pitch system of an offshore wind turbine is then carried out based on the occurrence frequency of light faults. The health condition of the pitch system of an offshore wind turbine is divided into multiple levels, and the probability of a fault occurring is predicted in advance according to the sequence anomaly rate, improving the reliability of the health assessment result of the pitch system of an offshore wind turbine to assist in making a reliable fault warning for the deterioration of the operating state of the pitch system of an offshore wind turbine. A specific implementable method is to set the time interval T for large window sliding as 24h. When light faults occur multiple times, it indicates that the system is gradually deteriorating. By counting the number of light faults before the occurrence of severe faults within 24h, it can be used to give an early warning of the system's deterioration, reminding the operator to perform some preventive maintenance in advance to avoid the occurrence of major faults and reduce the wind power operation and maintenance costs. For example Figure 6 , within the large window period on April 13th, when the second light fault occurs, the operator is timely reminded to pay attention to the unit status, which can avoid the occurrence of subsequent severe faults. More specifically, the health status of the pitch system of an offshore wind turbine is divided into three types: healthy, attention, and severely abnormal. Different colors can also be set to remind the operator. For example, the description of the health status and the suggestions after warning are constructed as shown in Table 2

[0094] Table 2 Description of Health Status and Suggestions after Warning

[0095]

[0096] It can be seen from Table 2 that within the 24h sliding window, the number of times n that the health score of the pitch system of an offshore wind turbine is less than the health threshold: n ≤ 1 time, the pitch system of an offshore wind turbine is in normal operation, and normal cycle maintenance and experiments can be carried out. The health status of the pitch system of an offshore wind turbine is healthy, and the operator is reminded in green; within the 24h sliding window, the number of times n that the health score of the pitch system of an offshore wind turbine is less than the health threshold: 1 < n ≤ 2 times, the pitch system of an offshore wind turbine may have abnormal conditions, which have not yet affected the equipment operation, but there is an early degradation trend and attention is needed. The health status of the pitch system of an offshore wind turbine is attention, and the operator is reminded in orange; within the 24h sliding window, the number of times n that the health score of the pitch system of an offshore wind turbine is less than the health threshold: n ≥ 3 times, the pitch system of an offshore wind turbine may have major faults, and shutdown maintenance and experiments should be arranged as soon as possible. The health status of the pitch system of an offshore wind turbine is severely abnormal, and the operator is reminded in red

[0097] Step 108, according to the health assessment result of the pitch system of the offshore wind turbine, make a fault warning for the deterioration of the operating state of the pitch system of the offshore wind turbine

[0098] As an example of this embodiment, after performing data analysis, data processing, and health assessment on the offshore wind turbine pitch control system according to steps 101 to 107, a health assessment result is obtained. Based on the health assessment result, different warnings are issued and different maintenance suggestions are provided. For details, see Table 3:

[0099] Table 3 Pitch system maintenance recommendations

[0100]

[0101]

[0102] The embodiment of the present invention combines fault mechanism analysis, multi-parameter fault characteristic value screening, Mahalanobis distance calculation and machine learning algorithms to evaluate and predict the health status of the variable pitch system of an offshore wind turbine. At the same time, a sliding window algorithm is used to monitor the probability of minor faults such as sporadic or self-healing faults, reduce the false alarm rate of the early warning model, and provide corresponding health warnings and maintenance suggestions. By providing early warnings, the reliability of the equipment is improved and the occurrence of major faults is avoided.

[0103] As another example of this embodiment, see Figure 7 , Figure 7 A schematic diagram of a fault warning method for an offshore wind turbine pitch control system according to an embodiment of the present invention Figure 2 ;like Figure 7As shown, a fault warning method for a pitch control system of an offshore wind turbine proposed in an embodiment of the present invention can also construct and train a health assessment model based on the technical content proposed in steps 101 to 105. The health assessment model is used to collect historical data, clean the historical data, construct a health data set and a fault data set, extract fault features to calculate the Mahalanobis distance, perform normalization processing, and finally perform health assessment. After the health assessment model executes steps 101 to 105, it can output a real-time health score in combination with the online data after data cleaning, and then execute steps 106 to 108 to realize the window sliding analysis abnormality rate. In addition, the online data greater than the health threshold is updated to the historical data, and the model database is updated to further improve the adaptability of model training and the reliability of data output. To explain more specifically, first, we fully analyze the failure mechanism of the wind turbine pitch system to understand the manifestation of the failure and the changes in SCADA data during the failure; then, we use historical data to clean the data, establish health and fault data sets, and extract fault characteristic values ​​based on the mechanism analysis of the pitch system failure; calculate the Mahalanobis distance between the sample data and the health data, and use the normalization algorithm to convert it into a health score; establish a health assessment model, and then use the trained health assessment model to calculate the real-time health score of the online data of the pitch system, and analyze the abnormality rate through window sliding. If the health score is less than the health threshold, an alarm will be issued and maintenance suggestions will be given. Finally, the historical database will be continuously updated to obtain more health data and fault data sets, and the health assessment model will be continuously updated.

[0104] The embodiment of the present invention proposes a fault warning method for an offshore wind turbine pitch system. By dividing the historical offshore wind turbine pitch system SCADA data into a health data set and a first fault data set, and then inserting the balance data into the first fault data set, the problem of small amount of fault data can be supplemented, and a reliable data basis can be provided for subsequent data analysis and health assessment. Then, the fault mechanism of the offshore wind turbine pitch system is analyzed, and the fault characteristics corresponding to the fault mechanism are extracted. The cause of the fault, the process of the fault, the main form of the fault and the change law of the fault characteristics can be analyzed, thereby improving the interpretability of subsequent abnormal analysis and health assessment. The offshore wind turbine pitch system is analyzed by combining the health data set and calculating the Mahalanobis distance. The health score is analyzed, which can consider the correlation between different characteristics in the process of health assessment of offshore wind turbine pitch system, and provide reliable support for the health assessment of the system. Finally, the offshore wind turbine pitch system is analyzed for abnormalities and the health of the offshore wind turbine pitch system is predicted and evaluated by constructing a sliding window with a preset time interval. The health assessment results of the offshore wind turbine pitch system are obtained according to the sequence abnormality rate to assist in making fault warnings for the deterioration of the operating status of the offshore wind turbine pitch system. The probability of fault occurrence is predicted in advance through abnormal analysis and reliable health assessment results of the offshore wind turbine pitch system are obtained to make quick response decisions and avoid the occurrence of major faults, thereby improving the reliability of fault warnings of the offshore wind turbine pitch system.

[0105] Example 2

[0106] See also Figure 8 , Figure 8 A schematic diagram of the module structure of a fault warning device for an offshore wind turbine pitch system according to an embodiment of the present invention. Figure 8 As shown, an embodiment of the present invention provides a fault warning device for a pitch control system of an offshore wind turbine, comprising:

[0107] Data set construction module 201, data set balancing module 202, fault mechanism analysis module 203, Mahalanobis distance calculation module 204, health score acquisition module 205, outlier analysis module 206, health assessment module 207 and fault warning module 208; the data set construction module 201 is used to obtain a number of historical offshore wind turbine pitch system SCADA data, build a health data set and a first fault data set; the data set balancing module 202 is used to generate a number of balanced data based on a preset first algorithm, and insert the several balanced data into the first fault data set to obtain a second fault data set; the fault mechanism analysis module 203 is used to perform fault mechanism analysis on the offshore wind turbine pitch system, and based on a preset second algorithm, extract the fault feature value corresponding to the fault mechanism in the second fault data set to obtain a third fault data set; the Mahalanobis distance calculation module 204, health score acquisition module 205, outlier analysis module 206, health assessment module 207 and fault warning module 208; the data set construction module 201 is used to obtain a number of historical offshore wind turbine pitch system SCADA data, build a health data set and a first fault data set; the data set balancing module 202 is used to generate a number of balanced data based on a preset first algorithm, and insert the several balanced data into the first fault data set to obtain a second fault data set; the fault mechanism analysis module 203 is used to perform fault mechanism analysis on the offshore wind turbine pitch system, and extract the fault feature value corresponding to the fault mechanism in the second fault data set based on a preset second algorithm to obtain a third fault data set; The distance calculation module 204 is used to calculate the Mahalanobis distance between the data sets based on the health data set and the third fault data set to obtain the fault Mahalanobis distance data set; the health score acquisition module 205 is used to obtain the health score of the offshore wind turbine pitch system based on the fault Mahalanobis distance data set; the outlier analysis module 206 is used to construct a health score sequence and a first sliding window based on a preset time interval and the health score of the offshore wind turbine pitch system, and analyze the outliers of the health score sequence in the first sliding window to obtain a sequence anomaly rate; the health assessment module 207 is used to obtain a health assessment result of the offshore wind turbine pitch system based on the sequence anomaly rate; the fault warning module 208 is used to issue a fault warning for the deterioration of the operating state of the offshore wind turbine pitch system based on the health assessment result of the offshore wind turbine pitch system.

[0108] The embodiment of the present invention proposes a fault warning device for an offshore wind turbine pitch system, which divides the historical offshore wind turbine pitch system SCADA data into a health data set and a first fault data set through a data set construction module, and then inserts the balanced data into the first fault data set through a data set balancing module, which can make up for the problem of small amount of fault data and provide a reliable data basis for subsequent data analysis and health assessment. Then, the fault mechanism analysis module is used to analyze the fault mechanism of the offshore wind turbine pitch system, and the fault characteristics corresponding to the fault mechanism are extracted, which can analyze the cause of the fault, the process of the fault, the main form of the fault and the change law of the fault characteristics, thereby improving the interpretability of subsequent abnormal analysis and health assessment, and combining the health data set with the Mahalanobis distance calculation module and the health score acquisition module and using the method of calculating the Mahalanobis distance. By analyzing the health score of the offshore wind turbine pitch system, the correlation between different features can be considered in the process of health assessment of the offshore wind turbine pitch system, providing reliable support for the health assessment of the system. Finally, the outlier analysis module is used to construct a sliding window with a preset time interval to perform abnormal analysis on the offshore wind turbine pitch system and predict the health of the offshore wind turbine pitch system. The health assessment module obtains the health assessment result of the offshore wind turbine pitch system according to the sequence abnormality rate to assist the fault warning module in making fault warnings for the deterioration of the operating status of the offshore wind turbine pitch system. The abnormal analysis is used to predict the probability of fault occurrence in advance and obtain reliable health assessment results of the offshore wind turbine pitch system, so as to make quick response decisions and avoid the occurrence of major faults, thereby improving the reliability of fault warning of the offshore wind turbine pitch system.

[0109] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

[0110] In the description of this specification, the reference terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0111] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, features specified as "first" or "second" may explicitly or implicitly include at least one of such features. In the description of this application, "plurality" means two or more, unless otherwise specifically defined.

Claims

1. A fault warning method for a pitch control system of an offshore wind turbine, characterized in that: include: Acquiring a number of historical offshore wind turbine pitch system SCADA data and constructing a healthy data set and a first fault data set, including: acquiring a number of historical offshore wind turbine pitch system SCADA data; based on the number of historical offshore wind turbine pitch system SCADA data, eliminating all data points that are less than a preset cut-in wind speed and greater than a preset cut-out wind speed to obtain a first cleaned data set; calculating the mean and standard deviation of the first cleaned data set, eliminating all data points that meet a preset criterion, to obtain a second cleaned data set; constructing a healthy data set and a first fault data set based on a preset time series and the second cleaned data set; Based on a preset first algorithm, a plurality of balanced data are generated, and the plurality of balanced data are inserted into the first fault data set to obtain a second fault data set, including: selecting a first sample point based on the first fault data set; finding a nearest neighbor sample point corresponding to the first sample point to obtain a plurality of second sample points corresponding to the first sample point; randomly selecting a second sample point as a third sample point based on the plurality of second sample points corresponding to the first sample point, and performing a subtraction with the first sample point to obtain a sample difference value; connecting the first sample point and the corresponding third sample point by a line, and randomly selecting a sample point on the line based on the sample difference value to generate balanced data corresponding to the first sample point; repeating the above steps to obtain a plurality of balanced data, and inserting the plurality of balanced data into the first fault data set to obtain a second fault data set; Performing a fault mechanism analysis on the offshore wind turbine pitch control system, and extracting fault characteristic values ​​corresponding to the fault mechanism from the second fault data set based on a preset second algorithm to obtain a third fault data set, including: when a fault occurs in the offshore wind turbine pitch control system, monitoring abnormal change parameters, and analyzing the fault mechanism based on the abnormal change parameters to obtain fault characteristic points; performing feature analysis based on the preset second algorithm using the fault characteristic points as principal components to extract fault characteristic values ​​corresponding to the fault mechanism to obtain the third fault data set; Based on the healthy data set and the third fault data set, calculating the Mahalanobis distance between the data sets to obtain a fault Mahalanobis distance data set; Obtaining a health score of an offshore wind turbine pitch system based on the fault Mahalanobis distance dataset; Based on a preset time interval and the health score of the offshore wind turbine pitch system, a health score sequence and a first sliding window are constructed, and outliers in the health score sequence within the first sliding window are analyzed to obtain a sequence anomaly rate, including: calculating the Mahalanobis distance between a fault-free operation data set of the offshore wind turbine pitch system in a non-limited power state and the health data set, and using the health score corresponding to the maximum Mahalanobis distance as a health threshold; based on the preset time interval, a health score sequence and a first sliding window are constructed; based on the health threshold, outliers in the health score sequence are marked through the first sliding window to obtain an outlier sequence within the first sliding window; and based on the outlier sequence within the first sliding window, a sequence anomaly rate is calculated; Based on the sequence abnormality rate, obtaining a health assessment result of the offshore wind turbine pitch system; According to the health assessment result of the offshore wind turbine pitch system, a fault warning is issued when the operating state of the offshore wind turbine pitch system deteriorates.

2. A fault warning method for a pitch control system of an offshore wind turbine according to claim 1, characterized in that: Calculating the Mahalanobis distance between the data sets based on the healthy data set and the third fault data set to obtain a fault Mahalanobis distance data set includes: Calculating a difference between the third fault data set and the healthy data set to obtain a data set difference; Based on the inverse matrix of the covariance matrix and the data set difference, a fault Mahalanobis distance data set is obtained.

3. The fault warning method for the pitch control system of an offshore wind turbine according to claim 1, characterized in that: Based on the fault Mahalanobis distance dataset, a health score of the offshore wind turbine pitch system is obtained, including: Calculating the data set mean and standard deviation of the fault Mahalanobis distance data set to obtain a normalized result; Based on the normalized result, a health score of the offshore wind turbine pitch system is obtained.

4. The fault warning method for a pitch control system of an offshore wind turbine according to claim 1, characterized in that: Calculating a sequence abnormality rate according to the abnormal value sequence in the first sliding window includes: When the health score sequence in the first sliding window is greater than the health threshold, there is no outlier sequence and the sequence anomaly rate is 0; When the health score sequence is less than the health threshold in the first sliding window and the health score sequence recovers to be greater than the health threshold in the first sliding window, there is an outlier sequence and the sequence anomaly rate is greater than 0 and less than 1; When the health score sequence is less than the health threshold in the first sliding window and the health score sequence cannot be restored to be greater than the health threshold in the first sliding window, there is an abnormal value sequence and the sequence abnormality rate is 1.

5. A fault warning method for a pitch control system of an offshore wind turbine according to claim 4, characterized in that: Based on the sequence abnormality rate, a health assessment result of the offshore wind turbine pitch system is obtained, including: Updating the preset time interval to construct a second sliding window; Obtain the number of times the sequence anomaly rate in the second sliding window is 1; When the number of times the sequence abnormality rate is 1 within the second sliding window is less than or equal to the preset first threshold, the health assessment result of the offshore wind turbine pitch system is healthy; When the number of times the sequence abnormality rate is 1 within the second sliding window is greater than the preset first threshold and less than or equal to the preset second threshold, the health assessment result of the offshore wind turbine pitch system is a concern; When the number of times the sequence abnormality rate is 1 within the second sliding window is greater than or equal to a preset third threshold, the health assessment result of the offshore wind turbine pitch control system is seriously abnormal.

6. A fault warning device for an offshore wind turbine pitch control system, characterized in that: include: Dataset construction module, data set balancing module, fault mechanism analysis module, Mahalanobis distance calculation module, health score acquisition module, outlier analysis module, health assessment module and fault warning module; The data set construction module is used to obtain a number of historical offshore wind turbine pitch system SCADA data to construct a healthy data set and a first fault data set, including: obtaining a number of historical offshore wind turbine pitch system SCADA data; based on the number of historical offshore wind turbine pitch system SCADA data, eliminating all data points that are less than a preset cut-in wind speed and greater than a preset cut-out wind speed to obtain a first cleaned data set; calculating the mean and standard deviation of the first cleaned data set, eliminating all data points that meet preset criteria, to obtain a second cleaned data set; based on a preset time series and the second cleaned data set, constructing a healthy data set and a first fault data set; The data set balancing module is used to generate a plurality of balanced data based on a preset first algorithm, and insert the plurality of balanced data into the first fault data set to obtain a second fault data set, including: selecting a first sample point based on the first fault data set; finding a nearest neighbor sample point corresponding to the first sample point to obtain a plurality of second sample points corresponding to the first sample point; randomly selecting a second sample point as a third sample point based on the plurality of second sample points corresponding to the first sample point, and performing a subtraction with the first sample point to obtain a sample difference value; connecting the first sample point and the corresponding third sample point by a line, and randomly selecting a sample point on the line based on the sample difference value to generate balanced data corresponding to the first sample point; repeating the above steps to obtain a plurality of balanced data, and inserting the plurality of balanced data into the first fault data set to obtain a second fault data set; The fault mechanism analysis module is used to perform a fault mechanism analysis on the offshore wind turbine pitch system, extracting fault characteristic values ​​corresponding to the fault mechanism from the second fault data set based on a preset second algorithm to obtain a third fault data set, including: when a fault occurs in the offshore wind turbine pitch system, monitoring abnormal change parameters, and analyzing the fault mechanism based on the abnormal change parameters to obtain fault characteristic points; based on the preset second algorithm, performing feature analysis with the fault characteristic points as principal components, extracting fault characteristic values ​​corresponding to the fault mechanism, and obtaining the third fault data set; The Mahalanobis distance calculation module is used to calculate the Mahalanobis distance between the data sets based on the healthy data set and the third fault data set to obtain a fault Mahalanobis distance data set; The health score acquisition module is used to acquire the health score of the offshore wind turbine pitch system based on the fault Mahalanobis distance data set; The outlier analysis module is used to construct a health score sequence and a first sliding window based on a preset time interval and the health score of the offshore wind turbine pitch system, analyze the outliers of the health score sequence in the first sliding window, and obtain a sequence anomaly rate, including: calculating the Mahalanobis distance between the fault-free operation data set of the offshore wind turbine pitch system in a non-limited power state and the health data set, and using the health score corresponding to the maximum Mahalanobis distance as a health threshold; constructing a health score sequence and a first sliding window based on the preset time interval; marking the outliers in the health score sequence through the first sliding window based on the health threshold to obtain an outlier value sequence in the first sliding window; and calculating the sequence anomaly rate based on the outlier value sequence in the first sliding window; The health assessment module is used to obtain a health assessment result of the offshore wind turbine pitch system based on the sequence abnormality rate; The fault warning module is used to issue a fault warning of deterioration of the operating state of the offshore wind turbine pitch system according to the health assessment result of the offshore wind turbine pitch system.

Images