Fault early warning method and system for gear box of offshore wind turbine generator

Abstract

The invention provides a fault early warning method and system for an offshore wind turbine generator gearbox, and the method comprises the steps: carrying out the fault mechanism analysis of the offshore wind turbine generator gearbox based on the historical operation data of the offshore wind turbine generator gearbox, and obtaining a fault monitoring index; based on a preset first algorithm and the fault monitoring index, obtaining an oversampling fault sample; obtaining a key fault feature index based on a preset second algorithm and the oversampling fault sample; based on a preset third algorithm, calculating a weighted mahalanobis distance value of the key fault feature index to obtain an actual health score of the offshore wind turbine generator gearbox; and based on the actual health score of the gear box of the offshore wind turbine generator, evaluating the health state of the gear box of the offshore wind turbine generator, and sending out a corresponding fault early warning. According to the method, the fault mechanism of the gear box of the offshore wind turbine generator is analyzed, and the health state of the gear box of the offshore wind turbine generator is evaluated, so that the reliability of fault early warning is improved.

Description

Technical Field

[0001] The present invention relates to the field of health assessment of offshore wind turbines, and particularly to a fault warning method and system for a gearbox of an offshore wind turbine. Background Art

[0002] With the increasing requirements for the reliability of offshore wind turbines, the accuracy requirements for fault prediction of offshore wind turbines are also getting higher and higher. Especially for the gearbox of an offshore wind turbine, the gearbox of an offshore wind turbine is an important transmission component, and its working performance directly affects the performance of the offshore wind turbine.

[0003] Currently, the existing fault warning methods mainly use methods such as industrial endoscopes and vibration tests to regularly detect the state of the gearbox, and comprehensively analyze the health state of the offshore gearbox with reference to relevant standards. Usually, machine learning algorithms are used to evaluate the health state, that is, by calculating the failure degree score value and comparing the score value with a preset threshold for health assessment. However, the fault warning methods adopted by the existing technologies lack analysis of the fault mechanism and cannot accurately predict the fault type, resulting in the problem of unreliable fault warning. Summary of the Invention

[0004] In order to solve the above problems, the present invention proposes a fault warning method and system for a gearbox of an offshore wind turbine, which realizes the analysis of the fault mechanism of the gearbox of the offshore wind turbine and the assessment of the health state of the gearbox of the offshore wind turbine, so as to improve the reliability of fault warning.

[0005] To achieve the above object, an embodiment of the present invention provides a fault warning method for a gearbox of an offshore wind turbine, including: based on the historical operation data of the gearbox of the offshore wind turbine, analyzing the fault mechanism of the gearbox of the offshore wind turbine to obtain fault monitoring indicators; obtaining oversampled fault samples based on a preset first algorithm and the fault monitoring indicators; obtaining key fault feature indicators based on a preset second algorithm and the oversampled fault samples; calculating the weighted Mahalanobis distance value of the key fault feature indicators based on a preset third algorithm to obtain the actual health score of the gearbox of the offshore wind turbine; and evaluating the health state of the gearbox of the offshore wind turbine based on the actual health score of the gearbox of the offshore wind turbine, and issuing a corresponding fault warning.

[0006] An embodiment of the present invention proposes a fault warning method for a gearbox of an offshore wind turbine. Based on the historical operation data of the gearbox of the offshore wind turbine, the fault mechanism of the gearbox of the offshore wind turbine is analyzed, so that the main factors causing faults during the historical operation can be analyzed. From this, the fault monitoring indicators that need to be obtained are determined. Since the fault samples belong to the minority class samples in the gearbox of the offshore wind turbine and the number of samples is small, in order to make the subsequent fault warning results of the gearbox of the offshore wind turbine more reliable, it is necessary to obtain the corresponding oversampled fault samples through a preset first algorithm and the fault monitoring indicators to increase the number of fault samples. In addition, since the fault types of the gearbox of the offshore wind turbine are diverse and the importance degrees among the fault types are different, the key fault feature indicators are reasonably extracted through a preset second algorithm and the oversampled fault samples to reduce the data dimension and ensure the reliability of the subsequent fault warning of the gearbox of the offshore wind turbine. Then, the weighted Mahalanobis distance values of each key fault feature indicator are calculated through a preset third algorithm, and weights are introduced to evaluate the importance of each key fault indicator for the health assessment of the gearbox of the offshore wind turbine, so as to accurately obtain the health score of the offshore wind power gear set. Finally, the health status of the gearbox of the offshore wind turbine is evaluated through the health score of the offshore wind power gear set, and corresponding fault warnings are issued according to the corresponding health status. By analyzing the fault mechanism of the gearbox of the offshore wind turbine and evaluating the health status of the equipment, the reliability of the fault warning is improved.

[0007] Further, based on the historical operation data of the gearbox of the offshore wind turbine, the fault mechanism of the gearbox of the offshore wind turbine is analyzed to obtain the fault monitoring indicators, including: based on the historical operation data of the gearbox of the offshore wind turbine, the historical operation fault types are obtained; based on the historical operation fault types, the fault mechanism of the gearbox of the offshore wind turbine is analyzed to obtain the fault mechanism change indicators corresponding to the occurrence of the fault; the fault mechanism change indicators corresponding to the occurrence of the fault that meet the preset change trend requirements are screened to obtain the fault monitoring indicators.

[0008] Through the above solution, by analyzing the historical operation data, the corresponding fault types are obtained, the fault mechanism of the gearbox of the offshore wind turbine is analyzed, and the fault mechanism change indicators whose change trends meet the requirements during the occurrence of the fault are screened out and used as the fault monitoring indicators, and the parameters that change most significantly during the occurrence of the fault are initially analyzed and screened out to improve the reliability of the subsequent fault warning results.

[0009] Further, obtaining oversampled fault samples based on the preset first algorithm and the fault monitoring metrics includes: obtaining the historical monitoring dataset of the gearbox of the offshore wind turbine based on the fault monitoring metrics; dividing the historical monitoring dataset of the gearbox of the offshore wind turbine into fault samples and normal samples; calculating the number of normal samples at a preset first nearest neighbor distance from each of the fault samples based on the preset first algorithm, obtaining the distribution density of the fault samples and calculating the target synthetic sample number; randomly selecting the neighbor samples of each of the fault samples based on a preset second nearest neighbor distance, and performing linear interpolation on each of the fault samples through the neighbor samples and the target synthetic sample number to obtain oversampled fault samples.

[0010] Through the above steps, since the fault samples belong to the minority class samples, it is necessary to perform oversampling processing on the fault samples through the preset first algorithm and synthesize samples through linear interpolation to increase the number of fault samples, which is beneficial to the subsequent health status assessment of the gearbox of the offshore wind turbine, and further improves the reliability of the fault warning result.

[0011] Further, obtaining the key fault feature indicators based on the preset second algorithm and the oversampled fault samples includes: constructing a number of decision trees based on the oversampled fault samples; classifying the oversampled fault samples through a preset voting mechanism based on the preset second algorithm and the number of decision trees to obtain a number of sample categories; performing out-of-bag sample evaluation on the features of the samples corresponding to each of the sample categories, calculating the prediction errors of the samples corresponding to each of the sample categories and performing feature importance ranking, and screening out the key fault feature indicators.

[0012] Through the above steps, among a large number of fault types, there is an importance order for the features of each fault sample. Through the preset second algorithm, a voting mechanism is used to classify the oversampled fault samples, and then the feature importance of each sample is evaluated based on the out-of-bag sample evaluation, and the key fault feature indicators are screened out to reduce the data dimension, ensure the data processing volume in the subsequent health assessment of the gearbox of the offshore wind turbine, and improve the reliability of the fault prediction result by improving the accuracy of the health status assessment.

[0013] Further, before performing the step of calculating the weighted Mahalanobis distance value based on the preset third algorithm and the key fault feature indicators to obtain the health score of the gearbox of the offshore wind turbine, it further includes: constructing an original data matrix based on the key fault feature indicators; performing unit vector data standardization on the original data matrix to obtain a standardized matrix; calculating the proportion of each sample in the standardized matrix, and calculating the key fault feature information entropy according to the proportion of each sample; obtaining the weights of each key fault feature based on the key fault feature information entropy.

[0014] Through the above solution, in order to more objectively reflect the importance of key fault feature indicators and reduce the influence of subjective factors, the information entropy of each key fault feature indicator is calculated to provide reasonable weights for the subsequent calculation of the health score, so as to ensure the reliability of the fault warning result.

[0015] Further, based on the preset third algorithm, calculate the weighted Mahalanobis distance value of the key fault feature indicators to obtain the actual health score of the offshore wind turbine gearbox, including: calculating the Mahalanobis distance of each key fault feature indicator in the original data matrix based on the original data matrix; calculating the weighted Mahalanobis distance value through the preset third algorithm based on each key fault feature weight; obtaining the actual health score of the offshore wind turbine gearbox based on the weighted Mahalanobis distance value. Further, based on the actual health score of the offshore wind turbine gearbox, evaluate the health status of the offshore wind turbine gearbox and issue a corresponding fault warning, including: if the actual health score of the offshore wind turbine gearbox is greater than or equal to the preset first health threshold, evaluate the health status of the offshore wind turbine gearbox as healthy; if the actual health score of the offshore wind turbine gearbox is less than the preset first health threshold and greater than the preset second health threshold, evaluate the health status of the offshore wind turbine gearbox as having a mild abnormality; if the actual health score of the offshore wind turbine gearbox is less than or equal to the preset second health threshold, it indicates that the health status of the evaluated offshore wind turbine gearbox is severely abnormal; issue a corresponding fault warning based on the health status of each offshore wind turbine gearbox. Further, issuing a corresponding fault warning based on the health status of each offshore wind turbine gearbox, including: if the health status of the offshore wind turbine gearbox is healthy, no fault warning is issued; if the health status of the offshore wind turbine gearbox is mildly abnormal, a mild fault warning is issued; if the health status of the offshore wind turbine gearbox is severely abnormal, a severe fault warning is issued.

[0016] Through the above solution, since there is a correlation between the operating parameters of the offshore wind turbine gearbox, calculating the Mahalanobis distance can reasonably reflect the relationship between parameters. Then, the weight relationship obtained from the pre-calculated information entropy of each key fault feature indicator is combined into the Mahalanobis distance, introducing weights to emphasize the importance of each parameter in health assessment, and then obtaining a reliable health score to ensure the accuracy of health status assessment. Finally, corresponding fault warnings are issued according to the different health statuses of the offshore wind turbine gearbox, thereby improving the reliability of the fault warning result.

[0017] Further, after performing the steps of evaluating the health status of the offshore wind turbine gearbox based on the actual health score of the offshore wind turbine gearbox and issuing corresponding fault warnings, the method further includes: obtaining the historical operation data of the offshore wind turbine gearbox, and dividing the fault monitoring indicators of the historical operation data of the offshore wind turbine gearbox into a training set and a test set according to a preset ratio; constructing an initial health score prediction model, and optimizing the model parameters of the initial health score prediction model through a preset fourth algorithm according to the training set to obtain a target health score prediction model; inputting the test set into the target health score prediction model to obtain a health score prediction value; calculating an error index based on the health score prediction value and the actual health score of the offshore wind turbine gearbox; if the error index meets the preset requirements, predicting the health status of the offshore wind turbine gearbox in a target time period through the target health score prediction model, and issuing corresponding fault warnings according to the health status of each offshore wind turbine gearbox in the target time period.

[0018] Through the above solution, based on the historical operation data of the offshore wind turbine gearbox, a target health score prediction model is trained. By obtaining the health score prediction value and calculating the error index in combination with the actual health score value, if the error index meets the requirements, it indicates that the prediction accuracy of the target health score prediction model meets the requirements. The target health score prediction model can be used to predict the health status of the offshore wind turbine gearbox in the future target time period, so as to perform early warning analysis on faults, thereby improving the reliability of the fault warning results.

[0019] An embodiment of the present invention further provides a fault warning system for an offshore wind turbine gearbox, including: a fault mechanism analysis module, an oversampling processing module, a key feature index acquisition module, a health score acquisition module, and a warning module; the fault mechanism analysis module is used to perform fault mechanism analysis on the offshore wind turbine gearbox based on the historical operation data of the offshore wind turbine gearbox to obtain fault monitoring indicators; the oversampling processing module is used to obtain oversampled fault samples based on a preset first algorithm and the fault monitoring indicators; the key feature index acquisition module is used to obtain key fault feature indicators based on a preset second algorithm and the oversampled fault samples; the health score acquisition module is used to calculate the weighted Mahalanobis distance value of the key fault feature indicators based on a preset third algorithm to obtain the actual health score of the offshore wind turbine gearbox; the warning module is used to evaluate the health status of the offshore wind turbine gearbox based on the actual health score of the offshore wind turbine gearbox and issue corresponding fault warnings.

[0020] An embodiment of the present invention proposes a fault warning system for an offshore wind turbine gearbox. The fault mechanism analysis module analyzes the historical operation data of the offshore wind turbine gearbox to analyze the fault mechanism of the offshore wind turbine gearbox, so as to analyze the main factors causing faults during the historical operation process, and thus determine the fault monitoring indicators that need to be obtained. Since the fault samples in the offshore wind turbine gearbox belong to the minority class samples and the sample size is small, in order to make the subsequent fault warning results of the offshore wind turbine gearbox more reliable, it is necessary to use a preset first algorithm and fault monitoring indicators through the oversampling processing module to obtain corresponding oversampled fault samples to increase the number of fault samples. In addition, due to the diverse fault types of the offshore wind turbine gearbox and the different importance levels among the fault types, the key feature index acquisition module uses a preset second algorithm and oversampled fault samples to reasonably extract key fault feature indicators to reduce the data dimension and ensure the reliability of the subsequent fault warning of the offshore wind turbine gearbox. Then, the health score acquisition module uses a preset third algorithm to calculate the weighted Mahalanobis distance values of each key fault feature indicator, introducing weights to evaluate the importance of each key fault indicator for the health assessment of the offshore wind turbine gearbox, so as to accurately obtain the health score of the offshore wind power gear set. Finally, the warning module evaluates the health status of the offshore wind turbine gearbox through the health score of the offshore wind power gear set and issues corresponding fault warnings according to the corresponding health status. By analyzing the fault mechanism of the offshore wind turbine gearbox and evaluating the health status of the equipment, the reliability of the fault warning is improved. Description of the Drawings

[0021] Figure 1 It is a schematic flow chart of the steps of a fault warning method for an offshore wind turbine gearbox provided by an embodiment of the present invention;

[0022] Figure 2 It is a sample enhancement schematic diagram of a fault warning method for an offshore wind turbine gearbox provided by an embodiment of the present invention Figure 1 ;

[0023] Figure 3 It is a sample enhancement schematic diagram of a fault warning method for an offshore wind turbine gearbox provided by an embodiment of the present invention Figure 2 ;

[0024] Figure 4 It is a sample enhancement schematic diagram of a fault warning method for an offshore wind turbine gearbox provided by an embodiment of the present invention Figure 3 ;

[0025] Figure 5 It is a sample enhancement schematic diagram of a fault warning method for an offshore wind turbine gearbox provided by an embodiment of the present invention Figure 4 ;

[0026] Figure 6 Sample enhancement schematic of a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention Figure 5 ;

[0027] Figure 7 Sample enhancement schematic of a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention Figure 6 ;

[0028] Figure 8 Schematic diagram of random forest decision tree error analysis for a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention;

[0029] Figure 9 Schematic diagram of the arrangement of feature importance for a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention;

[0030] Figure 10 Schematic diagram of the health score of a gearbox of an offshore wind turbine for a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention;

[0031] Figure 11 Schematic diagram of the prediction of the health score of a gearbox of an offshore wind turbine for a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention Figure 1 ;

[0032] Figure 12 Schematic diagram of the prediction of the health score of a gearbox of an offshore wind turbine for a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention Figure 2 ;

[0033] Figure 13 Schematic diagram of the module structure of a fault warning system for a gearbox of an offshore wind turbine provided by an embodiment of the present invention. Detailed implementation manners

[0034] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a 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 embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0035] The transmission components of an offshore wind turbine gearbox include: a first-stage planetary gearbox, a second-stage planetary gearbox, a main bearing, and a generator bearing. The first-stage planetary gearbox includes: a first-stage internal gear ring, first-stage planetary gears, first-stage planetary bearings, elastic pin shafts, a first-stage spline output, a first-stage sun gear, and a first-stage planetary carrier; the second-stage planetary gearbox includes: a second-stage internal gear ring, second-stage planetary gears, second-stage planetary bearings, second-stage planetary shafts, second-stage planetary carrier bearings, a second-stage sun gear, and a second-stage planetary carrier. The lubrication and cooling system of the offshore wind turbine gearbox transports gear oil from the oil tank to the online filter through an oil pump, and then distributes it to each lubrication point to forcibly lubricate and cool the transmission components. The lubricating oil flows through the filter after being pressurized by the oil pump, and after the temperature rises, it flows through an external cooler for cooling, and then is transported to each lubrication point, and finally returns to the oil tank to complete the cycle. In this cycle, the lubricating oil carries impurities and wear debris and is filtered out in the filter, and at the same time absorbs the heat generated by gear meshing and bearing friction, and discharges the heat in the cooler. It is worth mentioning that the offshore wind turbine gearbox is one of the components in the offshore wind turbine that causes the highest downtime and maintenance costs due to faults. Therefore, it is necessary to carry out health state assessment and fault prediction and early warning for the offshore wind turbine gearbox to ensure the stable operation of the offshore wind turbine;

[0036] The state prediction and fault early warning technology of offshore wind turbine gearboxes has been continuously developing, from early monitoring technologies to mathematics-based intelligent fault diagnosis technologies, and then to advanced artificial neural networks and deep learning. Among them, SCADA (Supervisory Control And Data Acquisition) data is widely used due to its low cost and practicality. However, the preprocessing of unstructured data is still a major challenge, which affects the development efficiency of machine learning models. Data quality, low-frequency resolution, and data processing in different degradation stages further complicate the development of effective PHM (Predictive Health Management) solutions. In addition, the challenges faced by predictive models include uncertainty management, the lack of interpretability of deep learning methods, and the need for online updates;

[0037] The health status assessment technology of offshore wind turbine gearboxes generally uses methods such as industrial endoscopes and vibration tests in the industry to regularly detect the status of gearboxes, and comprehensively analyzes their health status with reference to relevant standards. Or machine learning algorithms are adopted, such as the health assessment of wind turbine gearboxes based on conditional convolutional autoencoder Gaussian mixture models: This technical solution encodes and decodes sensor information and time series information through the encoder part and extracts compressed features, and then designs an energy design evaluation index based on the probability distribution of the signal itself according to the Gaussian mixture model for health assessment. This method uses kernel density estimation to determine the threshold and uses real wind farm data to verify the effectiveness of the method. In addition, the state monitoring and health assessment of wind turbine gearboxes based on KECA-GRNN are adopted: A technology that combines expert knowledge and neural networks is used for the state monitoring and health assessment of wind turbine gearboxes. This technology collects various sensor data such as gearbox vibration signals and temperature signals, establishes a KECA-GRNN model, conducts training and prediction, monitors and assesses the health status of the gearbox, and provides corresponding maintenance suggestions. The main process of the health status assessment of offshore wind turbine gearboxes: Obtain SCADA data, process the data, input it into a machine learning model and calculate the failure degree score value, and conduct a health assessment by comparing the score value with a preset threshold. However, the above technical means lack an analysis of the failure mechanism and may not be able to accurately predict all types of failures. The deep learning model may be too complex, increasing the calculation cost and training time. Most models are offline and need to be improved to achieve online update and real-time prediction;

[0038] To solve these problems, the embodiment of the present invention proposes a fault warning method for offshore wind turbine gearboxes, which realizes the analysis of the fault mechanism of offshore wind turbine gearboxes and the assessment of the health status of offshore wind turbine gearboxes to improve the reliability of fault warning. For specific explanations, refer to the embodiments and will not be elaborated here. It is worth mentioning that in this embodiment, a certain offshore wind farm in China is used as the research object for explanation. This wind farm is equipped with 36 5.5MW wind turbines, with a total installed capacity of 198MW, and is equipped with an offshore booster station of 220kV and an onshore centralized control center. Since it was put into operation in 2019, it has been operating stably for 5 years. In this embodiment, the SCADA operation data of 36 wind turbines at 10-minute intervals throughout 2023 is collected, and the No. 5 wind turbine with a wear fault in the gearbox is taken as an example for analysis.

[0039] Embodiment 1

[0040] See Figure 1 , Figure 1 which is a schematic diagram of the step flow of a fault warning method for an offshore wind turbine gearbox provided by an embodiment of the present invention. As Figure 1As shown in the figure, an embodiment of the present invention proposes a fault warning method for an offshore wind turbine gearbox, which includes steps 101 to 105, and the specific steps are as follows:

[0041] Step 101: Based on the historical operation data of the offshore wind turbine gearbox, conduct a fault mechanism analysis on the offshore wind turbine gearbox to obtain fault monitoring indicators;

[0042] As an example of this embodiment, based on the historical operation data of the offshore wind turbine gearbox, obtain the historical operation fault types; based on the historical operation fault types, conduct a fault mechanism analysis on the offshore wind turbine gearbox to obtain the fault mechanism change indicators corresponding to the occurrence of faults; screen the fault mechanism change indicators corresponding to the occurrence of faults that meet the requirements of the preset change trend to obtain fault monitoring indicators.

[0043] By analyzing the historical operation data, obtain the corresponding fault types, conduct a fault mechanism analysis on the offshore wind turbine gearbox, and screen out the fault mechanism change indicators whose change trends meet the requirements when the fault occurs, and use them as fault monitoring indicators to preliminarily analyze and screen out the parameters that change most significantly during the fault occurrence process, so as to improve the reliability of the subsequent fault warning results. In this embodiment, by analyzing the fault mechanism of the gearbox and its cooling and lubrication system, the changes in the material flow and energy flow paths are explored. A specific implementable manner is that according to the historical operation data of the offshore wind turbine, the common faults of the offshore wind turbine gearbox can include mechanical faults, bearing damage, filter blockage, oil pump damage, and pipeline leakage, etc.:

[0044] (i) Mechanical faults include: tooth breakage, pitting, and wear, etc. In terms of temperature and oil pressure, due to the heat generated by tooth surface friction being difficult to effectively discharge, it will cause the temperature of the gearbox oil sump and the high-speed shaft bearing to rise, and may block the filter, reduce the oil flow, lower the viscosity and fluidity of the lubricating oil, reduce the oil circulation volume in the system, and cause the inlet and outlet oil pressures to decrease; in terms of electricity, due to the reduction of transmission efficiency, it will cause the generator speed, active power, and current to decrease; in terms of oil, it may cause poor internal sealing and metal wear, affecting the oil metal content, but having no obvious effect on the oil moisture and oil level;

[0045] (ii) Bearing damage faults include: inner ring, outer ring, rolling element, and cage faults. In terms of temperature and oil pressure, it causes an increase in bearing heat generation. At the same time, bearing damage may lead to poor lubrication, reduce the oil circulation volume in the system, and cause the inlet and outlet oil pressures to decrease; in terms of electricity, it affects the generator speed, active power, and current; in terms of oil, it causes changes in the oil metal content, but having no obvious effect on the oil moisture and oil level;

[0046] (iii) Filter blockage faults include: blockage caused by impurities in the lubricating oil or debris generated by mechanical failures. In terms of temperature and oil pressure, due to the increased pressure difference across the filter element, the oil volume is insufficient, and heat cannot be effectively discharged, resulting in an increase in the temperature of the gearbox oil sump and the high-speed shaft bearing, and a decrease in the oil flow rate in the oil circuit and a decrease in the inlet oil pressure; in terms of electricity, it will not affect the generator speed and active power; in terms of the oil, it will not affect the oil water content and oil metal content;

[0047] (iv) Oil pump damage faults include: damage to the motor or controller. In terms of temperature and oil pressure, it will cause a decrease in the oil circuit pressure and an increase in the temperature of the oil sump and bearings; in terms of electricity, it will not directly affect the generator speed and active power; in terms of the oil, it will not affect the oil metal content, but may cause a decrease in the oil level;

[0048] (v) Pipeline leakage faults. In terms of temperature and oil pressure, it causes a decrease in the system oil pressure and a reduction in the oil volume, leading to an increase in the internal temperature of the gearbox; in terms of electricity, it will not affect the generator speed and active power; in terms of the oil, it will cause a decrease in the oil level, but will not affect the oil water content and oil metal content;

[0049] By analyzing the above common fault types, the corresponding fault mechanism change indicators (equivalent to SCADA indicators) when a fault occurs can be obtained. In this embodiment, taking the changes in the corresponding fault mechanism change indicators (equivalent to SCADA indicators) when the above 5 fault types occur as an example, see Table 1:

[0050] Table 1 Variation law table of SCADA indicators corresponding to the occurrence of faults in the gearbox of offshore wind turbines

[0051]

[0052] In Table 1, the arrow indicates the change trend of the parameter when the corresponding fault occurs. "↑" means the measured value will increase, "↓" means the measured value will decrease, and "→" means the measured value changes according to the original law without an increasing or decreasing trend; the temperature of the high-speed shaft bearing of the gearbox includes two parameters, namely the temperature of the driving end bearing of the high-speed shaft of the gear and the temperature of the non-driving end bearing of the high-speed shaft of the gear; finally, according to the change trends of each SCADA indicator, the parameter with the most obvious change in the oil circuit circulation during a fault is selected to obtain the fault monitoring indicators (constituted by SCADA indicators). One example is shown in Table 2:

[0053] Table 2: Fault monitoring indicator table of the gearbox of offshore wind turbines based on mechanism analysis

[0054]

[0055]

[0056] In Table 2, after conducting a mechanism analysis on the offshore wind turbine gearbox, the parameters that change most significantly during a fault are integrated as fault monitoring indicators. In the subsequent processing, attention will be paid to the fault monitoring indicators obtained in Table 2. Generally speaking, before conducting health assessment and prediction, it is necessary to clean the SCADA data in the fault monitoring indicators to ensure the accuracy and availability of the data. Taking the "wind speed - power" curve of an offshore wind turbine as an example, data points where the wind speed or active power is less than or equal to 0, as well as data points where the wind speed is lower than the cut-in wind speed but the active power is greater than 0 or the wind speed is higher than the cut-out wind speed but the active power is greater than 1.2 times the rated power, are excluded. At the same time, in the [cut-in wind speed, cut-out wind speed] interval, data cleaning is carried out on the data using methods such as the 3σ criterion and the median method. Thus, through the analysis of the fault mechanism of the offshore wind turbine gearbox, fault monitoring indicators are reasonably obtained to improve the reliability of the subsequent health status assessment, health score calculation, and fault warning results of the offshore wind turbine gearbox.

[0057] Step 102: Based on a preset first algorithm and the fault monitoring indicators, obtain oversampled fault samples;

[0058] As an example of this embodiment, based on the fault monitoring indicators, obtain the historical monitoring dataset of the offshore wind turbine gearbox; divide the historical monitoring dataset of the offshore wind turbine gearbox into fault samples and normal samples; based on a preset first algorithm, calculate the number of normal samples at a preset first nearest neighbor distance from each of the fault samples, obtain the distribution density of the fault samples, and calculate the target synthetic sample number; based on a preset second nearest neighbor distance, randomly select the neighbor samples of each of the fault samples, and perform linear interpolation on each of the fault samples through the neighbor samples and the target synthetic sample number to obtain oversampled fault samples.

[0059] Since in the fault analysis of offshore wind turbines, fault samples belong to minority class samples, and the data imbalance problem seriously affects the reliability of fault warning results, it is necessary to enhance the data of the small fault samples in the SCADA data. The fault samples are oversampled through a preset first algorithm, and synthetic samples are synthesized through linear interpolation to increase the number of fault samples, which is beneficial to the subsequent health status assessment of the offshore wind turbine gearbox, and further improves the reliability of fault warning results. A specific implementable method is that in this embodiment, the ADASYN algorithm (equivalent to the preset first algorithm) is used to oversample the small fault samples to increase the number of minority class samples. Suppose there are n samples in the SCADA dataset of the offshore wind turbine, among which there are n min minority class fault samples and n maj majority class normal samples, then the imbalance degree is To make the data reach the desired balance β×n (β > 1), the number G of minority class samples to be synthesized is as follows:

[0060] G = (β - 1)n min ;

[0061] In the formula, β is a control parameter, β > 1, and n min is the number of minority class fault samples;

[0062] For each fault sample s in the minority class i , calculate the number Δ of majority class normal samples in its k nearest neighbors i , and the distribution density δ of the fault sample s i is: i In the formula, k is the preset first nearest neighbor distance;

[0063]

[0064] For the fault sample s

[0065] The number g of samples to be synthesized i is: i In the formula, N is a normalization factor, and its calculation formula is as follows:

[0066]

[0067] Combined with the number of samples to be synthesized, for each minority class fault sample s

[0068]

[0069] Randomly select a sample s from its k' nearest neighbors i (j is the index of the randomly selected nearest neighbor sample), and generate a new sample through linear interpolation: ij where r is a random number between [0, 1], and k' is the preset second nearest neighbor distance;

[0070]

[0071] By strengthening the minority class fault samples, the number of fault samples can be balanced with the number of normal samples. To further illustrate the impact of data strengthening on the reliability of fault warning results, in the examples of this invention embodiment, for the 29 SCADA indicators in Table 2, fault small sample strengthening is performed, and some representative indicators are selected as examples. See

[0072] For the sample strengthening schematic diagram of a fault warning method for an offshore wind turbine gearbox provided in an embodiment of this invention Figure 2 , Figure 2 As shown in Figure 1 ; as Figure 2 shown,Figure 2 is a sample enhancement schematic diagram of the oil sump temperature fault. As can be seen from Figure 2 it, when the fault occurs, the oil sump temperature shows an upward trend; see Figure 3 , Figure 3 which is a sample enhancement schematic diagram of a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention. Figure 2 As Figure 3 shown, Figure 3 is a sample enhancement schematic diagram of the gearbox temperature fault. As can be seen from Figure 3 it, when the fault occurs, the gearbox temperature shows an upward trend; see Figure 4 , Figure 4 which is a sample enhancement schematic diagram of a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention. Figure 3 As Figure 4 shown, Figure 4 is a sample enhancement schematic diagram of the cooler temperature fault. As can be seen from Figure 4 it, when the fault occurs, the cooler temperature shows an upward trend; see Figure 5 , Figure 5 which is a sample enhancement schematic diagram of a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention. Figure 4 As Figure 5 shown, Figure 5 is a sample enhancement schematic diagram of the inlet oil pressure fault. As can be seen from Figure 5 it, when the fault occurs, the inlet oil pressure shows a downward trend; see Figure 6 , Figure 6 which is a sample enhancement schematic diagram of a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention. Figure 5 As Figure 6 shown, Figure 6 is a sample enhancement schematic diagram of the oil metal content fault. As can be seen from Figure 5 it, when the fault occurs, the oil metal content shows an upward trend; see Figure 7 , Figure 7 which is a sample enhancement schematic diagram of a fault warning method for a gearbox of an offshore wind turbine provided by an embodiment of the present invention. Figure 6 As Figure 7 shown, Figure 7 is a sample enhancement schematic diagram of the generator speed fault. As can be seen from Figure 5 it, when the fault occurs, the generator speed shows a downward trend; as can be seen from Figures 2 to 7It can be seen that after strengthening the fault samples, the normal data and the fault data can be effectively distinguished, and the change trend of the fault samples is consistent with the change of the parameters in Table 1. Thus, it can be seen that by strengthening the data of the minority-class fault samples, the problem caused by data imbalance between the normal data and the fault data can be solved to improve the reliability of subsequent fault warning.

[0073] Step 103: Obtain key fault feature indicators based on a preset second algorithm and the oversampled fault samples.

[0074] As an example of this embodiment, several decision trees are constructed based on the oversampled fault samples; based on a preset second algorithm and several decision trees, the oversampled fault samples are classified through a preset voting mechanism to obtain several sample categories; based on several sample categories, out-of-bag sample evaluation is performed on the features of the samples corresponding to each sample category, the prediction errors of the samples corresponding to each sample category are calculated, and the feature importance is sorted to screen out key fault feature indicators.

[0075] Among a large number of fault types, there is an importance order for the features of each fault sample. Through a preset second algorithm, a voting mechanism is used to classify the oversampled fault samples, and then out-of-bag sample evaluation is used to evaluate the feature importance of each sample to screen out key fault feature indicators, so as to reduce the dimension of the data, ensure the data processing volume in the subsequent health assessment process of the offshore wind turbine gearbox, improve the accuracy of the health state assessment, and thus improve the reliability of the fault prediction result. Specifically, in an implementable manner, the random forest algorithm (equivalent to the preset second algorithm) can play a role in reducing the data dimension during the selection of feature indicators and improve the efficiency of subsequent fault analysis; first, decision trees are constructed, and the definition of a decision tree is:

[0076]

[0077] In the formula, f is the splitting feature (f ∈ {1, 2, …, F}, F is the number of features), θ is the splitting threshold, is the Gini impurity of node t, C is the total number of categories (two categories: normal and fault), p tc is the proportion of samples in node t belonging to category c, N t is the number of samples in node t, and are the numbers of samples in the left and right child nodes after splitting respectively, GINI(t l ) and GINI(t r ) are the Gini impurities of the left and right child nodes respectively;

[0078] See Figure 8 , Figure 8Schematic diagram of random forest decision tree error analysis for a fault warning method of an offshore wind turbine gearbox provided by an embodiment of the present invention; as Figure 8 shown, when the number of decision trees is close to 0, the error is relatively high. When the curve is at the left endpoint (the number of decision trees is close to 0), the error is about 0.3. As the number of decision trees increases, the error decreases. When the number of decision trees exceeds a certain number (about after 50), the error curve gradually flattens out and the error change is very small. It can be seen that when the number of decision trees is moderate, the variance can be effectively reduced and the accuracy of the random forest model can be improved;

[0079] After constructing the decision tree, the random forest algorithm is further used for data classification. For the M decision trees in the random forest, for the oversampled fault sample y, each decision tree T m will give a predicted class c m , and the final predicted class of the oversampled fault sample y by the random forest algorithm is:

[0080]

[0081] In the formula, I(·) is an indicator function. When c m =c, I = 1, otherwise it is 0. Through this voting mechanism of the indicator function, the accuracy and stability of classification can be improved;

[0082] After classifying the oversampled fault sample y, several sample classes are obtained. In order to achieve the purpose of data dimensionality reduction, the random forest algorithm will be used to measure the feature importance by calculating the change in the prediction error of the out-of-bag (OOB) samples before and after the features are randomly perturbed. Specifically, let the prediction error of the original OOB samples be E. For feature j, after randomly shuffling the values in its OOB samples, a new prediction error E' j is obtained, then the importance I j of feature j is:

[0083] I j =E - E' j ;

[0084] Through the out-of-bag sample evaluation, the feature importance vector I=(I1, I2, …, I p ) is obtained, and based on this, the key features are screened to reduce the data dimension;

[0085] In this embodiment, one example of the selected number of decision trees is 100 and the minimum number of leaves is 1. For the selection of feature indicators of the random forest, see Figure 9 , Figure 9 Schematic diagram of the arrangement of feature importance for a fault warning method of an offshore wind turbine gearbox provided by an embodiment of the present invention; as Figure 9As shown, by measuring the feature importance, there are significant differences in the importance of different features. The importance of some features is close to 0.45, while some are close to 0, indicating that when a fault occurs, some features have the same impact on the system as other features or their impact is weaker than other factors; in Figure 9 In Figure 9 , the index number corresponds to Table 2. By selecting the index number, it can be obtained that the wear of the gearbox of the offshore wind turbine shows strong correlations in aspects such as oil sump temperature, gearbox temperature, high-speed shaft bearing temperature, cooler temperature, outlet oil pressure, inlet oil pressure, and oil fluid, and shows weak correlations in aspects such as cooling water temperature, oil level, rotational speed, active power, current, and wind speed. After feature selection, the SCADA indicators are reduced from the original 29 to 13. At the same time, the accuracy rate after random forest screening is 1, and the F1 score is 1, which also verifies the robustness of the random forest for feature extraction of wind turbines.

[0086] Before performing step 104, the embodiments of the present invention also propose: constructing an original data matrix based on key fault feature indicators; performing unit vector data standardization on the original data matrix to obtain a standardized matrix; calculating the proportion of each sample in the standardized matrix, and calculating the key fault feature information entropy according to the proportion of each sample;

[0087] In order to more objectively reflect the importance degree of key fault feature indicators and reduce the influence of subjective factors, by calculating the information entropy of each key fault feature indicator, a reasonable weight is provided for the subsequent calculation of the health score to ensure the reliability of the fault warning result. A specific implementable manner is as follows: Suppose there are m key fault feature indicators in the SCADA system of the offshore wind turbine and n historical data samples, and construct the SCADA original data matrix. The formula is as follows:

[0088] X = (x ij ) n×m (i = 1, 2,..., n; j = 1, 2,..., m);

[0089] In the formula, x ij represents the jth key fault feature indicator data of the ith historical data;

[0090] In order to eliminate the influence of the dimension, perform unit vector standardization on the original data matrix. The calculation formula is as follows:

[0091]

[0092] Through the above standardization formula, for the feature vector x i = (x i1 , x i2 ,..., x im ) of the ith sample, the standardized feature vector y i = (yi1 , y i2 , …, y im ); Standardization makes the sample feature vectors become unit vectors, eliminates the difference in dimension, and enhances the comparability of indicators;

[0093] After standardizing the original data matrix, it is necessary to calculate the proportion of each sample in the standardized matrix. By calculating the sample proportion, the relative size of the i-th sample of the j-th key fault feature index relative to all samples can be reflected, preparing for the subsequent information entropy calculation. Specifically, using the standardized matrix Y = (y ij ) n×m , calculate the proportion p ij of the i-th sample under the j-th key fault feature index as:

[0094]

[0095] Immediately afterwards, calculate the information entropy. The information entropy e j of the j-th key fault feature index is:

[0096]

[0097] where, when p ij = 0, according to the limit definition and 1 / ln n makes the value of the information entropy fall within [0, 1]. This shows that when the proportions of samples under the corresponding key fault feature index are closer to being equal, the information entropy is larger, the uncertainty of the index is higher, and the amount of information is smaller;

[0098] Finally, calculate the weight ω ij of the j-th index according to the information entropy. The specific formula is as follows:

[0099]

[0100] Among them, the larger the information entropy, the smaller the amount of information of the index, and the smaller the impact on the evaluation result. This weight calculation objectively reflects the importance of the key fault feature index, reduces the influence of subjective factors, and is used for subsequent weighted summation. Combining the features selected by the random forest provides reasonable weights for the weighted Mahalanobis distance calculation;

[0101] In the example of the embodiment of the present invention, combined with Figure 6 the selected key fault feature indexes and the historical data of the SCADA system, the weights of each feature index can be obtained by the entropy weight method. One explanation is shown in Table 3:

[0102] Table 3 Weights of Key Fault Feature Indexes of the SCADA System for Offshore Wind Turbine Gearboxes

[0103]

[0104]

[0105] In Table 3, the weight of the oil sump temperature is relatively small, indicating that the influence of the oil sump temperature on the overall system performance of the gearbox is relatively small; the weights of the two indicators of the gearbox temperature are the same and slightly higher than the weight of the oil sump temperature, but still at a relatively low level; the weight of the high-speed shaft bearing temperature is slightly higher than the aforementioned gearbox temperature indicators, indicating that the high-speed shaft bearing temperature has a slightly greater impact on the system performance; the weight of the cooler temperature is relatively high, especially the weight of the outlet oil temperature is the highest, indicating that the cooler temperature has a great impact on the gearbox performance. The inlet and outlet oil temperatures of the cooler can reflect the efficiency of the cooling system and are crucial for the thermal management and stability of the gearbox; the weights of the outlet oil pressure indicators are different, and the weight of the outlet pressure of the filter pump is relatively high, meaning that the outlet pressure of the filter pump has a greater impact on the system performance because it is directly related to the oil filtration effect and system lubrication; the weights of the inlet oil pressure are generally low, indicating that these pressure indicators have a relatively small impact on the system performance; the weight of the oil is relatively high, indicating that the metal content in the oil has a greater impact on the gearbox. An increase in the metal content means increased internal wear of the gearbox.

[0106] Step 104, based on a preset third algorithm, calculate the weighted Mahalanobis distance value of the key fault feature indicators to obtain the actual health score of the gearbox of the offshore wind turbine.

[0107] As an example of this embodiment, based on the original data matrix, calculate the Mahalanobis distance of each key fault feature indicator in the original data matrix; based on the weights of each key fault feature, calculate the weighted Mahalanobis distance value through the preset third algorithm; based on the weighted Mahalanobis distance value, obtain the actual health score of the gearbox of the offshore wind turbine. Based on the key fault feature information entropy, obtain the weights of each key fault feature.

[0108] Since there are correlations among the operating parameters of the gearbox of the offshore wind turbine, for example, temperature and pressure will affect each other. Therefore, calculating the Mahalanobis distance can reasonably reflect the relationship between parameters. Then, combining the weight relationship obtained from the information entropy of each key fault feature indicator calculated in advance into the Mahalanobis distance, introducing weights to emphasize the importance of each parameter in health assessment, and further obtaining a reliable health score to ensure the accuracy of health status assessment. A specific implementable way is to evaluate the health status by calculating the adjusted difference of the covariance between the operating parameters (such as temperature, vibration, pressure, etc.) of the offshore wind turbine and the normal operating state parameters; first, calculate the Mahalanobis distance. The SCADA original data matrix is X, the mean of each key fault feature indicator of the data set is μ=(u1, u2, …, u m ), and the covariance matrix is Σ=(σ ij ) m×m , and the sample point xi =(x i1 , x i2 , …, x im ) to the Mahalanobis distance d from the mean μ i is as follows:

[0109]

[0110] where:

[0111]

[0112] The Mahalanobis distance considers the data covariance structure, excludes the interference of variable correlation, accurately measures the distance between the data point and the mean, reflects the data dispersion degree, and can initially judge whether the sample deviates from the normal operation state;

[0113] After calculating the Mahalanobis distance, the weighted Mahalanobis distance is calculated by introducing weights, emphasizing the differences in the importance of different indicators, effectively judging the deviation degree of the data point from the normal distribution in anomaly detection, and combining the weight ω obtained by the entropy weight method ij , the weighted Mahalanobis distance D from the sample point x i to the mean μ is calculated as follows: i The calculation formula is as follows:

[0114]

[0115] where, the calculation formula of the diagonal matrix Λ is as follows:

[0116]

[0117] After calculating the weighted Mahalanobis distance, the Mahalanobis distance is normalized to obtain the corresponding health score. The specific calculation method can be realized by the existing technology and will not be elaborated here.

[0118] Step 105, based on the actual health score of the offshore wind turbine gearbox, evaluate the health state of the offshore wind turbine gearbox and issue a corresponding fault warning.

[0119] As an example of this embodiment, if the actual health score of the offshore wind turbine gearbox is greater than or equal to the preset first health threshold, the health status of the offshore wind turbine gearbox is evaluated as healthy; if the actual health score of the offshore wind turbine gearbox is less than the preset first health threshold and greater than the preset second health threshold, the health status of the offshore wind turbine gearbox is evaluated as having a mild abnormality; if the actual health score of the offshore wind turbine gearbox is less than or equal to the preset second health threshold, it indicates that the health status of the offshore wind turbine gearbox is evaluated as having a serious abnormality; based on the health status of each offshore wind turbine gearbox, corresponding fault warnings are issued. If the health status of the offshore wind turbine gearbox is healthy, no fault warning is issued; if the health status of the offshore wind turbine gearbox is having a mild abnormality, a mild fault warning is issued; if the health status of the offshore wind turbine gearbox is having a serious abnormality, a serious fault warning is issued.

[0120] The prior art lacks an analysis of the fault mechanism, and the deep learning model used for the health status assessment of the offshore wind turbine gearbox is too complex to accurately predict the fault type, resulting in the problem of unreliable fault warnings. To solve such a problem, the embodiments of the present invention propose to analyze the fault mechanism of the offshore wind turbine gearbox, and then integrate the fault characteristic indicators. Through sample-enhanced data dimensionality reduction, accurate and simple data is obtained to ensure the simplification of data processing. Finally, the weights of the key fault characteristic indicators are calculated by the entropy weight method, and then the Mahalanobis distance is calculated, and the weighted Mahalanobis distance is obtained by introducing the weights. Finally, the health score is calculated, and the health status of the offshore wind turbine gearbox is evaluated through the health score. Specifically, see Figure 10 , Figure 10 is a schematic diagram of the health score of the offshore wind turbine gearbox for a fault warning method of an offshore wind turbine gearbox provided in an embodiment of the present invention; as Figure 10 shown, the health score calculated by combining the weights of the key fault characteristic indicators selected in Table 3. The horizontal axis (X-axis): labeled "Time", and the time range is from 00:00:00 on January 1, 2023 to 00:00:00 on January 1, 2024, showing the time span of data collection. The vertical axis (Y-axis): labeled "Health Score", with a range from 0 to 1, representing the health condition of the gearbox. The higher the score, the better the health condition of the gearbox. At Figure 10Among them, the health score of the wind turbine gearbox fluctuates between 0.6 and 0.9 for most of the time, showing a relatively stable health condition. However, at the time point of 18:20:00 on May 21, 2023, the health score dropped significantly to 0.05; at the time point of 13:50:00 on July 18, 2023, the health score dropped to 0.16 again; at the time point of 20:00:00 on October 21, 2023, the health score dropped to 0.10 again. These three time points are also consistent with the time when the gearbox of the unit suffered wear, verifying the accuracy and rationality of using the weighted Mahalanobis distance to calculate the health score. The health status of the offshore wind turbine gearbox is evaluated according to the health score, and corresponding fault warnings are issued. For example, one warning can be carried out according to the rules shown in Table 4:

[0121] Table 4 Fault Warning Rule Table for Offshore Wind Turbine Gearbox

[0122] Health score Health status Fault warning result ≥0.6 Health No warning <0.6 and >0.5 Mild anomaly Mild fault warning ≤0.5 Severe anomaly Severe fault warning

[0123] In the example shown in Table 4, when the health score is greater than or equal to 0.6 (equivalent to the preset first health threshold), the anti-state of the offshore wind turbine gearbox is evaluated as healthy at this time, and no fault warning needs to be issued; when the health score is less than 0.6 (equivalent to the preset first health threshold), the anti-state of the offshore wind turbine gearbox is evaluated as healthy at this time, and no fault warning needs to be issued; when the health score is less than 0.6 (equivalent to the preset first health threshold) and greater than 0.5 (equivalent to the preset second health threshold), the anti-state of the offshore wind turbine gearbox is evaluated as mildly abnormal at this time, and a mild fault warning is issued to remind the operators to check the fault hidden dangers in time; when the health score is less than or equal to 0.5 (equivalent to the preset second health threshold), the anti-state of the offshore wind turbine gearbox is evaluated as severely abnormal at this time, and a severe fault warning is issued to remind the operators to shut down the offshore wind turbine gearbox in time and carry out maintenance in time, and handle the fault according to the key fault characteristic indicators;

[0124] After performing step 105, an embodiment of the present invention further proposes: obtaining historical operation data of an offshore wind turbine gearbox, dividing the fault monitoring indicators of the historical operation data of the offshore wind turbine gearbox into a training set and a test set according to a preset ratio; constructing an initial health score prediction model, and optimizing the model parameters of the initial health score prediction model through a preset fourth algorithm according to the training set to obtain a target health score prediction model; inputting the test set into the target health score prediction model to obtain a health score prediction value; calculating an error index based on the health score prediction value and the actual health score of the offshore wind turbine gearbox; if the error index meets the preset requirements, predicting the health status of the offshore wind turbine gearbox in a target time period through the target health score prediction model, and issuing a corresponding fault warning according to the health status of each offshore wind turbine gearbox in the target time period.

[0125] According to the historical operation data of the offshore wind turbine gearbox, a target health score prediction model is trained. By obtaining the health score prediction value and combining the actual health score value, an error index is calculated. If the error index meets the requirements, it indicates that the prediction accuracy of the target health score prediction model meets the requirements. The target health score prediction model can be used to predict the health status of the offshore wind turbine gearbox in the future target time period to perform early warning analysis on faults, thereby improving the reliability of the fault warning result. A specific implementable manner is to utilize the ability of the LSTM model to handle long-term dependencies in sequence data and effectively capture the dynamic characteristics in the operation data of the offshore wind turbine to predict the health score in the future target time period, so as to predict the possible occurrence of faults in advance. Specifically, the SCADA data of the offshore wind turbine (including indicators such as wind speed, power, temperature, and vibration) is divided into a training set and a test set according to a ratio of 80:20 (equivalent to the preset ratio). The training set is used to train the LSTM model, and the test set is used to evaluate the generalization ability and prediction performance of the model. In this embodiment, the architecture of the model is set as an LSTM model including an input layer, an LSTM layer, a fully connected layer, and an output layer. Among them, the input layer receives the normalized monitoring data sequence; the LSTM layer learns long-term dependencies through internal gate control structures (forget gate, update gate, processing gate, and output gate); the fully connected layer maps the output features of the LSTM layer; the output layer outputs the predicted health score;

[0126] After designing the model architecture, the LSTM model parameters are optimized using the training set data through the backpropagation algorithm. During training, the model is monitored based on performance evaluation metrics to ensure convergence and good performance. The trained LSTM model is used to predict the test set data to obtain the predicted health score sequence. By calculating various error metrics (RMSE, MAE, MBE, MAPE) and the coefficient of determination R2 between the predicted values and the actual values, the performance of the model on the test set is evaluated to show that the prediction accuracy of the target health score prediction model meets the requirements.

[0127] To further explain the availability of the target health score prediction model for health score prediction, this embodiment presents an example. In May 2023 and October 2023, the SCADA system both had excessively high gearbox oil temperature, which was actually a gearbox wear fault. To verify the accuracy of the prediction model, historical data for two months was selected to predict the fault data for the next two weeks, and traditional algorithms such as BP (neural network), SVM (support vector machine), and RF (random forest) were used as reference objects to predict the health score. The comparison of each prediction method is as follows:

[0128] Table 5 Comparison results of each prediction method

[0129] Serial number Prediction method RMSE <![CDATA[R 2 > MAE MBES MAPE 1 LSTM 0.55 0.93 0.35 -0.08 0.03 2 BP 0.87 0.80 0.51 -0.22 0.03 3 SVM 3.08 0.60 2.01 -1.25 0.09 4 RF 1.22 0.75 0.75 -0.48 0.04

[0130] In Table 5, the LSTM model (equivalent to the target health score prediction model) proposed in the embodiment of the present invention, as well as the various error metrics and the coefficient of determination R of the existing traditional BP (neural network), SVM (support vector machine), and RF (random forest), can be compared. 2 From Table 5, it can be seen that the various error metrics and the coefficient of determination of the LSTM model proposed in the embodiment of the present invention are all better than the traditional algorithms, and all meet the specification requirements (equivalent to the preset requirements) that the MAPE of the medium-term prediction results of the wind farm should not be greater than 0.3. At the same time, the R of LSTM 2 is 0.93, which means that the model can explain 93% of the variability of the test data. Although it is slightly lower than the training set, this high value still indicates that the model has good generalization ability. The mean absolute percentage error (MAPE) is 2%, indicating that the average deviation between the predicted values and the actual values is small, reflecting the high prediction accuracy of the model.

[0131] See Figure 11 and Figure 12 , Figure 11 is a schematic diagram of the health score prediction of the gearbox of an offshore wind turbine provided in an embodiment of the present invention for a fault warning method of the gearbox of an offshore wind turbine. Figure 1 ; Figure 12 is a schematic diagram of the health score prediction of the gearbox of an offshore wind turbine provided in an embodiment of the present invention for a fault warning method of the gearbox of an offshore wind turbine.Figure 2 ; As shown in Figure 11 and Figure 12 , the variation of the health score of the gearbox of a wind turbine with time is shown. It can be seen from the figure that in the overall trend, the health score (red line) predicted by the target health score prediction model proposed in the embodiment of the present invention can better track the change of the actual health score (blue line). Thus, it can be known that the accuracy of the target health score prediction model obtained by training with the LSTM model in predicting the health score of the gearbox of a wind turbine. Although there are certain deviations, these deviations are within an acceptable range, which proves the reliability of the model prediction. Furthermore, it can effectively evaluate and predict the health status of the gearbox of an offshore wind turbine, providing a scientific basis for the maintenance and fault prevention of the equipment.

[0132] The embodiment of the present invention proposes a fault warning method for the gearbox of an offshore wind turbine. According to the historical operation data of the gearbox of the offshore wind turbine, the fault mechanism of the gearbox of the offshore wind turbine is analyzed, so that the main factors causing faults during the historical operation process can be analyzed. Thus, it is determined that fault monitoring indicators need to be obtained. Since the fault samples belong to minority class samples in the gearbox of the offshore wind turbine and the sample quantity is small, in order to make the subsequent fault warning result of the gearbox of the offshore wind turbine more reliable, it is necessary to obtain corresponding oversampled fault samples through a preset first algorithm and the fault monitoring indicators to increase the number of fault samples. In addition, since the fault types of the gearbox of the offshore wind turbine are diverse and the importance degrees among the fault types are different, key fault feature indicators are reasonably extracted through a preset second algorithm and the oversampled fault samples to reduce the data dimension and ensure the reliability of the subsequent fault warning of the gearbox of the offshore wind turbine. Then, the weighted Mahalanobis distance values of each key fault feature indicator are calculated through a preset third algorithm, and weights are introduced to evaluate the importance of each key fault indicator for the health assessment of the gearbox of the offshore wind turbine, so as to accurately obtain the health score of the offshore wind gear set. Finally, the health status of the gearbox of the offshore wind turbine is evaluated through the health score of the offshore wind gear set, and corresponding fault warnings are issued according to the corresponding health status. By analyzing the fault mechanism of the gearbox of the offshore wind turbine and evaluating the health status of the equipment, the reliability of the fault warning is improved.

[0133] Embodiment 2

[0134] See Figure 13 , Figure 13 , which is a schematic diagram of the module structure of a fault warning system for the gearbox of an offshore wind turbine provided by an embodiment of the present invention. As shown in Figure 13As shown in the figure, an embodiment of the present invention further provides a fault warning system for an offshore wind turbine gearbox, including: a fault mechanism analysis module 201, an oversampling processing module 202, a key feature index acquisition module 203, a health score acquisition module 204, and a warning module 205; the fault mechanism analysis module 201 is configured to perform fault mechanism analysis on the offshore wind turbine gearbox based on the historical operation data of the offshore wind turbine gearbox to obtain fault monitoring indexes; the oversampling processing module 202 is configured to obtain oversampled fault samples based on a preset first algorithm and the fault monitoring indexes; the key feature index acquisition module 203 is configured to obtain key fault feature indexes based on a preset second algorithm and the oversampled fault samples; the health score acquisition module 204 is configured to calculate the weighted Mahalanobis distance value of the key fault feature indexes based on a preset third algorithm to obtain the actual health score of the offshore wind turbine gearbox; the warning module 205 is configured to evaluate the health state of the offshore wind turbine gearbox based on the actual health score of the offshore wind turbine gearbox and issue corresponding fault warnings.

[0135] An embodiment of the present invention proposes a fault warning system for an offshore wind turbine gearbox. The fault mechanism analysis module analyzes the fault mechanism of the offshore wind turbine gearbox according to the historical operation data of the offshore wind turbine gearbox, so as to analyze the main factors causing faults during the historical operation process, and thus determine the need to obtain fault monitoring indexes. Since the fault samples in the offshore wind turbine gearbox belong to the minority class samples and the sample quantity is small, in order to make the subsequent fault warning results of the offshore wind turbine gearbox more reliable, it is necessary for the oversampling processing module to use a preset first algorithm and the fault monitoring indexes to obtain corresponding oversampled fault samples to increase the number of fault samples. In addition, due to the diverse fault types of the offshore wind turbine gearbox and the different importance levels among the fault types, the key feature index acquisition module uses a preset second algorithm and the oversampled fault samples to reasonably extract key fault feature indexes to reduce the data dimension and ensure the reliability of the subsequent fault warning of the offshore wind turbine gearbox. Then, the health score acquisition module uses a preset third algorithm to calculate the weighted Mahalanobis distance value of each key fault feature index, introducing weights to evaluate the importance of each key fault index for the health assessment of the offshore wind turbine gearbox, so as to accurately obtain the health score of the offshore wind power gear set. Finally, the warning module evaluates the health state of the offshore wind turbine gearbox through the health score of the offshore wind power gear set and issues corresponding fault warnings according to the corresponding health state. By analyzing the fault mechanism of the offshore wind turbine gearbox and evaluating the health state of the equipment, the reliability of the fault warning is improved.

[0136] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

[0137] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples", etc. means that the specific features, structures, materials or characteristics described in connection 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 can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0138] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present application, "a plurality" means two or more, unless otherwise specifically defined.

Claims

1. A fault warning method for an offshore wind turbine gearbox, characterized in that: include: Based on the historical operation data of offshore wind turbine gearboxes, the fault mechanism of offshore wind turbine gearboxes is analyzed to obtain fault monitoring indicators; Based on the preset first algorithm and the fault monitoring indicator, an oversampled fault sample is obtained; Based on the preset second algorithm and the oversampled fault samples, a key fault characteristic indicator is obtained; Based on a preset third algorithm, a weighted Mahalanobis distance value of the key fault characteristic indicator is calculated to obtain an actual health score of the gearbox of the offshore wind turbine; Based on the actual health score of the offshore wind turbine gearbox, the health status of the offshore wind turbine gearbox is evaluated, and a corresponding fault warning is issued.

2. A fault warning method for an offshore wind turbine gearbox according to claim 1, characterized in that: The method of analyzing the fault mechanism of the offshore wind turbine gearbox based on the historical operation data of the offshore wind turbine gearbox and obtaining fault monitoring indicators includes: Based on the historical operation data of the gearbox of the offshore wind turbine generator set, obtaining the historical operation fault type; Based on the historical operation fault types, a fault mechanism analysis is performed on the gearbox of the offshore wind turbine to obtain a corresponding fault mechanism change index when the fault occurs; The fault mechanism change index corresponding to the fault occurring when the fault meets the preset change trend requirement is screened to obtain the fault monitoring index.

3. A fault warning method for an offshore wind turbine gearbox according to claim 1, characterized in that: The obtaining of oversampled fault samples based on the preset first algorithm and the fault monitoring indicator includes: Based on the fault monitoring indicators, a historical monitoring data set of the offshore wind turbine gearbox is obtained; Dividing the offshore wind turbine gearbox historical monitoring data set into fault samples and normal samples; Based on a preset first algorithm, the number of normal samples at a preset first nearest neighbor distance from each of the faulty samples is calculated, the distribution density of the faulty samples is obtained, and the target number of synthetic samples is calculated; Based on a preset second nearest neighbor distance, a nearest neighbor sample of each of the fault samples is randomly selected, and each of the fault samples is linearly interpolated by the nearest neighbor sample and the target number of synthetic samples to obtain an oversampled fault sample.

4. A fault warning method for an offshore wind turbine gearbox according to claim 1, characterized in that: The key fault characteristic index is obtained based on the preset second algorithm and the oversampled fault sample, including: Based on the oversampled fault samples, construct several decision trees; Based on the preset second algorithm and the plurality of decision trees, the oversampled fault samples are classified by a preset voting mechanism to obtain a plurality of sample categories; Based on several of the sample categories, out-of-bag sample evaluation is performed on the features of samples corresponding to each of the sample categories, the prediction errors of samples corresponding to each of the sample categories are calculated and the features are ranked by importance, so as to screen out key fault feature indicators.

5. A fault warning method for an offshore wind turbine gearbox according to claim 1, characterized in that: Before executing the step of calculating the weighted Mahalanobis distance value based on the preset third algorithm and the key fault characteristic index to obtain the health score of the offshore wind turbine gearbox, the method further includes: Based on the key fault characteristic indicators, the original data matrix is ​​constructed; Performing unit vector data normalization on the original data matrix to obtain a normalized matrix; Calculating the weight of each sample in the standardized matrix, and calculating the key fault feature information entropy according to the weight of each sample; Based on the key fault feature information entropy, the weight of each key fault feature is obtained.

6. A fault warning method for an offshore wind turbine gearbox as claimed in claim 5, characterized in that: The method of calculating the weighted Mahalanobis distance value of the key fault characteristic index based on the preset third algorithm to obtain the actual health score of the offshore wind turbine gearbox includes: Based on the original data matrix, calculating the Mahalanobis distance of each of the key fault characteristic indicators in the original data matrix; Based on the weights of the key fault features, calculating the weighted Mahalanobis distance value by the preset third algorithm; Based on the weighted Mahalanobis distance value, an actual health score of the offshore wind turbine gearbox is obtained.

7. A fault warning method for an offshore wind turbine gearbox according to claim 1, characterized in that: Based on the actual health score of the offshore wind turbine gearbox, the health status of the offshore wind turbine gearbox is evaluated, and a corresponding fault warning is issued, including: If the actual health score of the offshore wind turbine gearbox is greater than or equal to a preset first health threshold, the health status of the offshore wind turbine gearbox is evaluated to be healthy; If the actual health score of the offshore wind turbine gearbox is less than a preset first health threshold and greater than a preset second health threshold, the health status of the offshore wind turbine gearbox is evaluated as being slightly abnormal; If the actual health score of the offshore wind turbine gearbox is less than or equal to the preset second health threshold, it means that the health status of the offshore wind turbine gearbox is evaluated to be seriously abnormal; Based on the health status of each of the offshore wind turbine gearboxes, a corresponding fault warning is issued.

8. A fault warning method for an offshore wind turbine gearbox as claimed in claim 7, characterized in that: Based on the health status of each of the offshore wind turbine gearboxes, a corresponding fault warning is issued, including: If the health status of the offshore wind turbine gearbox is healthy, no fault warning is issued; If the health status of the offshore wind turbine gearbox is a slight abnormality, a slight fault warning is issued; If the health status of the offshore wind turbine gearbox is seriously abnormal, a serious fault warning is issued.

9. A fault warning method for an offshore wind turbine gearbox according to any one of claims 1 to 8, characterized in that: After executing the steps of evaluating the health status of the offshore wind turbine gearbox based on the actual health score of the offshore wind turbine gearbox and issuing a corresponding fault warning, the method further includes: Acquire historical operation data of the gearbox of the offshore wind turbine set, and divide the fault monitoring indicators of the historical operation data of the gearbox of the offshore wind turbine set into a training set and a test set according to a preset ratio; Constructing an initial health score prediction model, and optimizing the model parameters of the initial health score prediction model by a preset fourth algorithm according to the training set to obtain a target health score prediction model; Inputting the test set into the target health score prediction model to obtain a health score prediction value; Calculating an error index based on the predicted health score and an actual health score of the offshore wind turbine gearbox; If the error index meets the preset requirements, the health status of the offshore wind turbine gearbox in the target time period is predicted by the target health score prediction model, and a corresponding fault warning is issued according to the health status of each offshore wind turbine gearbox in the target time period.

10. A fault warning system for an offshore wind turbine gearbox, characterized in that: include: Fault mechanism analysis module, oversampling processing module, key characteristic index acquisition module, health score acquisition module and early warning module; The fault mechanism analysis module is used to perform fault mechanism analysis on the gearbox of the offshore wind turbine and obtain fault monitoring indicators; The oversampling processing module is used to obtain an oversampling fault sample based on a preset first algorithm and the fault monitoring indicator; The key characteristic indicator acquisition module is used to obtain the key fault characteristic indicator based on the preset second algorithm and the oversampled fault sample; The health score acquisition module is used to calculate the weighted Mahalanobis distance value of the key fault characteristic indicator based on a preset third algorithm to obtain an actual health score of the offshore wind turbine gearbox; The early warning module is used to evaluate the health status of the offshore wind turbine gearbox based on the actual health score of the offshore wind turbine gearbox and issue a corresponding fault early warning.

Images