Offshore wind turbine generator set fault prediction method and device considering environmental factor influence

Abstract

The invention discloses an offshore wind turbine generator fault prediction method and device considering the influence of environmental factors, and the method comprises the steps: dividing an offshore wind turbine generator into a sensor system, a mechanical system and a tower structure system according to the correlation between the fault rate and the environmental factors, and building fault detection models respectively; time convolutional network prediction models are respectively established in combination with unit operation data and environment data to predict the operation state of each system, and fault occurrence is predicted in combination with a fault prediction model; coupling each system fault prediction model, constructing a coupling fault prediction model based on unit operation and an offshore wind turbine generator fault tree, carrying out label classification on predicted faults of each system fault model, carrying out verification through the fault tree, carrying out analysis and classification on fault causes, and supplementing the fault tree when judging that a new fault type appears; according to the method, the correlation between the offshore environment and the fault generation increments of different parts in the wind turbine generator is considered, so that the fault generation prediction accuracy of the offshore wind turbine generator is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of offshore wind turbine fault prediction, and more particularly to an offshore wind turbine fault prediction method and device taking environmental factors into consideration. Background Art

[0002] Offshore wind power systems have developed rapidly as an important clean energy power generation method. However, compared with the higher availability of onshore wind turbines, offshore wind power systems operate in a complex and harsh marine environment, with a high failure rate, poor accessibility, difficult maintenance, and much higher maintenance costs than onshore wind turbines.

[0003] In related technologies, troubleshooting is usually done only when the unit experiences a fault or abnormal condition. On-site operators are unable to handle and prevent the fault in a timely manner, which can have a significant impact on the normal operation of the unit. Therefore, if the occurrence of a fault can be predicted, the operation and maintenance costs of offshore wind turbines will be greatly reduced.

[0004] Fault prediction technology takes the current usage status of the equipment as the starting point, and combines the known characteristics, environmental conditions and historical data of the prediction object to predict future faults of the equipment in order to eliminate the faults in advance; most of the existing wind turbine fault prediction models use machine learning algorithms to mine historical data, analyze and process a large amount of historical operation data of wind turbines, and predict wind turbine faults by establishing fault prediction models.

[0005] However, previous studies on offshore wind turbine fault prediction usually assume that the unit equipment gradually deteriorates over time and then fails based on the natural degradation process of the unit. This ignores the impact of the objective offshore environment and makes it difficult to accurately describe the actual laws of unit failures under harsh offshore operating environments.

[0006] Therefore, how to provide a solution that takes into account the impact of environmental factors and accurately predicts the occurrence of offshore wind turbine failures is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0007] In view of this, the present invention provides a method and device for predicting faults of offshore wind turbines taking into account the influence of environmental factors to solve the technical problems mentioned in the background technology.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] A method for predicting faults of offshore wind turbines taking into account the influence of environmental factors comprises the following steps:

[0010] S1. Offshore wind turbines are divided into sensor systems, mechanical systems and tower structure systems according to the correlation between their failure rates and environmental factors, and different fault prediction models are established according to the characteristics of each system, including the sensor KPCA-KD fault detection model, the mechanical system fault diagnosis model taking environmental factors into account and the tower structure damage detection model;

[0011] S2. Combine the operating data of the offshore wind turbine with the corresponding environmental data, and establish TCN time convolution network prediction models according to the divided systems to predict the operating status of each system of the offshore wind turbine and input them into the corresponding fault prediction model to predict the occurrence of faults;

[0012] S3. Couple the fault prediction models of each system, build a coupled fault prediction model based on unit operation and a fault tree for offshore wind turbines, and classify the faults predicted by the fault models from different systems;

[0013] S4. Verify the fault prediction through the fault tree of offshore wind turbines, analyze and classify the causes of the faults, and supplement the fault tree when new fault types are detected;

[0014] S4. Collect unit data and environmental data in real time to modify fault prediction models and fault trees.

[0015] Preferably, the specific content of step S2 is:

[0016] In different systems, environmental data are extracted using the ReliefF algorithm to select environmental factor features with greater correlation as input variables;

[0017] By combining the historical operating data and corresponding environmental data of each system, a TCN prediction model is constructed to predict the operating status parameters of each system, and the parameters are input into the corresponding fault prediction model as input to perform fault diagnosis of the unit.

[0018] Preferably, the specific content of the sensor KPCA-KD fault detection model is:

[0019] The KPCA model is constructed based on the trouble-free operation data of offshore wind turbines, and the nonlinear principal component score T1 is calculated;

[0020] Substitute the predicted unit operation data into the KPCA model, calculate the nonlinear principal component score T2 at this time, and calculate the KD value between T1 and T2;

[0021] The kernel density estimation method is used to calculate the detection threshold in the KD scenario;

[0022] Compare the KD value with the detection threshold to determine the occurrence of sensor failure;

[0023] The offshore wind turbine database is updated through equipment operation data, and the KPCA model and detection threshold are corrected.

[0024] Preferably, the specific content of the mechanical system fault diagnosis model taking environmental factors into account is:

[0025] The ReliefF algorithm is used to select the optimal characteristic value of the environmental factors that affect the failure rate of the mechanical system;

[0026] Construct a mechanical system failure rate model that takes environmental influences into account, and estimate model parameters based on maximum likelihood method and particle swarm algorithm;

[0027] Combined with the impact of equipment preventive maintenance on the failure rate, the equipment degradation process is modeled to predict the remaining life of the equipment;

[0028] Considering the corrosion effect of offshore environment on wind turbine blades, a blade corrosion model is constructed;

[0029] Combining the corrosion model and the remaining life prediction value, when the remaining life prediction value or the corrosion degree exceeds the preset value, a fault warning is issued to the mechanical system;

[0030] The mechanical system fault diagnosis model is corrected by combining real-time environmental monitoring data and real-time unit operation status monitoring data.

[0031] Preferably, the specific content of the tower structure damage detection model is:

[0032] According to the tower database of offshore wind turbines, the intact tower structure and the damaged tower structure are classified, and the frequency change rate and square change rate of the natural frequency of the intact tower structure and the damaged tower structure are normalized;

[0033] The processed data is used as the neural network input parameter through the BP neural network to build a diagnostic model of the damage location and damage degree of the offshore wind power tower;

[0034] The tower structure database is supplemented by the real-time data of the tower structure, and the tower structure damage detection model is corrected.

[0035] Preferably, the specific content of step S3 is:

[0036] The operating data and corresponding environmental data of offshore wind turbines are selected to form a training data set for the fault scenarios of each system of the unit and the multiple fault scenarios between systems, and the fault scenarios are labeled with scenario labels;

[0037] The ReliefF algorithm is used to calculate the weights of each system eigenvalue under different scenarios;

[0038] The weights of the characteristic values ​​of each system under different fault scenarios are determined using the fuzzy C-means clustering algorithm to determine the cluster center and its membership degree, and the corresponding fault scenario label is marked on each cluster center;

[0039] When the model makes a fault prediction, it determines the membership of each scenario by calculating the weight data of the characteristic values ​​of each system, divides the predicted fault into sensors, mechanical systems and tower structures according to the fault scenario labels, and verifies it by the fault tree.

[0040] Preferably, the specific contents of constructing the fault tree of the offshore wind turbine are:

[0041] Based on the top event, the fault tree of offshore wind turbines is divided into pitch subsystem failure, transmission chain subsystem failure, generator subsystem failure, yaw subsystem failure, control subsystem failure, converter subsystem failure and tower structure failure;

[0042] The deductive method is used to analyze the subsystem failures of the offshore wind turbine fault tree from top to bottom, and the causes of the failures are analyzed step by step and refined to the components in the subsystem;

[0043] The working method combines fault tree and rule reasoning, and transforms the fault tree into reasoning rules according to its composition;

[0044] The fault source is searched using a reverse reasoning strategy with the fault probability as the search order.

[0045] Preferably, the specific content of step S4 is:

[0046] The fault tree verifies the confidence level of the probability of fault occurrence through real-time data and prediction data in each system;

[0047] After verification, if the prediction result is judged to be true or a fault occurs but the fault source is different, a fault warning verified by the fault tree is output to accurately locate the fault source;

[0048] If it is judged that no fault has occurred, the fault prediction record will be temporarily retained. When the same fault prediction record appears multiple times within a certain period of time and the fault tree judges that there is no fault, the fault warning is still output. If it is verified that a new fault has occurred, the new fault type, cause and probability will be added to the fault tree for correction.

[0049] A fault prediction device for offshore wind turbines considering the influence of environmental factors, based on the method for predicting faults of offshore wind turbines considering the influence of environmental factors, comprises a data acquisition module, a sensor KPCA-KD fault detection model, a mechanical system fault diagnosis model considering environmental factors, a tower structure damage detection model, a TCN time convolution network prediction model, a coupling fault prediction model based on unit operation, an offshore wind turbine fault tree and a correction module;

[0050] Data acquisition module, used for the operation data of offshore wind turbines and the corresponding environmental data;

[0051] TCN temporal convolutional network prediction model is used to combine the operating data of offshore wind turbines with the corresponding environmental data. According to the divided sensor system, mechanical system and tower structure system, TCN temporal convolutional network prediction models are established to predict the operating status of each system of offshore wind turbines and input them into the corresponding sensor KPCA-KD fault detection model, the mechanical system fault diagnosis model taking into account environmental factors, and the tower structure damage detection model to predict the occurrence of faults;

[0052] A coupled fault prediction model based on unit operation is used to predict faults and classify labels for faults predicted by fault models from different systems;

[0053] Offshore wind turbine fault tree, used to verify fault predictions, analyze and classify the causes of faults, and supplement the fault tree when new fault types are detected;

[0054] The correction module is used to correct the fault prediction model and fault tree based on the unit data and environmental data collected in real time.

[0055] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method for predicting offshore wind turbine faults that takes environmental factors into consideration.

[0056] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and device for predicting faults of offshore wind turbines that take into account the influence of environmental factors. The correlation between the offshore environment and the incremental fault occurrence of different components in the wind turbine is considered, and the offshore wind turbine is divided into a control system, a mechanical system and a tower structure. Fault prediction models are established for each of them to make accurate predictions of the occurrence of offshore wind turbine faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0058] Figure 1 A schematic flow chart of a method for predicting faults of offshore wind turbines taking into account the influence of environmental factors provided by the present invention;

[0059] Figure 2A schematic diagram of the sensor KPCA-KD fault detection model process provided by the present invention;

[0060] Figure 3 A schematic diagram of a flow chart of a mechanical system fault diagnosis model taking environmental factors into account provided by the present invention;

[0061] Figure 4 This is a schematic diagram of the tower structure damage detection model process provided by the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0063] The embodiment of the present invention discloses a method for predicting faults of offshore wind turbines taking into account the influence of environmental factors, such as Figure 1 , including the following steps:

[0064] S1. Offshore wind turbines are divided into sensor systems, mechanical systems and tower structure systems according to the correlation between their failure rates and environmental factors, and different fault prediction models are established according to the characteristics of each system, including the sensor KPCA-KD fault detection model, the mechanical system fault diagnosis model taking environmental factors into account and the tower structure damage detection model;

[0065] S2. Combine the operating data of the offshore wind turbine with the corresponding environmental data, and establish TCN time convolution network prediction models according to the divided systems to predict the operating status of each system of the offshore wind turbine and input them into the corresponding fault prediction model to predict the occurrence of faults;

[0066] S3. Couple the fault prediction models of each system, build a coupled fault prediction model based on unit operation and a fault tree for offshore wind turbines, and classify the faults predicted by the fault models from different systems;

[0067] S4. Verify the fault prediction through the fault tree of offshore wind turbines, analyze and classify the causes of the faults, and supplement the fault tree when new fault types are detected;

[0068] S4. Collect unit data and environmental data in real time to modify fault prediction models and fault trees.

[0069] In this embodiment, the offshore wind turbine data includes SCADA data and vibration data obtained by sensors, and the offshore environment monitoring data where the unit is located includes wind speed, wind direction, wave height, wave speed, salt spray concentration, temperature, seawater salinity, dissolved oxygen content, pH value, etc.

[0070] In order to further implement the above technical solution, the specific content of step S2 is:

[0071] In different systems, environmental data are extracted using the ReliefF algorithm to select environmental factor features with greater correlation as input variables;

[0072] By combining the historical operating data and corresponding environmental data of each system, a TCN prediction model is constructed to predict the operating status parameters of each system, and the parameters are input into the corresponding fault prediction model as input to perform fault diagnosis of the unit.

[0073] TCN is an improvement on the traditional convolutional neural network. It retains the powerful feature extraction ability of convolutional neural network. Through convolution operation, it can be performed on the entire time series at the same time with good parallel processing ability. In addition, by using multiple convolutional layers and a sufficiently long historical input window, TCN can learn and capture long-term dependencies in time series.

[0074] TCN consists of three structures: causal convolution for time series, dilated convolution for historical data memory, and residual module to eliminate gradient disappearance / explosion.

[0075] Causal Convolution:

[0076]

[0077] Dilated Convolution:

[0078]

[0079] Among them, d represents the expansion coefficient of each layer, k is the convolution kernel size, sd·i is the index of the previous time series information, and a double hidden layer structure can be used. From the input layer to the output layer, d takes the values ​​of 1, 2, and 4 respectively;

[0080] Residual Connection:

[0081] o=Activation[x+F(x)]

[0082] Specifically, the branch result F(x) after a series of transformations F is added to the module initial input x as the output of the residual module.

[0083] In order to further implement the above technical solution, Figure 2 ,The specific content of the sensor KPCA-KD fault detection model is:

[0084] Construct a KPCA model from the fault-free operation data of an offshore wind turbine and calculate the non-linear principal component score T1;

[0085] Specifically, through the non-linear mapping Φ, map the sample data set X from the input space to the high-dimensional feature space F to obtain the mapping matrix Φ(X), and perform PCA dimensionality reduction in the high-dimensional feature space;

[0086] Denote the covariance matrix of Φ(X) as C F , solve the eigenvector V L and eigenvalue λ:

[0087]

[0088] Among them,

[0089]

[0090] That is, there exists a coefficient set α i , such that V L is linearly represented by all samples in the high-dimensional feature space F;

[0091] Introduce the kernel matrix K, there is:

[0092] K i，j =(Φ(x i )·Φ(x j ))

[0093] Nλα = Kα

[0094] Among them, α is the eigenvector; so the eigenvector V of C F can be obtained from the eigenvector α of the kernel matrix K L ; the eigenvalues of the matrix K are denoted as λ1≥λ2≥…≥λ M , α1, α2, …α M are its corresponding eigenvectors, and the cumulative contribution rate method is used to determine the number of non-linear principal components A (A < M), and the cumulative contribution rate is set to 85%, so as to retain the first A eigenvalues;

[0095] Calculate the projection of the mapping matrix Φ(X) on the eigenvector V L , that is, the non-linear principal component score T1 of the mapped data:

[0096] T1 = Φ(X) T V L

[0097] Among them, a = 1, 2, ···, A;

[0098] Substitute the predicted unit operation data into the KPCA model, calculate the nonlinear principal component score T2 at this time, and calculate the KD value between T1 and T2;

[0099]

[0100] Among them, μ a , μ b represents the average value of T1 and T2, ∑ a ,∑ b Represent the covariance matrices of T1 and T2 respectively;

[0101] The kernel density estimation method is used to calculate the detection threshold in the KD scenario:

[0102]

[0103] Among them, d i is the ith observation, h is the smoothing bandwidth parameter, which is 14, a is the sample standard deviation, n is the number of samples, and K is the kernel function;

[0104] Compare the KD value between T1 and T2 with the detection threshold to determine the occurrence of sensor failure;

[0105] In this embodiment, common faults of sensors in wind turbines include bias faults, drift faults, freezing faults, precision loss or degradation faults;

[0106] The offshore wind turbine database is updated through equipment operation data, and the KPCA model and detection threshold are corrected.

[0107] In practical applications, since sensor data is typical time series data, in order to ensure that the model can extract effective fault features, improve detection accuracy, and avoid the impact of noise data on the training model, it is necessary to fill in missing data and denoise the sensor data, and standardize the data.

[0108] In this embodiment, for the mechanical system, the influence of the offshore environment on the unit components is considered, the model parameters are solved by the particle swarm optimization algorithm, the failure rate model is constructed, and the influence of the equipment failure rate after the preventive maintenance is coupled to predict the remaining life of the equipment. After combining with the corrosion model, the occurrence of environmental mechanical system failure can be predicted; since the blades of the offshore wind turbine are affected by salt spray, ultraviolet rays and humidity for a long time, and because of the high linear velocity of the blade tip, the leading edge blade corrosion damage is the most serious, and the leading edge corrosion of the blade will lead to an increase in the airfoil resistance, which will eventually lead to the loss of power generation of the unit. In severe cases, corrosion may even cause accidents such as cracking and fracture of the leading edge of the blade; the offshore environmental factors mainly include wind speed, wind direction, wave height, wave speed, salt spray concentration, temperature and air pressure; the mechanical system of the offshore wind turbine mainly includes transmission components such as gears, rotors, bearings and blades, and its failures include bearing damage or wear, gear fracture or surface bonding, blade damage or fracture, etc.

[0109] In order to further implement the above technical solution, Figure 3 , the specific content of the mechanical system fault diagnosis model taking environmental factors into account is:

[0110] The ReliefF algorithm is used to select the optimal characteristic value of the environmental factors that affect the failure rate of the mechanical system;

[0111] In this embodiment, the ReliefF algorithm has the advantages of high operating efficiency and excellent handling of multi-category problems and regression problems in high-order data;

[0112] Construct a mechanical system failure rate model that takes environmental influences into account, and estimate model parameters based on maximum likelihood method and particle swarm algorithm;

[0113] Specifically, the mechanical system failure rate model considering environmental impact is:

[0114]

[0115] Where λ(t, X) is the equipment failure rate, X is the external environmental factor that affects the equipment operation status, X = (x1, x2, ..., x p ), p is the number of environmental factors, γ is the coefficient of environmental factors, γ=(γ1,γ2,…,γ p );

[0116] Then the reliability function R(t, X) and failure probability density function f(t, X) of the equipment can be obtained as follows:

[0117]

[0118]

[0119] The particle swarm algorithm is used to optimize and estimate the unknown parameters η, β, and γ, and the log-likelihood function is used as the fitness function of the model. The larger the value of the fitness function, the closer the parameter estimation value obtained by the particle swarm algorithm is to the true value of the parameter;

[0120] In the actual maintenance of equipment, although preventive maintenance can improve equipment performance and reliability, it is difficult to completely restore the performance of the equipment to its original state. Therefore, it is essentially an incomplete maintenance method.

[0121] Combined with the impact of equipment preventive maintenance on the failure rate, the equipment degradation process is modeled to predict the remaining life of the equipment;

[0122] Specifically, let the failure rate of the equipment after the i-th preventive maintenance be i (t,X) is:

[0123] λ i (t,X)=b i-1 ×λ i-1 (t+a i-1 T i-1 , X)

[0124] Among them, b i-1 is the failure rate increasing factor of the equipment after the i-th preventive maintenance, a i-1 is the service age regression factor of the equipment after the i-th preventive maintenance;

[0125]

[0126]

[0127] Then, by combining the above formulas, we can obtain the recursive relationship between the failure rate after the i-th preventive maintenance and the failure rate before the maintenance.

[0128] The expected value of the remaining life of the equipment at time t is used as the predicted value of the remaining life, which is:

[0129]

[0130] Where RUL(t, X) is the remaining life of the equipment at time t;

[0131] Considering the corrosion effect of offshore environment on wind turbine blades, a blade corrosion model is constructed;

[0132] The relationship between corrosion thickness d(t) and time can be expressed as:

[0133]

[0134] Where d is the final corrosion thickness of the component, d = 10 mm, Tst is the time when the component corrosion starts, that is, the corrosion protection system has failed, Tst = 2a, β and η are shape factors, β = 2.5 and η = 19;

[0135] Combining the corrosion model and the remaining life prediction value, when the remaining life prediction value or the corrosion degree exceeds the preset value, a fault warning is issued to the mechanical system;

[0136] The mechanical system fault diagnosis model is corrected by combining real-time environmental monitoring data and real-time unit operation status monitoring data.

[0137] In this embodiment, for the tower structure, considering the offshore environmental factors, vibration analysis technology is used, that is, by comparing the natural frequencies of the intact structure and the damaged structure, a tower fault diagnosis model is established, which can predict the damage degree and damage location of the tower structure.

[0138] In order to further implement the above technical solution, Figure 4 , the specific content of the tower structure damage detection model is:

[0139] The K-means data clustering method is used to classify the intact tower structure and damaged tower structure in the tower database of offshore wind turbines, and the frequency change rate NFCR and square change rate NSFR of the natural frequency of the intact tower structure and the damaged tower structure are normalized.

[0140] Specifically, the frequency change rate of the structure is:

[0141]

[0142] where ω i is the initial natural frequency of the structure, ωj is the natural frequency after the structure is damaged, and its rate of change is related to the degree and location of damage to the structure, which can also be expressed as:

[0143] FCR = g(r)f(ΔK, ΔM)

[0144] Where r is the vector of the structural damage position; the Taylor series expansion of each system eigenvalue pair f about ΔK = 0 and ΔM = 0, and ignoring the high-order terms, is:

[0145]

[0146] Since f(0,0)=0, we have:

[0147]

[0148] Among them, the partial derivative of function f at ΔK = 0 and ΔM = 0 is a constant, which is:

[0149] FCR=ΔKn(r)+ΔMm(r)

[0150] In fact, structural damage mainly refers to the change of structural stiffness. Therefore, assuming that = 0, we get:

[0151] FCR=ΔKn(r)

[0152] At this time, NFCR has:

[0153]

[0154] Where m is the frequency order of the structure under test;

[0155] The NSFR is used to identify the extent of damage and is:

[0156]

[0157] Among them, {φ i} is ω i The corresponding eigenvector;

[0158] The processed data is used as the neural network input parameter through the BP neural network to build a diagnostic model of the damage location and damage degree of the offshore wind power tower;

[0159] The tower structure database is supplemented by the real-time data of the tower structure, and the tower structure damage detection model is corrected.

[0160] In this embodiment, according to the divided systems, based on actual engineering experience and the unit operation database, it is preliminarily determined that each system contains an initial fault feature set of 9 fault features (including wind speed, wind direction, wave height, wave speed, salt spray concentration, temperature, seawater salinity, dissolved oxygen content, pH value, etc.), and the feature weights are calculated using the ReliefF algorithm:

[0161] First, the prediction weight of the feature is initialized to 0, and then a sample R is randomly selected from the data set. K samples similar to R are found from the same category and different categories, and H is used to i and M i Indicates that, finally, the feature weight calculation formula is used to iteratively update each feature weight. The feature weight calculation formula is:

[0162]

[0163] Where W(A) and W'(A) are the feature weights before and after iterative updating, respectively; n is the total number of samples, i = 1, 2, ···, n; Class(R) is the R sample type, C is the C class sample, and M i(C) is the C-type sample among the samples of different classes that are closest to sample R, P[Class(R)] is the proportion of R sample type to the total number of samples, H i With M i are the samples of the same type and different types that are closest to sample R in the training set, m is the number of samplings, K is the number of neighboring samples, P C is the ratio of the number of samples in the Cth category to the total number of samples, diff(A,R,H i ) are samples R and H i Regarding the difference of feature A, feature A is represented by a k-dimensional array, A = {a1, a2, ..., a k}, the sample characteristic difference calculation formula is:

[0164]

[0165] The first four feature weight values ​​are selected as the effective fault feature set, and 80% of the original data set is randomly selected as the training data set to train the model, and the remaining 20% ​​is used as the test data set to verify the model. Take the tower unit as an example: the corresponding environmental factors and unit operation data are selected as training data, and the vibration frequency of the tower is used as the simulation output, which is input into the tower fault to judge the occurrence of the fault.

[0166] Due to the complexity of wind turbines and the harsh operating environment, when diagnosing wind turbine faults, one fault source often triggers multiple alarms, and the fault source cannot be directly determined. Traditional fault diagnosis methods use data collected by a SCADA (supervisory control and data acquisition) system to perform health status assessment and fault warning; however, due to the imperfect alarm function of the SCADA system and the low level of fault analysis of on-site operation and maintenance personnel, repetitive faults and the safety hazards behind them are difficult to discover and handle in a timely manner. Therefore, this embodiment couples the prediction models of each system to construct a coupled fault prediction model based on unit operation and a fault tree for offshore wind turbines.

[0167] Among them, the ReliefF algorithm itself has the advantages of processing multi-classification problems, robustness to noise, high efficiency and high accuracy. It performs well in eliminating redundant features. By assigning different weights to different features, unimportant features can be effectively removed, thereby improving the accuracy of the model while reducing calculations. Therefore, in this embodiment, the weights of the characteristic values ​​of each system are calculated by the ReliefF algorithm through classification processing, thereby achieving accurate judgment of unit fault diagnosis.

[0168] In order to further implement the above technical solution, the specific content of step S3 is:

[0169] The operating data and corresponding environmental data of offshore wind turbines are selected to form a training data set for the fault scenarios of each system of the unit and the multiple fault scenarios between systems, and the fault scenarios are labeled with scenario labels;

[0170] The ReliefF algorithm is used to calculate the weights of each system eigenvalue under different scenarios;

[0171] The characteristic values ​​of each system include vibration data of the tower structure, operation data of the sensor, and relevant data of the mechanical system;

[0172] The weights of the characteristic values ​​of each system under different fault scenarios are determined using the fuzzy C-means clustering algorithm FCM to determine the cluster center and its membership degree, and the corresponding fault scenario label is marked on each cluster center;

[0173] When the model makes a fault prediction, it determines the membership of each scenario by calculating the weight data of the characteristic values ​​of each system, divides the predicted fault into sensors, mechanical systems and tower structures according to the fault scenario labels, and verifies it by the fault tree.

[0174] In order to further implement the above technical solution, the specific contents of constructing the fault tree of offshore wind turbines are as follows:

[0175] Based on the top event, the fault tree of offshore wind turbines is divided into pitch subsystem failure, transmission chain subsystem failure, generator subsystem failure, yaw subsystem failure, control subsystem failure, converter subsystem failure and tower structure failure;

[0176] The deductive method is used to analyze the subsystem failures of the offshore wind turbine fault tree from top to bottom, and the causes of the failures are analyzed step by step and refined to the components in the subsystem;

[0177] The working method combines fault tree and rule reasoning. According to the composition of the fault tree, it is transformed into reasoning rules. The reasoning rules include the credibility of the event occurrence, etc.

[0178] The fault source is searched using a reverse reasoning strategy with the fault probability as the search order.

[0179] In order to further implement the above technical solution, the specific content of step S4 is:

[0180] The fault tree verifies the confidence level of the probability of fault occurrence through real-time data and prediction data in each system;

[0181] After verification, if the prediction result is judged to be true or a fault occurs but the fault source is different, a fault warning verified by the fault tree is output to accurately locate the fault source;

[0182] If it is judged that no fault has occurred, the fault prediction record will be temporarily retained. When the same fault prediction record appears multiple times within a certain period of time and the fault tree judges that there is no fault, the fault warning is still output. If it is verified that a new fault has occurred, the new fault type, cause and probability will be added to the fault tree for correction.

[0183] A fault prediction device for offshore wind turbines considering the influence of environmental factors, based on a fault prediction method for offshore wind turbines considering the influence of environmental factors, including a data acquisition module, a sensor KPCA-KD fault detection model, a mechanical system fault diagnosis model considering environmental factors, a tower structure damage detection model, a TCN time convolution network prediction model, a coupling fault prediction model based on unit operation, an offshore wind turbine fault tree and a correction module;

[0184] Data acquisition module, used for the operation data of offshore wind turbines and the corresponding environmental data;

[0185] TCN temporal convolutional network prediction model is used to combine the operating data of offshore wind turbines with the corresponding environmental data. According to the divided sensor system, mechanical system and tower structure system, TCN temporal convolutional network prediction models are established to predict the operating status of each system of offshore wind turbines and input them into the corresponding sensor KPCA-KD fault detection model, the mechanical system fault diagnosis model taking into account environmental factors, and the tower structure damage detection model to predict the occurrence of faults;

[0186] A coupled fault prediction model based on unit operation is used to predict faults and classify labels for faults predicted by fault models from different systems;

[0187] Offshore wind turbine fault tree, used to verify fault predictions, analyze and classify the causes of faults, and supplement the fault tree when new fault types are detected;

[0188] The correction module is used to correct the fault prediction model and fault tree based on the unit data and environmental data collected in real time.

[0189] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method for predicting faults of offshore wind turbines that takes environmental factors into consideration.

[0190] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0191] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest range of system characteristics consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting faults of offshore wind turbines considering the influence of environmental factors, characterized in that: The following steps are involved: S1. Offshore wind turbines are divided into sensor systems, mechanical systems and tower structure systems according to the correlation between their failure rates and environmental factors, and different fault prediction models are established according to the characteristics of each system, including the sensor KPCA-KD fault detection model, the mechanical system fault diagnosis model taking environmental factors into account and the tower structure damage detection model; S2. Combine the operating data of the offshore wind turbine with the corresponding environmental data, and establish TCN time convolution network prediction models according to the divided systems to predict the operating status of each system of the offshore wind turbine and input them into the corresponding fault prediction model to predict the occurrence of faults; S3. Couple the fault prediction models of each system, build a coupled fault prediction model based on unit operation and a fault tree for offshore wind turbines, and classify the faults predicted by the fault models from different systems; S4. Verify the fault prediction through the fault tree of offshore wind turbines, analyze and classify the causes of the faults, and supplement the fault tree when new fault types are detected; S4. Collect unit data and environmental data in real time to modify fault prediction models and fault trees.

2. A method for predicting offshore wind turbine faults taking environmental factors into consideration according to claim 1, characterized in that: The specific content of step S2 is: In different systems, environmental data are extracted using the ReliefF algorithm to select environmental factor features with greater correlation as input variables; By combining the historical operating data and corresponding environmental data of each system, a TCN prediction model is constructed to predict the operating status parameters of each system, and the parameters are input into the corresponding fault prediction model as input to perform fault diagnosis of the unit.

3. The method for predicting offshore wind turbine faults taking into account environmental factors according to claim 1, characterized in that: The specific content of the sensor KPCA-KD fault detection model is: The KPCA model is constructed based on the trouble-free operation data of offshore wind turbines, and the nonlinear principal component score T1 is calculated; Substitute the predicted unit operation data into the KPCA model, calculate the nonlinear principal component score T2 at this time, and calculate the KD value between T1 and T2; The kernel density estimation method is used to calculate the detection threshold in the KD scenario; Compare the KD value with the detection threshold to determine the occurrence of sensor failure; The offshore wind turbine database is updated through equipment operation data, and the KPCA model and detection threshold are corrected.

4. The method for predicting offshore wind turbine faults taking into account environmental factors according to claim 1, characterized in that: The specific contents of the mechanical system fault diagnosis model taking environmental factors into account are: The ReliefF algorithm is used to select the optimal characteristic value of the environmental factors that affect the failure rate of the mechanical system; Construct a mechanical system failure rate model that takes environmental influences into account, and estimate model parameters based on maximum likelihood method and particle swarm algorithm; Combined with the impact of equipment preventive maintenance on the failure rate, the equipment degradation process is modeled to predict the remaining life of the equipment; Considering the corrosion effect of offshore environment on wind turbine blades, a blade corrosion model is constructed; Combining the corrosion model and the remaining life prediction value, when the remaining life prediction value or the corrosion degree exceeds the preset value, a fault warning is issued to the mechanical system; The mechanical system fault diagnosis model is corrected by combining real-time environmental monitoring data and real-time unit operation status monitoring data.

5. The method for predicting offshore wind turbine faults taking into account environmental factors according to claim 1, characterized in that: The specific contents of the tower structure damage detection model are as follows: According to the tower database of offshore wind turbines, the intact tower structure and the damaged tower structure are classified, and the frequency change rate and square change rate of the natural frequency of the intact tower structure and the damaged tower structure are normalized; The processed data is used as the neural network input parameter through the BP neural network to build a diagnostic model of the damage location and damage degree of the offshore wind power tower; The tower structure database is supplemented by the real-time data of the tower structure, and the tower structure damage detection model is corrected.

6. The method for predicting offshore wind turbine faults taking environmental factors into consideration according to claim 1, characterized in that: The specific content of step S3 is: The operating data and corresponding environmental data of offshore wind turbines are selected to form a training data set for the fault scenarios of each system of the unit and the multiple fault scenarios between systems, and the fault scenarios are labeled with scenario labels; The ReliefF algorithm is used to calculate the weights of each system eigenvalue under different scenarios; The weights of the characteristic values ​​of each system under different fault scenarios are determined using the fuzzy C-means clustering algorithm to determine the cluster center and its membership degree, and the corresponding fault scenario label is marked on each cluster center; When the model makes a fault prediction, it determines the membership of each scenario by calculating the weight data of the characteristic values ​​of each system, divides the predicted fault into sensors, mechanical systems and tower structures according to the fault scenario labels, and verifies it by the fault tree.

7. The method for predicting offshore wind turbine faults taking environmental factors into consideration according to claim 1, characterized in that: The specific contents of constructing the fault tree of offshore wind turbines are: Based on the top event, the fault tree of offshore wind turbines is divided into pitch subsystem failure, transmission chain subsystem failure, generator subsystem failure, yaw subsystem failure, control subsystem failure, converter subsystem failure and tower structure failure; The deductive method is used to analyze the subsystem failures of the offshore wind turbine fault tree from top to bottom, and the causes of the failures are analyzed step by step and refined to the components in the subsystem; The working method combines fault tree and rule reasoning, and transforms the fault tree into reasoning rules according to its composition; The fault source is searched using a reverse reasoning strategy with the fault probability as the search order.

8. The method for predicting offshore wind turbine faults taking environmental factors into consideration according to claim 1, characterized in that: The specific content of step S4 is: The fault tree verifies the confidence level of the probability of fault occurrence through real-time data and prediction data in each system; After verification, if the prediction result is judged to be true or a fault occurs but the fault source is different, a fault warning verified by the fault tree is output to accurately locate the fault source; If it is judged that no fault has occurred, the fault prediction record will be temporarily retained. When the same fault prediction record appears multiple times within a certain period of time and the fault tree judges that there is no fault, the fault warning is still output. If it is verified that a new fault has occurred, the new fault type, cause and probability will be added to the fault tree for correction.

9. A fault prediction device for offshore wind turbines taking into account the influence of environmental factors, characterized in that: A method for predicting faults of offshore wind turbines taking into account the influence of environmental factors based on any one of claims 1 to 8, comprising a data acquisition module, a sensor KPCA-KD fault detection model, a mechanical system fault diagnosis model taking into account environmental factors, a tower structure damage detection model, a TCN time convolution network prediction model, a coupled fault prediction model based on unit operation, an offshore wind turbine fault tree, and a correction module; Data acquisition module, used for the operation data of offshore wind turbines and the corresponding environmental data; TCN temporal convolutional network prediction model is used to combine the operating data of offshore wind turbines with the corresponding environmental data. According to the divided sensor system, mechanical system and tower structure system, TCN temporal convolutional network prediction models are established to predict the operating status of each system of offshore wind turbines and input them into the corresponding sensor KPCA-KD fault detection model, the mechanical system fault diagnosis model taking into account environmental factors, and the tower structure damage detection model to predict the occurrence of faults; A coupled fault prediction model based on unit operation is used to predict faults and classify labels for faults predicted by fault models from different systems; Offshore wind turbine fault tree, used to verify fault predictions, analyze and classify the causes of faults, and supplement the fault tree when new fault types are detected; The correction module is used to correct the fault prediction model and fault tree based on the unit data and environmental data collected in real time.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting faults of offshore wind turbines taking into account the influence of environmental factors as described in any one of claims 1 to 8 is implemented.

Images