Wind turbine generator fault tracking and redisk method based on multilayer data playback

Abstract

The invention discloses a wind turbine generator fault tracking and redisk method based on multilayer data playback, and the method comprises the steps: S1, obtaining a wind turbine generator operation data set, and carrying out the preprocessing of the wind turbine generator operation data set; s2, dividing multi-level data according to the time dimension, the space dimension and the equipment module of the wind turbine generator operation data set; s3, constructing a multi-layer data playback model based on the multi-layer data; s4, generating a causal chain of the operation parameters of the wind turbine generator, and establishing a causal inference analysis model; s5, analyzing the multi-level data playback result by using a causal inference analysis model, tracking the occurrence time point, the influence factor and the evolution process of the fault of the wind turbine generator, and positioning the root of the fault; and S6, generating a wind turbine generator fault recovery report. According to the invention, the intelligence and reliability of wind turbine generator maintenance are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine generator systems, and in particular to a wind turbine generator system fault tracing and replay method based on multi-layer data playback. Background Art

[0002] During the operation of wind turbines, various faults are prone to occur due to long-term exposure to complex natural environments and high-intensity mechanical loads. Faults may be caused by multiple factors such as mechanical wear, environmental changes, electrical failures, and human operational errors. They have complex coupling relationships and multi-dimensional characteristics. Once a fault occurs, it will not only affect the power generation efficiency of the wind turbine but may also lead to high maintenance costs or even serious safety accidents. Therefore, timely and accurate fault tracking and review are important technical requirements to ensure the safe operation of wind turbines.

[0003] In the existing technology, the fault diagnosis method of wind turbines usually relies on a single-level data monitoring and analysis method. The existing method can preliminarily locate specific types of faults to a certain extent, but due to the multi-level coupling characteristics of the wind turbine system, a single data source is difficult to fully reflect the entire process of the fault. In addition, traditional fault analysis methods often use empirical models or statistical methods to infer faults, which are easily affected by insufficient data or noise interference, making it difficult to identify the source of the fault.

[0004] Another common technology is the fault prediction model based on historical data. The fault prediction model based on historical data uses the historical operating data of the wind turbine to establish a regression model or a machine learning algorithm to try to predict possible future faults. However, there are the following problems: First, there is a lack of multi-level data integration and only relies on a single type of historical data, which cannot fully utilize the correlation information between the data; second, the existing prediction model is difficult to handle the complex cause-and-effect relationship in the dynamic fault evolution process, resulting in low accuracy of the prediction results; third, after the fault occurs, the existing technology usually only focuses on the repair of the fault point, lacks the review and pattern extraction of the entire fault process, and it is difficult to provide systematic support for subsequent optimization strategies.

[0005] In addition, most of the current systems for wind turbine fault analysis rely on engineers' subjective experience and judgment, lack an automated fault tracking and review mechanism, and only rely on a single level of fault analysis in complex faults caused by multiple factors. The source of the fault may be unclear or key influencing factors may be missed, resulting in inefficient troubleshooting and delayed maintenance decisions.

[0006] In summary, the existing technology has the following defects in wind turbine fault analysis and recovery:

[0007] 1. The fault analysis method is single and cannot integrate multi-level data for comprehensive analysis;

[0008] 2. Traditional models are not capable of processing dynamic causal relationships, making it difficult to accurately infer the fault evolution path;

[0009] 3. Lack of systematic fault review mechanism, unable to extract fault occurrence mode and guide subsequent optimization;

[0010] 4. There is a high reliance on engineers, a low degree of automation, and insufficient troubleshooting efficiency and reliability. Summary of the invention

[0011] One object of the present invention is to propose a wind turbine fault tracing and replay method based on multi-layer data playback, which improves the intelligence and reliability of wind turbine maintenance.

[0012] A wind turbine fault tracing and replay method based on multi-layer data playback according to an embodiment of the present invention includes the following steps:

[0013] S1. Obtaining a wind turbine operation data set and preprocessing the wind turbine operation data set;

[0014] S2. Divide the multi-level data according to the time dimension, space dimension and equipment module of the wind turbine operation data set;

[0015] S3. Construct a multi-layer data playback model based on multi-layer data;

[0016] S4. Based on the multi-layer data playback model, the causal relationship between multi-dimensional data is analyzed using the causal inference algorithm to generate the causal chain of wind turbine operating parameters and establish a causal inference analysis model;

[0017] S5. Use the causal inference analysis model to analyze the multi-level data playback results, track the occurrence time, influencing factors and evolution process of wind turbine failures, and locate the root cause of the failure;

[0018] S6. Based on the multi-layer data playback results and fault location analysis, the entire process of wind turbine failure is reviewed, the fault occurrence mode is extracted, and a wind turbine failure review report is generated.

[0019] Optionally, the S1 includes:

[0020] S11. Obtain the wind speed V of the wind turbine w , power generation P, speed ω, current I and voltage U, construct the wind turbine operation data set, and preliminarily organize the wind turbine operation data set according to the time series T to form the time series data set D t :

[0021] Dt ={(V w (t i ),P(t i ),ω(t i ),I(t i ),U(t i ))∣t i ∈T};

[0022] Among them, t i represents the data record at the i-th time point, and n is the number of sampling points;

[0023] S12, using the sliding mean filtering method to filter the time series data set D t Perform smoothing to filter out random noise and generate the denoised wind turbine operation data set D t_smooth ;

[0024] S13. Running data set D for denoised wind turbines t_smooth The parameters in are normalized so that their range is mapped to the interval [0,1];

[0025] S14. For the missing values ​​in the normalized wind turbine operation data set, a time interpolation method is used to fill them, and a preprocessed wind turbine operation data set D is generated. processed :

[0026] D processed ={(V ″ w (t i ),P ″ (t i ),ω ″ (t i ),I ″ (t i ),U ″ (t i ))∣t i ∈T};

[0027] Among them, V ″ w (t i ),P ″ (t i ),ω ″ (t i ),I ″ (t i ),U ″ (t i ) represents the parameter data after denoising, normalization and missing value filling.

[0028] Optionally, S2 includes:

[0029] S21, based on the preprocessed wind turbine operation data set D processed Define the time window length Δt and divide the time series T into m time periods:

[0030]

[0031] T k =[t k-1 ,t k );

[0032] t k =t0+k·Δt;

[0033] Among them, t0 is the starting time, t k is the end time of the kth time layer, t end is the end time;

[0034] S22. For each time period T k Extract the corresponding operation data to form a multi-level data set in the time dimension

[0035]

[0036] S23, based on the preprocessed wind turbine operation data set D processed Define a spatial position set S = {s1, s2, ..., s p}, where s p represents the pth physical position or component of the wind turbine. For each spatial position s p Extract the operating data of the corresponding location to form a multi-level data set in spatial dimension

[0037]

[0038] Among them, s i At time point t i The location of the sensors that collect the data;

[0039] S24, based on the preprocessed wind turbine operation data set D processed Define the device module set M = {m1,m2,…,m q}, where m q represents the qth functional module of the wind turbine. For each equipment module m q Extract the operation data of the corresponding module to form a multi-level data set of equipment modules

[0040]

[0041] Among them, mi At time point t i The device module to which the collected data belongs;

[0042] S25. Multi-level datasets with time dimension Multi-level datasets in spatial dimension and equipment module multi-level dataset Integrate to form a multi-level dataset

[0043]

[0044] in, Represents a set of data that meets the conditions at the time layer k, space layer l, and device module layer j at the same time.

[0045] Optionally, the S3 includes:

[0046] S31, based on multi-level data sets For each time period T k Using time dimension multi-level dataset Constructing the time state matrix

[0047]

[0048] The state transfer equation of the time dimension is established. The state transfer equation of the time dimension describes the dynamic behavior of the wind turbine in time:

[0049]

[0050] in, is the state transfer matrix in the time dimension, is the control input matrix of the time dimension, is the control input vector of the time dimension, is the process noise vector in the time dimension;

[0051] Use the state transfer equation to calculate the time period T k The state in the time is replayed time by time to generate a time-by-time replay model

[0052] S32, based on multi-level datasets For each spatial position s l Exploiting spatial dimensions of multi-level datasets Constructing the space state matrix

[0053]

[0054] Establish the state propagation equation of the spatial dimension, which describes the mutual influence between different spatial positions:

[0055]

[0056] in, is the spatial coupling matrix, representing the spatial position s m For spatial position s l The influence weight of (l,m) is the spatial position s m For spatial position s l The propagation delay of is the process noise vector of spatial dimension;

[0057] Use the state propagation equation to perform distributed playback of the states of each spatial position to generate a distributed playback model

[0058] S33, by replaying the model hour by hour and distributed playback model Constructing the global state matrix Integrate data from time, space and device module dimensions:

[0059]

[0060] in, is the device module m j The state matrix of

[0061] A global state transfer equation is established to describe the comprehensive dynamic behavior of wind turbines in multiple dimensions:

[0062]

[0063] in, is the global state transfer matrix, is the global control input matrix, is the global control input vector, is the global process noise vector;

[0064] The global state transfer equation is used to replay the global state of the wind turbine to generate a global data replay model M. global .

[0065] Optionally, the S4 includes:

[0066] S41, based on global data playback model M global Extract multi-dimensional operating parameter data and construct a causal analysis data set D causal ;

[0067] S42. Causal analysis data set D causal Perform preprocessing and define the random variable set X(t);

[0068] S43, using causal inference algorithm to determine the variable X(t) using Granger causality test method i (t) Is it a variable X? j Granger causes of (t), where X i (t),X j (t)∈X(t);

[0069] Construct a univariate autoregressive model:

[0070]

[0071] Among them, X j (t) is the target variable, N is the lag order, Δt is the time step, a jn is the coefficient of the autoregressive model, describing X j (t) Dependence on its own historical value, ∈ j (t) is the model residual;

[0072] Combined with the dependent variable X i (t) for the target variable X j (t) constructs a multivariate causal regression model:

[0073]

[0074] Among them, b jn represents the target variable X j The autoregressive coefficient of (t), c in Denotes the dependent variable X i (t) for the target variable X j The causal influence strength of (t), ∈′ j (t) is the model residual;

[0075] Construct the causal test objective function and calculate the residual sum of squares RSS1 and RSS2 of the univariate model and the multivariate model:

[0076]

[0077] Calculate the F statistic for the Granger causality test ij :

[0078]

[0079] Where T is the number of time series samples, F ij Denotes the dependent variable X i (t) Whether it significantly affects the target variable Xj The Granger causal strength of (t);

[0080] Find the critical value F of the F distribution according to the preset significance level α α , if F ij >F α , then the dependent variable X i (t) is the target variable X j Granger cause of (t), there is a causal relationship;

[0081] S44. Based on the causal relationship analysis results, a causal chain C of the wind turbine operating parameters is generated. The causal chain C represents the causal path between the operating parameters, where the arrows represent the causal influence direction:

[0082] C={(X i →X j )|F ij >F α};

[0083] S45. Establishing causal inference analysis model M causal The causal inference analysis model consists of a causal chain C and the corresponding causal strength F ij It is used to analyze and locate the causes of wind turbine failures.

[0084] Optionally, the S5 includes:

[0085] S51. Construct fault-related operating parameter set X fault (t), based on the causal inference model M causal The causal chain C extracts the target variable X j The causal path of (t):

[0086]

[0087] S52, fault-related operating parameter set X fault (t) Each dependent variable X i (t) Calculate its rate of change ΔX i (t):

[0088]

[0089] Among them, Δt1 is the time step, γ ik For other parameters X k (t) for X i (t) The cross-influence factor describes the dynamic coupling between the internal parameters of the wind turbine;

[0090] S53, based on the rate of change ΔX i (t) passes the fault trigger threshold θ iDetermine the fault triggering time point set T fault :

[0091]

[0092] Among them, t f Indicates the time point when the trigger condition is met. The max operation selects the parameter with the most significant change as the sign of fault triggering;

[0093] S54, fault triggering time point set T fault At each time point t f Review of its key impact factor set I fault (t f );

[0094] S55, based on key impact factor set I fault (t f ) and causal inference model M causal , combined with the playback result M global Generate fault evolution process E fault :

[0095] E fault ={(t f ,I fault (t f ),X j (t))|t f ∈T fault ,X j (t) is the target fault parameter};

[0096] Among them, E fault Describe the occurrence time of the fault, key influencing factors and parameter evolution path;

[0097] S56, according to the fault evolution process E fault and causal chain C to locate the root cause of the wind turbine fault R fault :

[0098]

[0099] Among them, R fault It represents the set of root cause parameters obtained through causal path tracing analysis, identifying the specific fault source of the wind turbine.

[0100] Optionally, the S6 includes:

[0101] S61. Comprehensively review the failure process of the wind turbine generator system based on the multi-layer data playback model and causal inference analysis model, and extract data of all time nodes, spatial locations and equipment modules related to the failure;

[0102] S62, analyzing the set of fault triggering time points, determining the time axis of the fault occurrence, marking the change characteristics of the operating parameters in different time periods and the evolution trend of the fault to form a fault replay sequence in the time dimension;

[0103] S63. Analyze the impact of different sensor locations and equipment modules on the target fault during the fault evolution process by combining the spatial dimension multi-level data in the global data playback model, and identify the coupling relationship between spatial locations and their contribution to the fault;

[0104] S64, extracting the operation characteristics of the equipment module dimension, analyzing the role of different functional modules of the wind turbine in the occurrence and propagation of faults, and constructing a fault correlation map between modules;

[0105] S65. Summarize and simplify the causal chain in the causal inference analysis model, extract the main causal path of the fault, and generate a causal network structure that reflects the laws of fault triggering, propagation and diffusion;

[0106] S66. Based on the comprehensive analysis of time axis, space dimension, equipment module and causal path, a wind turbine fault mode library is constructed to classify the fault occurrence modes and compare them with the existing historical data to extract representative fault characteristics and evolution modes;

[0107] S67. Generate a wind turbine fault reconstructive report, which includes the following contents:

[0108] The time of occurrence of the failure and key influencing factors;

[0109] The fault propagation process and the spatial locations and equipment modules involved;

[0110] Causal paths and root cause parameters of failures;

[0111] Failure modes and failure mode characteristics;

[0112] Optimization suggestions and early warning strategies are proposed based on the review results.

[0113] The beneficial effects of the present invention are:

[0114] (1) The present invention constructs a multi-level data set in three dimensions of time, space and equipment module, and dynamically reproduces the data based on a global data playback model to fully reproduce the entire process of wind turbine failure. Compared with the existing single-level data analysis method, the multi-level data playback model can capture the correlation characteristics of faults in different dimensions, the trend changes on the time axis, the coupling relationship between spatial positions and the functional differences of equipment modules, which significantly improves the accuracy and comprehensiveness of fault location and provides reliable data support for the in-depth analysis of complex faults.

[0115] (2) The present invention generates a causal chain between the operating parameters of the wind turbine set through a causal inference algorithm, and combines the causal chain to trace and analyze the fault path, effectively overcoming the limitation of traditional statistical methods that are difficult to handle complex causal relationships. In the causal inference analysis, the improved algorithms of Granger causality test, causal network modeling and causal strength quantification are introduced, so that the occurrence, propagation and evolution process of the fault can be clearly expressed in the form of a causal path.

[0116] (3) The present invention establishes a full-process review mechanism for wind turbine failures by combining a global data playback model with a causal inference analysis model, covering multi-dimensional data analysis of time, space and equipment modules. By extracting the key time points, key influencing factors and causal paths of fault triggering, a fault review report with a global perspective is generated, and the failure mode is summarized and concluded based on the review results. Compared with the traditional method of only performing local repairs on a single fault point, the present invention can systematically identify the evolution law of wind turbine failures and provide guidance for subsequent early warning, maintenance optimization and operation strategy formulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0117] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0118] Figure 1 The present invention provides a flow chart of a method for wind turbine fault tracing and recovery based on multi-layer data playback. DETAILED DESCRIPTION

[0119] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0120] refer to Figure 1 , a box-type transformer cooling coordinated control method based on a dynamic area allocation algorithm comprises the following steps:

[0121] S1. Obtaining a wind turbine operation data set and preprocessing the wind turbine operation data set;

[0122] S2. Divide the multi-level data according to the time dimension, space dimension and equipment module of the wind turbine operation data set;

[0123] S3. Construct a multi-layer data playback model based on multi-layer data;

[0124] S4. Based on the multi-layer data playback model, the causal relationship between multi-dimensional data is analyzed using the causal inference algorithm to generate the causal chain of wind turbine operating parameters and establish a causal inference analysis model;

[0125] S5. Use the causal inference analysis model to analyze the multi-level data playback results, track the occurrence time, influencing factors and evolution process of wind turbine failures, and locate the root cause of the failure;

[0126] S6. Based on the multi-layer data playback results and fault location analysis, the entire process of wind turbine failure is reviewed, the fault occurrence mode is extracted, and a wind turbine failure review report is generated.

[0127] In this implementation, S1 includes:

[0128] S11. Obtain the wind speed V of the wind turbine w , power generation P, speed ω, current I and voltage U, construct the wind turbine operation data set, and preliminarily organize the wind turbine operation data set according to the time series T to form the time series data set D t :

[0129] D t ={(V w (t i ),P(t i ),ω(t i ),I(t i ),U(t i ))∣t i ∈T};

[0130] Among them, t i represents the data record at the i-th time point, and n is the number of sampling points;

[0131] S12, using the sliding mean filtering method to filter the time series data set D t Perform smoothing to filter out random noise and generate the denoised wind turbine operation data set D t_smooth ;

[0132] S13. Running data set D for denoised wind turbines t_smooth The parameters in are normalized so that their range is mapped to the interval [0,1];

[0133] S14. For the missing values ​​in the normalized wind turbine operation data set, a time interpolation method is used to fill them, and a preprocessed wind turbine operation data set D is generated. processed :

[0134] D processed ={(V ″ w (t i ),P ″ (t i ),ω ″ (t i),I ″ (t i ),U ″ (t i ))∣t i ∈T};

[0135] Among them, V ″ w (t i ),P ″ (t i ),ω ″ (t i ),I ″ (t i ),U ″ (t i ) represents the parameter data after denoising, normalization and missing value filling.

[0136] In this implementation, S2 includes:

[0137] S21, based on the preprocessed wind turbine operation data set D processed Define the time window length Δt and divide the time series T into m time periods:

[0138]

[0139] T k =[t k-1 ,t k );

[0140] t k =t0+k·Δt;

[0141] Among them, t0 is the starting time, t k is the end time of the kth time layer, t end is the end time;

[0142] S22. For each time period T k Extract the corresponding operation data to form a multi-level data set in the time dimension

[0143]

[0144] S23, based on the preprocessed wind turbine operation data set D processed Define a spatial position set S = {s1, s2, ..., s p}, where s p represents the pth physical position or component of the wind turbine. For each spatial position s p Extract the operating data of the corresponding location to form a multi-level data set in spatial dimension

[0145]

[0146] Among them, s i At time point t i The location of the sensors that collect the data;

[0147] S24, based on the preprocessed wind turbine operation data set D processed Define the device module set M = {m1,m2,…,m q}, where m q represents the qth functional module of the wind turbine. For each equipment module m q Extract the operation data of the corresponding module to form a multi-level data set of equipment modules

[0148]

[0149] Among them, m i At time point t i The device module to which the collected data belongs;

[0150] S25. Multi-level datasets with time dimension Multi-level datasets in spatial dimension and equipment module multi-level dataset Integrate to form a multi-level dataset

[0151]

[0152] in, Represents a set of data that meets the conditions at the time layer k, space layer l, and device module layer j at the same time.

[0153] In this implementation, S3 includes:

[0154] S31, based on multi-level data sets For each time period T k Using time dimension multi-level dataset Constructing the time state matrix

[0155]

[0156] The state transfer equation of the time dimension is established. The state transfer equation of the time dimension describes the dynamic behavior of the wind turbine in time:

[0157]

[0158] in, is the state transfer matrix in the time dimension, is the control input matrix of the time dimension, is the control input vector of the time dimension, is the process noise vector in the time dimension;

[0159] Use the state transfer equation to calculate the time period T k The state in the time is replayed time by time to generate a time-by-time replay model

[0160] S32, based on multi-level datasets For each spatial position s l Exploiting spatial dimensions of multi-level datasets Constructing the space state matrix

[0161]

[0162] Establish the state propagation equation of the spatial dimension, which describes the mutual influence between different spatial positions:

[0163]

[0164] in, is the spatial coupling matrix, representing the spatial position s m For spatial position s l The influence weight of (l,m) is the spatial position s m For spatial position s l The propagation delay of is the process noise vector of spatial dimension;

[0165] Use the state propagation equation to perform distributed playback of the states of each spatial position to generate a distributed playback model

[0166] S33, by replaying the model hour by hour and distributed playback model Constructing the global state matrix Integrate data from time, space and device module dimensions:

[0167]

[0168] in, is the device module m j The state matrix of

[0169] A global state transfer equation is established to describe the comprehensive dynamic behavior of wind turbines in multiple dimensions:

[0170]

[0171] in, is the global state transfer matrix, is the global control input matrix, is the global control input vector, is the global process noise vector;

[0172] The global state transfer equation is used to replay the global state of the wind turbine to generate a global data replay model M. global .

[0173] In this implementation, S4 includes:

[0174] S41, based on global data playback model M global Extract multi-dimensional operating parameter data and construct a causal analysis data set D causal ;

[0175] S42. Causal analysis data set D causal Perform preprocessing and define the random variable set X(t);

[0176] S43, using causal inference algorithm to determine the variable X(t) using Granger causality test method i (t) Is it a variable X? j Granger causes of (t), where X i (t),X j (t)∈X(t);

[0177] Construct a univariate autoregressive model:

[0178]

[0179] Among them, X j (t) is the target variable, N is the lag order, Δt is the time step, a jn is the coefficient of the autoregressive model, describing X j (t) Dependence on its own historical value, ∈ j (t) is the model residual;

[0180] Combined with the dependent variable X i (t) for the target variable X j (t) constructs a multivariate causal regression model:

[0181]

[0182] Among them, b jn represents the target variable X j The autoregressive coefficient of (t), c inDenotes the dependent variable X i (t) for the target variable X j The causal influence strength of (t), ∈′ j (t) is the model residual;

[0183] Construct the causal test objective function and calculate the residual sum of squares RSS1 and RSS2 of the univariate model and the multivariate model:

[0184]

[0185] Calculate the F statistic for the Granger causality test ij :

[0186]

[0187] Where T is the number of time series samples, F ij Denotes the dependent variable X i (t) Whether it significantly affects the target variable X j The Granger causal strength of (t);

[0188] Find the critical value F of the F distribution according to the preset significance level α α , if F ij >F α , then the dependent variable X i (t) is the target variable X j Granger cause of (t), there is a causal relationship;

[0189] S44. Based on the causal relationship analysis results, a causal chain C of the wind turbine operating parameters is generated. The causal chain C represents the causal path between the operating parameters, where the arrows represent the causal influence direction:

[0190] C={(X i →X j )|F ij >F α};

[0191] S45. Establishing causal inference analysis model M causal The causal inference analysis model consists of a causal chain C and the corresponding causal strength F ij It is used to analyze and locate the causes of wind turbine failures.

[0192] In this implementation, S5 includes:

[0193] S51. Construct fault-related operating parameter set X fault (t), based on the causal inference model M causal The causal chain C extracts the target variable X j The causal path of (t):

[0194]

[0195] S52, fault-related operating parameter set X fault (t) Each dependent variable X i (t) Calculate its rate of change ΔX i (t):

[0196]

[0197] Among them, Δt1 is the time step, γ ik For other parameters X k (t) for X i (t) The cross-influence factor describes the dynamic coupling between the internal parameters of the wind turbine;

[0198] S53, based on the rate of change ΔX i (t) passes the fault trigger threshold θ i Determine the fault triggering time point set T fault :

[0199]

[0200] Among them, t f Indicates the time point when the trigger condition is met. The max operation selects the parameter with the most significant change as the sign of fault triggering;

[0201] S54, fault triggering time point set T fault At each time point t f Review of its key impact factor set I fault (t f );

[0202] S55, based on key impact factor set I fault (t f ) and causal inference model M causal , combined with the playback result M global Generate fault evolution process E fault :

[0203] E fault ={(t f ,I fault (t f ),X j (t))|t f ∈T fault ,X j (t) is the target fault parameter};

[0204] Among them, E fault Describe the occurrence time of the fault, key influencing factors and parameter evolution path;

[0205] S56, according to the fault evolution process E fault and causal chain C to locate the root cause of the wind turbine fault R fault :

[0206]

[0207] Among them, R fault It represents the set of root cause parameters obtained through causal path tracing analysis, identifying the specific fault source of the wind turbine.

[0208] In this implementation, S6 includes:

[0209] S61. Comprehensively review the failure process of the wind turbine generator system based on the multi-layer data playback model and causal inference analysis model, and extract data of all time nodes, spatial locations and equipment modules related to the failure;

[0210] S62, analyzing the set of fault triggering time points, determining the time axis of the fault occurrence, marking the change characteristics of the operating parameters in different time periods and the evolution trend of the fault to form a fault replay sequence in the time dimension;

[0211] S63. Analyze the impact of different sensor locations and equipment modules on the target fault during the fault evolution process by combining the spatial dimension multi-level data in the global data playback model, and identify the coupling relationship between spatial locations and their contribution to the fault;

[0212] S64, extracting the operation characteristics of the equipment module dimension, analyzing the role of different functional modules of the wind turbine in the occurrence and propagation of faults, and constructing a fault correlation map between modules;

[0213] S65. Summarize and simplify the causal chain in the causal inference analysis model, extract the main causal path of the fault, and generate a causal network structure that reflects the laws of fault triggering, propagation and diffusion;

[0214] S66. Construct a wind turbine fault mode library based on a comprehensive analysis of the time axis, spatial dimension, equipment module and causal path, classify the fault occurrence modes and compare them with existing historical data to extract representative fault characteristics and evolution patterns;

[0215] S67. Generate a wind turbine fault reconstructive report, which includes the following contents:

[0216] The time of occurrence of the failure and key influencing factors;

[0217] The fault propagation process and the spatial locations and equipment modules involved;

[0218] Causal paths and root cause parameters of failures;

[0219] Failure modes and failure mode characteristics;

[0220] Optimization suggestions and early warning strategies are proposed based on the review results.

[0221] Embodiment 1:

[0222] In order to verify the feasibility and effectiveness of the wind turbine fault tracing and replay method based on multi-layer data playback proposed in the present invention, this embodiment 1 takes the wind turbine fault in the actual operation of a wind farm as the background, simulates the real fault occurrence and diagnosis scenario through multi-layer data playback, causal analysis and fault replay process, and demonstrates the advantages of the present invention in complex fault location and analysis.

[0223] A coastal wind farm has a total of 50 wind turbines with a total operation life of 5 years. In June 2024, a wind turbine numbered WT-23 suddenly experienced an abnormal drop in power generation during operation, from the rated power of 2MW to 1.2MW, accompanied by obvious abnormal main shaft vibration. The wind speed where the wind turbine is located is 12.5m / s, and other surrounding wind turbines are operating normally. This type of fault is highly hidden, and traditional methods make it difficult to quickly locate the cause of the fault, which may result in long-term shutdown and cause significant economic losses.

[0224] This embodiment 1 uses the method proposed in the present invention to completely track and replay the fault, and obtains the operating data set through the SCADA system of the wind turbine set, including five types of data: wind speed, power generation, main shaft vibration, current and voltage. The time range is from June 1, 2024 to June 7, 2024, and the data is recorded every 10 seconds, with a total of approximately 60,480 data points.

[0225] After preprocessing, the data is divided into multiple levels according to the time dimension, space dimension and equipment module dimension. For example, the time dimension is divided into three time periods per day (0:00-8:00, 8:00-16:00, 16:00-24:00), the space dimension corresponds to the sensor installation location (main shaft, blade, generator), and the equipment module includes the converter, generator and control cabinet. Multi-layer data is input into the global data playback model for dynamic reproduction, gradually restoring the entire process of the fault occurrence.

[0226] During the playback process, it was found that the spindle vibration data began to gradually increase at 16:00 on June 3, and the vibration amplitude increased from 5mm / s to 15mm / s, and reached a peak on June 5. Subsequently, the power generation began to decline at 10:00 on June 5, which was associated with the increase in converter temperature. In addition, the causal inference model showed that there was a strong causal relationship between the spindle vibration and the converter temperature, with a causal strength of 18.5, which was significantly higher than the set threshold of 5.0, indicating that abnormal spindle vibration was the main cause of converter overheating and power drop.

[0227] Further analysis based on the cause-and-effect chain showed that spindle bearing wear was the root cause of the vibration, and this wear was related to stress concentration caused by uneven blade load distribution. Through system review, the case revealed the complete fault chain from abnormal blade load to spindle wear and then to power drop, providing clear guidance for subsequent maintenance.

[0228] In this embodiment, the method of the present invention is compared with a traditional single-dimensional data analysis method.

[0229] Fault location time: The method of the present invention completes multi-dimensional data playback and causal chain analysis within 2 hours, locating the root cause of the fault as spindle bearing wear. The traditional method requires multiple rounds of manual investigation and component disassembly, which takes an average of 72 hours.

[0230] Positioning accuracy: The fault positioning accuracy of the method of the present invention reaches 92% in the data of 30 training sample wind turbines, while the accuracy of the traditional method is about 65%, especially in multi-causal coupling faults, the accuracy drops significantly.

[0231] Economic loss assessment: The method of the present invention quickly reviewed the fault and clarified the fault repair plan, reducing the downtime to 3 hours and controlling the economic loss to RMB 50,000. The traditional method had a long downtime (about 72 hours) and the economic loss was as high as RMB 300,000.

[0232] It can be seen from this Example 1 that the method of the present invention can quickly and accurately track and review the entire process of complex faults, greatly improving the fault diagnosis efficiency and reliability of maintenance decisions of wind turbines. Through multi-layer data playback and causal inference analysis, the present invention is superior to traditional methods in terms of fault location accuracy, analysis efficiency and economic benefits, and has significant technical and practical value.

[0233] The present invention constructs a multi-level data set in three dimensions of time, space and equipment module, and dynamically reproduces the data based on a global data playback model to fully reproduce the entire process of wind turbine failure. Compared with the existing single-level data analysis method, the multi-level data playback model can capture the correlation characteristics of faults in different dimensions, the trend changes on the time axis, the coupling relationship between spatial positions and the functional differences of equipment modules, which significantly improves the accuracy and comprehensiveness of fault location, and provides reliable data support for the in-depth analysis of complex faults. Experimental verification shows that the fault location accuracy of this method is improved by more than 15% compared with the traditional single-dimensional data analysis model.

[0234] The present invention generates a causal chain between the operating parameters of the wind turbine set through a causal inference algorithm, and combines the causal chain to trace and analyze the fault path, effectively overcoming the limitation of traditional statistical methods that are difficult to handle complex causal relationships. In the causal inference analysis, improved algorithms for Granger causality test, causal network modeling and causal strength quantification are introduced, so that the occurrence, propagation and evolution process of the fault can be clearly expressed in the form of a causal path. Compared with the existing fault diagnosis method based on correlation analysis, the present invention can clearly identify the direct factors and indirect factors that affect key parameters, improve the accuracy of fault tracking by 20%, and significantly enhance the processing capability of multi-source coupled faults.

[0235] The present invention establishes a full-process review mechanism for wind turbine failures by combining a global data playback model with a causal inference analysis model, covering multi-dimensional data analysis of time, space and equipment modules. By extracting the key time points, key influencing factors and causal paths of fault triggering, a fault review report with a global perspective is generated, and the failure modes are summarized and concluded based on the review results. Compared with the traditional method of only performing local repairs on a single fault point, the present invention can systematically identify the evolution laws of wind turbine failures and provide guidance for subsequent early warning, maintenance optimization and operation strategy formulation. Experimental data show that the failure mode extracted by this method is more than 90% consistent with the actual failure process, which significantly improves the intelligence and reliability of wind turbine maintenance.

[0236] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A wind turbine fault tracing and replay method based on multi-layer data playback, characterized in that: The steps include: S1. Obtaining a wind turbine operation data set and preprocessing the wind turbine operation data set; S2. Divide the multi-level data according to the time dimension, space dimension and equipment module of the wind turbine operation data set; S3. Construct a multi-layer data playback model based on multi-layer data; S4. Based on the multi-layer data playback model, the causal relationship between multi-dimensional data is analyzed using the causal inference algorithm to generate the causal chain of wind turbine operating parameters and establish a causal inference analysis model; S5. Use the causal inference analysis model to analyze the multi-level data playback results, track the occurrence time, influencing factors and evolution process of wind turbine failures, and locate the root cause of the failure; S6. Based on the multi-layer data playback results and fault location analysis, the entire process of wind turbine failure is reviewed, the fault occurrence mode is extracted, and a wind turbine failure review report is generated.

2. A wind turbine fault tracing and replay method based on multi-layer data playback according to claim 1, characterized in that: The S1 includes: S11. Obtain the wind speed V of the wind turbine w , power generation P, speed ω, current I and voltage U, construct the wind turbine operation data set, and preliminarily organize the wind turbine operation data set according to the time series T to form the time series data set D t : D t ={(V w (t i ),P(t i ),ω(t i ),I(t i ),U(t i ))|t i ∈T}; Among them, t i represents the data record at the i-th time point, and n is the number of sampling points; S12, using the sliding mean filtering method to filter the time series data set D t Perform smoothing to filter out random noise and generate the denoised wind turbine operation data set D t_smooth .

3. A wind turbine fault tracing and replay method based on multi-layer data playback according to claim 1, characterized in that: The S1 further comprises: S13. Running data set D for denoised wind turbines t_smooth The parameters in are normalized so that their range is mapped to the interval [0,1]; S14. For the missing values ​​in the normalized wind turbine operation data set, a time interpolation method is used to fill them, and a preprocessed wind turbine operation data set D is generated. processed : D processed ={(V″ w (t i ),P"(t i ),ω"(t i ),I″(t i ),U″(t i ))|t i ∈T}; Among them, V w (t i ),P″(t i ),ω″(t i ),I″(t i ),U″(t i ) represents the parameter data after denoising, normalization and missing value filling.

4. A wind turbine fault tracing and replay method based on multi-layer data playback according to claim 1, characterized in that: The S2 includes: S21, based on the preprocessed wind turbine operation data set D processed Define the time window length Δt and divide the time series T into m time periods: T k =[t k-1 ,t k ); t k =t0+k·Δt; Among them, t0 is the starting time, t k is the end time of the kth time layer, t end is the end time; S22. For each time period T k Extract the corresponding operation data to form a multi-level data set in the time dimension S23, based on the preprocessed wind turbine operation data set Dprocessed, define the spatial position set S = {s1, s2, ..., s p }, where s p represents the pth physical position or component of the wind turbine. For each spatial position s p Extract the operating data of the corresponding location to form a multi-level data set in spatial dimension Among them, s i At time point t i The location of the sensors that collect the data; S24, based on the preprocessed wind turbine operation data set D processed Define the device module set M = {m1,m2,…,m q }, where m q represents the qth functional module of the wind turbine. For each equipment module m q Extract the operation data of the corresponding module to form a multi-level data set of equipment modules Among them, m i At time point t i The device module to which the collected data belongs; S25. Multi-level datasets with time dimension Multi-level datasets in spatial dimension and equipment module multi-level dataset Integrate to form a multi-level dataset in, Represents a set of data that meets the conditions at the time layer k, space layer l, and device module layer j at the same time.

5. The method for wind turbine fault tracing and replaying based on multi-layer data playback according to claim 1 is characterized in that: Based on multi-level datasets For each time period T k Using time dimension multi-level dataset Constructing the time state matrix The state transfer equation of the time dimension is established. The state transfer equation of the time dimension describes the dynamic behavior of the wind turbine in time: in, is the state transfer matrix in the time dimension, is the control input matrix of the time dimension, is the control input vector of the time dimension, is the process noise vector in the time dimension; Use the state transfer equation to calculate the time period T k The state in the time is replayed time by time to generate a time-by-time replay model .

6. A wind turbine fault tracing and replay method based on multi-layer data playback according to claim 5, characterized in that: Based on multi-level datasets For each spatial position s l Exploiting spatial dimensions of multi-level datasets Constructing the space state matrix Establish the state propagation equation of the spatial dimension, which describes the mutual influence between different spatial positions: in, is the spatial coupling matrix, representing the spatial position s m For spatial position s l The influence weight of (l,m) is the spatial position s m For spatial position s l The propagation delay of is the process noise vector of spatial dimension; Use the state propagation equation to perform distributed playback of the states of each spatial position to generate a distributed playback model .

7. A wind turbine fault tracing and replay method based on multi-layer data playback according to claim 6, characterized in that: By replaying the model moment by moment and distributed playback model Constructing the global state matrix Integrate data from time, space and device module dimensions: in, is the device module m j The state matrix of A global state transfer equation is established to describe the comprehensive dynamic behavior of wind turbines in multiple dimensions: in, is the global state transfer matrix, is the global control input matrix, is the global control input vector, is the global process noise vector; The global state transfer equation is used to replay the global state of the wind turbine to generate a global data replay model M. global .

8. The method for wind turbine fault tracing and replaying based on multi-layer data playback according to claim 1 is characterized in that: The S4 includes: S41, based on global data playback model M global Extract multi-dimensional operating parameter data and construct a causal analysis data set D causal ; S42. Causal analysis data set D causal Perform preprocessing and define the random variable set X(t); S43, using causal inference algorithm to determine the variable X(t) using Granger causality test method i (t) Is it a variable X? j Granger causes of (t), where X i (t),X j (t)∈X(t); Construct a univariate autoregressive model: Among them, X j (t) is the target variable, N is the lag order, Δt is the time step, a jn is the coefficient of the autoregressive model, describing X j (t) Dependence on its own historical value, ∈ j (t) is the model residual; Combined with the dependent variable X i (t) for the target variable X j (t) constructs a multivariate causal regression model: Among them, b jn represents the target variable X j The autoregressive coefficient of (t), c in Denotes the dependent variable X i (t) for the target variable X j The causal influence strength of (t), ∈′ j (t) is the model residual; Construct the causal test objective function and calculate the residual sum of squares RSS1 and RSS2 of the univariate model and the multivariate model: Calculate the F statistic for the Granger causality test ij : Where T is the number of time series samples, F ij Denotes the dependent variable X i (t) Whether it significantly affects the target variable X j The Granger causal strength of (t); Find the critical value F of the F distribution according to the preset significance level α α , if F ij >F α , then the dependent variable X i (t) is the target variable X j Granger cause of (t), there is a causal relationship; S44. Based on the causal relationship analysis results, a causal chain C of the wind turbine operating parameters is generated. The causal chain C represents the causal path between the operating parameters, where the arrows represent the causal influence direction: C={(X i →X j )|F ij >F α }; S45. Establishing causal inference analysis model M causal The causal inference analysis model consists of a causal chain C and the corresponding causal strength F ij It is used to analyze and locate the causes of wind turbine failures.

9. The method for wind turbine fault tracing and replaying based on multi-layer data playback according to claim 1 is characterized in that: The S5 includes: S51. Construct fault-related operating parameter set X fault (t), based on the causal inference model M causal The causal chain C extracts the target variable X j The causal path of (t): S52, fault-related operating parameter set X fault (t) Each dependent variable X i (t) Calculate its rate of change ΔX i (t): Among them, Δt1 is the time step, γ ik For other parameters X k (t) for X i (t) The cross-influence factor describes the dynamic coupling between the internal parameters of the wind turbine; S53, based on the rate of change ΔX i (t) passes the fault trigger threshold θ i Determine the fault triggering time point set T fault : Among them, t f Indicates the time point when the trigger condition is met. The max operation selects the parameter with the most significant change as the sign of fault triggering; S54, fault triggering time point set T fault At each time point t f Review of its key impact factor set I fault (t f ); S55, based on key impact factor set I fault (t f ) and causal inference model M causal , combined with the playback result M global Generate fault evolution process E fault : E fault ={(t f ,I fault (t f ),X j (t))|t f ∈T fault ,X j (t) is the target fault parameter}; Among them, E fault Describe the occurrence time of the fault, key influencing factors and parameter evolution path; S56, according to the fault evolution process E fault and causal chain C to locate the root cause of the wind turbine fault R fault : Among them, R fault It represents the set of root cause parameters obtained through causal path tracing analysis, identifying the specific fault source of the wind turbine.

10. A wind turbine fault tracing and replay method based on multi-layer data playback according to claim 1, characterized in that: The S6 includes: S61. Comprehensively review the failure process of the wind turbine generator system based on the multi-layer data playback model and causal inference analysis model, and extract data of all time nodes, spatial locations and equipment modules related to the failure; S62, analyzing the set of fault triggering time points, determining the time axis of the fault occurrence, marking the change characteristics of the operating parameters in different time periods and the evolution trend of the fault to form a fault replay sequence in the time dimension; S63. Analyze the impact of different sensor locations and equipment modules on the target fault during the fault evolution process by combining the spatial dimension multi-level data in the global data playback model, and identify the coupling relationship between spatial locations and their contribution to the fault; S64, extracting the operation characteristics of the equipment module dimension, analyzing the role of different functional modules of the wind turbine in the occurrence and propagation of faults, and constructing a fault correlation map between modules; S65. Summarize and simplify the causal chain in the causal inference analysis model, extract the main causal path of the fault, and generate a causal network structure that reflects the laws of fault triggering, propagation and diffusion; S66. Based on the comprehensive analysis of time axis, space dimension, equipment module and causal path, a wind turbine fault mode library is constructed to classify the fault occurrence modes and compare them with the existing historical data to extract representative fault characteristics and evolution modes; S67. Generate a wind turbine fault review report.

11. A wind turbine fault tracing and replay method based on multi-layer data playback according to claim 1, characterized in that: The review report includes: The time of occurrence of the failure and key influencing factors; The fault propagation process and the spatial locations and equipment modules involved; Causal paths and root cause parameters of failures; Failure modes and failure mode characteristics; Optimization suggestions and early warning strategies are proposed based on the review results.

Images