Early fault warning method and device for wind turbine based on specific causal network

Abstract

This invention discloses a method and device for early fault warning of wind turbines based on specific causal networks, relating to the field of data processing. The method includes: calculating the directed transfer entropy between any two variables in each time window using conditional mutual information based on the data matrix of each time window; constructing a specific causal network for each time window and calculating the variational causal characteristics of the current time window; calculating the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window and the comprehensive state index value of each time window based on the variational causal characteristics of each time window, and calculating the change in the comprehensive state index value between the current time window and the previous time window; and determining whether the current time window is a warning window based on the change in the comprehensive state index value. This invention solves the problem of difficulty in providing early warnings based on SCADA data.

Description

Technical Field

[0001] This invention relates to the field of data processing, and specifically to a method and device for early fault warning of wind turbines based on specific causal networks. Background Technology

[0002] With the rapid growth of wind power installed capacity, the operation and maintenance of wind turbine units are facing severe challenges. On the one hand, wind farms are typically located in remote mountainous areas with abundant wind resources or offshore areas far from land, requiring them to withstand complex operating conditions such as high wind speeds and extreme weather, which places extremely high demands on the long-term reliability of the equipment. On the other hand, maintenance work itself is costly, involving huge investments of human, material, and financial resources, constituting a key challenge to the sustainable development of the industry. Excessive downtime can lead to significant reductions in power generation efficiency and economic losses. The root cause lies in core technical challenges such as insufficient accuracy in fault prediction and slow maintenance response mechanisms, which greatly exacerbate the complexity of maintenance work. The complexity of wind turbine units stems from the highly nonlinear and strongly coupled relationships between their mechanical, electrical, and aerodynamic subsystems. This complexity often results in a many-to-many correlation between the causes and phenomena of faults, meaning that a single trigger may lead to a chain of faults, while a single fault is often the result of multiple factors working together. Therefore, establishing an efficient and reliable early fault warning technology system for wind turbines is of great importance. It not only helps to improve the safety of equipment operation and reduce maintenance costs, but also ensures the economical and stable operation of wind power systems, and has important engineering application value and practical significance.

[0003] Currently, most wind turbine equipment is equipped with Supervisory Control and Data Acquisition (SCADA) systems. SCADA systems can collect equipment status data at a frequency of seconds / minutes. Their biggest advantage is that they do not require additional sensors; they can achieve economical and efficient status monitoring simply by utilizing the database recorded by the SCADA system itself, and have demonstrated good results in production practice. Therefore, early warning and fault diagnosis methods for wind turbines based on SCADA data have attracted much attention from scholars both domestically and internationally, with related research continuously deepening and significant results verified by engineering practice. Existing early warning and fault diagnosis methods for wind turbines based on SCADA data can be mainly divided into two categories: one is supervised learning methods based on classification models, and the other is unsupervised or semi-supervised methods based on Normal Behavior Models (NBMs). Because fault samples in SCADA data are scarce and highly imbalanced, it is difficult to construct a classification model covering all fault types, thus limiting the practical application of classification methods. In contrast, normal behavior models use historical data on the health status of the unit to establish a normal operating baseline and identify anomalies by detecting significant deviations between new data and model fit values. The core of these models lies in their ability to accurately simulate the normal operating behavior of the unit.

[0004] While research has deepened and SCADA data-based early warning and fault diagnosis methods for wind turbines have achieved some success in engineering practice, existing technologies still have certain shortcomings: SCADA data is susceptible to interference and noise, and often contains missing or outlier values, increasing the difficulty of preprocessing. Furthermore, since wind turbines operate normally most of the time and faults rely on manual recording, there is a severe shortage of labeled anomaly samples available for model training, which becomes a fundamental bottleneck for supervised learning methods. At the model and algorithm level, although many advanced machine learning models can capture complex nonlinear relationships, their "black box" nature leads to poor interpretability of the decision-making process, making it difficult for maintenance personnel to trust the results and locate the root cause of faults. Additionally, models are often insensitive to early, subtle performance degradation features, mostly only detecting existing faults and struggling to achieve true predictive maintenance. The setting of alarm thresholds lacks a unified standard; excessive sensitivity can lead to false alarms, while excessive leniency can result in missed alarms. This dilemma is highly dependent on experience and difficult to optimize. Secondly, wind turbine anomalies often involve the interaction of multiple SCADA parameters, and multivariate coupling analysis places high demands on algorithms and computing resources. In addition, existing models mostly rely on historical fault data for training, and their ability to detect unknown or new faults is insufficient. Summary of the Invention

[0005] The purpose of this application is to propose a method and device for early fault warning of wind turbine units based on specific causal networks to address the aforementioned technical problems.

[0006] In a first aspect, the present invention provides a method for early fault warning of wind turbines based on specific causal networks, comprising the following steps:

[0007] The time series data of all numerical variables continuously monitored in the SCADA system of the wind turbine are acquired and preprocessed to obtain preprocessed time series data; a data matrix of several time windows is constructed based on the preprocessed time series data using a sliding window.

[0008] Based on the data matrix of each time window, the directed transit entropy between any two variables in each time window is calculated using conditional mutual information, and a specific causal network for each time window is constructed.

[0009] The variant causal characteristics of the current time window are calculated based on the specific causal network of the current time window and the specific causal network of the previous time window. The variant causal characteristics include the dynamic causal fluctuation intensity of each node in the specific causal network and all nodes in its first-order neighborhood, the average specific transfer entropy coefficient between each node and any node in its first-order neighborhood, and the average specific transfer entropy coefficient between any node in its first-order neighborhood and any node in its second-order neighborhood. Based on the variant causal characteristics of each time window, the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window is calculated.

[0010] The comprehensive state index value for each time window is calculated based on the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window, and the change in the comprehensive state index value between the current time window and the previous time window is calculated; based on the change in the comprehensive state index value between the current time window and the previous time window, it is determined whether the current time window is an early warning window.

[0011] Preferably, the directed transit entropy between any two variables within each time window is calculated using conditional mutual information based on the data matrix of each time window. Specifically, this includes:

[0012] Let the data matrix for each time window be: ;

[0013] in, The length of each time window, The total number of variables. Indicates the first The first variable within the time window The value at each moment. ;

[0014] Based on the data matrix for each time window, construct sample triples for any two variables, represented as: ;in, The total number of sampling moments within each time window. , For the time lag, This indicates taking the first element from the data matrix in each time window. The variables range from the first sampling time to the second sampling time. The first vector is composed of the values ​​at each sampling time. This indicates taking the first element from the data matrix in each time window. The variable from the first From the sampling time to the... The second vector is composed of the values ​​at each sampling time. This indicates taking the first element from the data matrix in each time window. The variables range from the first sampling time to the second sampling time. The third vector is composed of the values ​​at each sampling time.

[0015] In the current time window Within, given the order of nearest neighbors From the sample triplet, the first The value at the sampling time and the first sampling time The values ​​at each sampling time point are respectively used to construct the first... Joint vector at each sampling time point and the Joint vector at each sampling time point , where [] indicates splicing, , and The first vector is the first one. The element and the first One element, and The second vector is respectively the first The element and the first One element, and The third vector is the first one. The element and the first Elements; constructing a joint space through joint vectors. ;

[0016] In the joint space Internal search nearest neighbor radius Based on the nearest neighbor radius in the subspace , and The internal neighbor count is represented as:

[0017] ;

[0018] ;

[0019] ;

[0020] in, This represents the maximum norm distance between two vectors. Represents the count of a set. , and They represent the first Centered on each sample point in the subspace , and Falling within the nearest radius The number of other sample points in the neighborhood;

[0021] The directed transitive entropy is calculated explicitly using KSG estimation based on conditional mutual information, as shown in the following equation:

[0022] ;

[0023] in, Indicates the current time window From the inside The variable points to the first... The directed transitive entropy of each variable, For conditional mutual information, This is the digamma function.

[0024] As a preferred approach, the construction process for the specific causal network for each time window is as follows:

[0025] Within each time window, using each variable as a node, starting from the first... The variable points to the first... The directed transitive entropy of the nth variable is used as the... The node points to the first The weights of the edges of each node are used to form the directed transit entropy matrix, which is the directed transit entropy matrix between any two variables. This matrix is ​​then used as the weighted directed adjacency matrix of the specific causal network for each time window, and a directed weighted graph is constructed.

[0026] As a preferred embodiment, the formula for calculating the dynamic causal fluctuation intensity of each node and all nodes in its first-order neighborhood in a specific causal network is as follows:

[0027] ;

[0028] in, Indicates the first Each node in the current time window The intensity of dynamic causal fluctuations of all nodes in the next first-order neighborhood set. For the first Each node in the current time window The next first-order neighborhood set, by the first It consists of nodes that are connected by edges. For the first Each node in the current time window The number of nodes in the next first-order neighborhood. For the first Each node in the current time window The average of the following values, For the first Each node in the current time window The standard deviation of the values ​​below, the first The node is the _th Each node in the current time window The next first-order neighbor node, For the first Each node in the current time window The average of the following values, For the first Each node in the current time window The standard deviation of the values ​​below;

[0029] The formula for calculating the average specific transfer entropy coefficient between each node in a specific causal network and any node in its first-order neighborhood is as follows:

[0030] ;

[0031] in, Indicates the first Each node and its position in the current time window The average value of the specific transfer entropy coefficient between each node in the next first-order neighborhood set. Indicates the previous time window From the inside The node points to the first The directed transitive entropy of each node ; Indicates the first Each node and its position in the current time window The first-order neighborhood set below The specific transfer entropy coefficient between nodes;

[0032] The formula for calculating the average idiosyncratic transfer entropy coefficient between any node in the first-order neighborhood and any node in the second-order neighborhood of each node in a idiosyncratic causal network is as follows:

[0033] ;

[0034] in, Indicates the first Each node in the current time window The average value of the specific transfer entropy coefficient between each node in the first-order neighborhood set and each node in the second-order neighborhood set; Indicates the first Each node in the current time window The first in the next first-order neighborhood set The node points to the first node in the second-order neighborhood set. The directed transitive entropy of each node Indicates the first The nodes in the previous time window The first-order neighborhood set under -1 The node points to the first node in the second-order neighborhood set. The directed transitive entropy of each node; Indicates the first Each node in the current time window The first-order neighborhood set below The node and its second-order neighborhood set The specific transfer entropy coefficient between nodes.

[0035] As a preferred embodiment, the formula for calculating the dynamic network biomarker score of causal variation potential for each node in the specific causal network for each time window is as follows:

[0036] ;

[0037] in, Indicates the current time window The first specific causal network The dynamic network marker score of the causal variation potential of each node;

[0038] The formula for calculating the comprehensive status index value for each time window is as follows:

[0039] ;

[0040] in, Indicates the current time window The comprehensive state index value, where m represents the total number of variables;

[0041] The formula for calculating the change in the comprehensive status index value between the current time window and the previous time window is as follows:

[0042] ;

[0043] in, This indicates the change in the overall status index value between the current time window and the previous time window. Indicates the previous time window The comprehensive status index value.

[0044] Preferably, the preprocessing process includes normalization, removal of outliers using the local outlier factor algorithm, and filling in the removed values ​​using the K-nearest neighbor algorithm.

[0045] Secondly, the present invention provides a wind turbine early fault warning device based on a specific causal network, comprising:

[0046] The preprocessing module is configured to acquire and preprocess the time-series data of all numerical variables continuously monitored in the SCADA system of the wind turbine to obtain preprocessed time-series data; and to construct a data matrix of several time windows based on the preprocessed time-series data using a sliding window.

[0047] The specific causal network construction module is configured to calculate the directed transit entropy between any two variables in each time window using conditional mutual information based on the data matrix of each time window, and construct the specific causal network for each time window.

[0048] The feature calculation module is configured to calculate the variant causal features of the current time window based on the specific causal network of the current time window and the specific causal network of the previous time window. The variant causal features include the dynamic causal fluctuation intensity of each node in the specific causal network and all nodes in its first-order neighborhood, the average specific transfer entropy coefficient between each node and any node in its first-order neighborhood, and the average specific transfer entropy coefficient between any node in its first-order neighborhood and any node in its second-order neighborhood. Based on the variant causal features of each time window, the module calculates the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window.

[0049] The early warning module is configured to calculate the comprehensive state index value of each time window based on the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window, and calculate the change value of the comprehensive state index value between the current time window and the previous time window; and determine whether the current time window is an early warning window based on the change value of the comprehensive state index value between the current time window and the previous time window.

[0050] Thirdly, the present invention provides an electronic device including one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0051] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any of the implementations of the first aspect.

[0052] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method as described in any of the implementations in the first aspect.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) The wind turbine early fault warning method based on specific causal network mentioned in this invention calculates the directed transfer entropy matrix between variables based on conditional mutual information estimation under the sliding time window according to the variables continuously monitored by the SCADA system of the target wind turbine, and maps it to a specific causal network. By calculating three indicators in the variation causal characteristics of the dynamic network markers in the specific causal network, the comprehensive state index value of the wind turbine is obtained, thereby realizing the identification and early warning of the evolution of the wind turbine's operating state from normal to abnormal (critical).

[0055] (2) The wind turbine early fault warning method based on specific causal network mentioned in this invention does not require a large number of fault labeled samples for machine learning training, nor does it require the establishment of a complex mechanism model, which is conducive to realizing early defect detection and early warning of wind turbines, reducing operation and maintenance costs and improving operational reliability.

[0056] (3) The wind turbine early fault warning method based on specific causal network mentioned in this invention Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating an embodiment of the wind turbine early fault warning method based on a specific causal network according to this application.

[0059] Figure 2A three-dimensional distribution diagram of key variables, time windows, and scores in the wind turbine SCADA system of the wind turbine early fault warning method based on specific causal networks, as an embodiment of this application.

[0060] Figure 3 The specific causal network diagram of the key variables of the wind turbine SCADA system in the 100th time window of the wind turbine early fault warning method based on specific causal network in the embodiment of this application.

[0061] Figure 4 The specific causal network diagram of the key variables of the wind turbine SCADA system in the 2500th time window of the wind turbine early fault warning method based on specific causal network in the embodiment of this application.

[0062] Figure 5 This is a schematic diagram illustrating the trend and early warning of the comprehensive status index value of a wind turbine as a function of a time window in the wind turbine early fault warning method based on a specific causal network, as described in an embodiment of this application.

[0063] Figure 6 This is a schematic diagram of an early fault warning device for wind turbines based on a specific causal network, as an embodiment of this application.

[0064] Figure 7 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0066] Figure 1 An embodiment of this application illustrates an early fault warning method for wind turbines based on specific causal networks, comprising the following steps:

[0067] S1. Obtain the time series data of all numerical variables continuously monitored in the SCADA system of the wind turbine and preprocess them to obtain preprocessed time series data; construct a data matrix of several time windows using a sliding window based on the preprocessed time series data.

[0068] In a specific embodiment, the preprocessing process includes normalization, removal of outliers using the local anomaly factor algorithm, and filling in the removed values ​​using the K-nearest neighbor algorithm.

[0069] Specifically, the time series data of all numerical variables in the SCADA system of the wind turbine collected in the embodiments of this application need to be preprocessed. The preprocessing method includes normalizing the time series data, removing outliers from the normalized time series data using the Local Outlier Factor (LOF) algorithm, removing points that are determined to be outliers by LOF, and filling the removed points using K Nearest Neighbors (KNN).

[0070] S2, based on the data matrix of each time window, calculate the directed transit entropy between any two variables in each time window using conditional mutual information, and construct a specific causal network for each time window.

[0071] Specifically, the preprocessed time-series data is divided into sliding windows and a data matrix with several time windows is constructed. The length of the sliding window is set to... Step size is The total number of monitored variables is Continuous data acquisition Variables in The data is obtained at the nth time point, and a data matrix is ​​constructed for each time window. That is, the nth time point in the preprocessed time series data is selected. To the The SCADA data at each sampling time is used as a data matrix for a time window. The above steps are repeated K times to obtain a data matrix for K time windows.

[0072] In a specific embodiment, the directed transit entropy between any two variables within each time window is calculated using conditional mutual information based on the data matrix of each time window. This specifically includes:

[0073] Let the data matrix for each time window be: ;

[0074] in, The length of each time window, The total number of variables. Indicates the first The first variable within the time window The value at each moment. ;

[0075] Based on the data matrix for each time window, construct sample triples for any two variables, represented as: ;in, The total number of sampling moments within each time window. , For the time lag, This indicates taking the first element from the data matrix in each time window. The variables range from the first sampling time to the second sampling time. The first vector is composed of the values ​​at each sampling time. This indicates taking the first element from the data matrix in each time window. The variable from the first From the sampling time to the... The second vector is composed of the values ​​at each sampling time. This indicates taking the first element from the data matrix in each time window. The variables range from the first sampling time to the second sampling time. The third vector is composed of the values ​​at each sampling time.

[0076] In the current time window Within, given the order of nearest neighbors From the sample triplet, the first The value at the sampling time and the first sampling time The values ​​at each sampling time point are respectively used to construct the first... Joint vector at each sampling time point and the Joint vector at each sampling time point , where [] indicates splicing, , and The first vector is the first one. The element and the first One element, and The second vector is respectively the first The element and the first One element, and The third vector is the first one. The element and the first Elements; constructing a joint space through joint vectors. ;

[0077] In the joint space Internal search nearest neighbor radius Based on the nearest neighbor radius in the subspace , and The internal neighbor count is represented as:

[0078] ;

[0079] ;

[0080] ;

[0081] in, This represents the maximum norm distance between two vectors. Represents the count of a set. , and They represent the first Centered on each sample point in the subspace , and Falling within the nearest radius The number of other sample points in the neighborhood;

[0082] The directed transitive entropy is calculated explicitly using KSG estimation based on conditional mutual information, as shown in the following equation:

[0083] ;

[0084] in, Indicates the current time window From the inside The variable points to the first... The directed transitive entropy of each variable, For conditional mutual information, This is the digamma function.

[0085] Specifically, for each time window, the directed transitive entropy matrix between variables is calculated based on conditional mutual information estimation; the specific causal network for each time window is then constructed using this matrix. The elements of the directed transitive entropy matrix represent the directed transitive entropy between any two variables. The specific calculation process is as follows:

[0086] For each time window, establish a variable set for the numerically monitored variables. For any two variables in the variable set, take the first variable as the first variable. The first variable and the second Using the first variable as an example, sample triples are constructed within each time window, that is, the first variable is obtained from the data matrix of each time window by lag time. The variables range from the first sampling time within the time window to the [missing value]. The first vector is composed of the values ​​at each sampling time. Obtain the first data from the data matrix in each time window. The variable from the first From the sampling time to the... The second vector is composed of the values ​​at each sampling time. Obtain the first data from the data matrix in each time window. The variables range from the first sampling time to the second sampling time. The third vector is composed of the values ​​at each sampling time. Through the first The first vector of the nth variable and the nth variable The conditional mutual information between the second and third vectors of each variable is calculated from the variables. Pointer variable The directed transitive entropy. A KSG-kNN-based conditional mutual information estimator is used to explicitly estimate this conditional mutual information; specifically, given the nearest neighbor order... Define a joint vector for each sampling moment within the current time window. The first term of this joint vector is the element corresponding to the sampling moment in the first vector, the second term is the element corresponding to the sampling moment in the second vector, and the third term is the element corresponding to the sampling moment in the third vector. These three elements are concatenated to form the joint vector. Using these three elements as three coordinate dimensions, a joint space is constructed. In the joint space Internal search nearest neighbor radius Based on the nearest neighbor radius, in the subspace , and Internal neighbor count, subspace It is the subspace constructed by the first and third elements of the joint vector. It refers to the subspace constructed by the second and third elements of the joint vector. This refers to the subspace constructed by the third element of the joint vector. The Chebyshev distance (L) is used. ∞ (Measurement) The joint vector of one sampling moment within each time window and the joint vector of all other sampling moments in the subspace. , and Calculate the nearest neighbor distance for the coordinates in the matrix, and for those that satisfy the condition less than the first nearest neighbor... By counting the set of sampling times of the nearest neighbor radii, we can obtain the subspace centered on one of the sampling times. , and Within, falling within the nearest radius of The number of other sampling times in the neighborhood. Substituting this number into the KSG explicit estimation formula yields the conditional mutual information. Taking the first... Joint vector at each sampling time point and the Joint vector at each sampling time point As an example, the formula for calculating the Chebyshev distance in the embodiments of this application is as follows:

[0087] ;

[0088] in, For the first Joint vector at each sampling time point The The component and the first Joint vector at each sampling time point The One portion, This indicates taking the maximum value. The dimension index of the joint vector. This represents the total number of dimensions of the joint vector.

[0089] For all variable pairs within each time window Repeating the above process, we obtain the directed transfer entropy matrix for this time window, expressed as:

[0090] .

[0091] In a specific embodiment, the construction process of the specific causal network for each time window is as follows:

[0092] Within each time window, using each variable as a node, starting from the first... The variable points to the first... The directed transitive entropy of the nth variable is used as the... The node points to the first The weights of the edges of each node are used to form the directed transit entropy matrix, which is the directed transit entropy matrix between any two variables. This matrix is ​​then used as the weighted directed adjacency matrix of the specific causal network for each time window, and a directed weighted graph is constructed.

[0093] Specifically, variables are mapped one-to-one with nodes, and the set of nodes is mapped one-to-one with all variables, using directed entropy transfer between nodes. The current time window is constructed using edge weights. A specific causal network. For any The elements of the directed transfer entropy matrix As from the first The node points to the first The weights of the edges of each node, in order to As a weighted directed adjacency matrix of the specific causal network for this time window, a directed weighted graph is constructed for all time windows. =1,2,…,K Repeat the above construction process to obtain a specific causal network sequence that varies with the time window. Each directed weighted graph reflects the causal relationship between variables within the corresponding time window, thus realizing the construction and representation of specific causal networks for each time window.

[0094] S3. Calculate the variational causal characteristics of the current time window based on the specific causal network of the current time window and the specific causal network of the previous time window. The variational causal characteristics include the dynamic causal fluctuation intensity of each node in the specific causal network and all nodes in its first-order neighborhood, the average specific transfer entropy coefficient between each node and any node in its first-order neighborhood, and the average specific transfer entropy coefficient between any node in its first-order neighborhood and any node in its second-order neighborhood. Calculate the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window based on the variational causal characteristics of each time window.

[0095] In a specific embodiment, the formula for calculating the dynamic causal fluctuation intensity of each node in the specific causal network and all nodes in its first-order neighborhood is as follows:

[0096] ;

[0097] in, Indicates the first Each node in the current time window The intensity of dynamic causal fluctuations of all nodes in the next first-order neighborhood set. For the first Each node in the current time window The next first-order neighborhood set, by the first It consists of nodes that are connected by edges. For the first Each node in the current time window The number of nodes in the next first-order neighborhood. For the first Each node in the current time window The average of the following values, For the first Each node in the current time window The standard deviation of the values ​​below, the first The node is the _th Each node in the current time window The next first-order neighbor node, For the first Each node in the current time window The average of the following values, For the first Each node in the current time window The standard deviation of the values ​​below;

[0098] The formula for calculating the average specific transfer entropy coefficient between each node in a specific causal network and any node in its first-order neighborhood is as follows:

[0099] ;

[0100] in, Indicates the first Each node and its position in the current time window The average value of the specific transfer entropy coefficient between each node in the next first-order neighborhood set. Indicates the previous time window From the inside The node points to the first The directed transitive entropy of each node ; Indicates the first Each node and its position in the current time window The first-order neighborhood set below The specific transfer entropy coefficient between nodes;

[0101] The formula for calculating the average idiosyncratic transfer entropy coefficient between any node in the first-order neighborhood and any node in the second-order neighborhood of each node in a idiosyncratic causal network is as follows:

[0102] ;

[0103] in, Indicates the first Each node in the current time window The average value of the specific transfer entropy coefficient between each node in the first-order neighborhood set and each node in the second-order neighborhood set; Indicates the first Each node in the current time window The first in the next first-order neighborhood set The node points to the first node in the second-order neighborhood set. The directed transitive entropy of each node Indicates the first The nodes in the previous time window The first-order neighborhood set under -1 The node points to the first node in the second-order neighborhood set. The directed transitive entropy of each node; Indicates the first Each node in the current time window The first-order neighborhood set below The node and its second-order neighborhood set The specific transfer entropy coefficient between nodes.

[0104] Specifically, embodiments of this application also require evaluating the mutation causal characteristics of each node in the specific causal network of the current time window. The specific process is as follows:

[0105] In the initial stage, when At that time, the data matrix of the reference time window is constructed using the preprocessed time series data of the sampling time corresponding to the time window; the corresponding directed transfer entropy matrix is ​​calculated based on the data matrix of the reference time window using the above method.

[0106] As the sampling time gradually increases, the time window With step size Slide forward to get different time windows. The data samples within the time window are also updated synchronously. Calculate the... The first variable and the second The data matrix corresponding to each variable in the current time window value, and The differences between them are due to changes in the time series data within the time window, representing the causal variability between the data matrix of the current time window and the data matrix of the previous time window. Therefore, the 1st... The first variable and the second The specific transfer entropy coefficient (STEC) of each variable in the current time window, denoted by the symbol... It is represented as shown in the following formula:

[0107] ;

[0108] First, regarding the first The first variable and the second The directed transit entropy of each variable is differencing between two adjacent time windows to obtain... To characterize adjacent time windows, from arrive Time The node (variable) and the first The change in directed transitive entropy between nodes (variables). Further... The absolute value is used to represent the magnitude of change, and it is used as a measure variable. The basis for determining whether causal relationships change significantly between adjacent time windows is then used to calculate subsequent causal variation characteristics.

[0109] Each node in the specific causal network within the current time window is used as a Dynamic Network Marker (DNM), and each node is assigned a causal variation potential dynamic network marker score (CVP-DNM score). Essentially, a DNM is a set of interacting nodes that exhibit three characteristics when the state of the specific causal network develops to near a critical state:

[0110] 1. The average dynamic causal fluctuation intensity of a node in a specific causal network and all nodes in its first-order neighborhood increases.

[0111] 2. The causal strength of nodes within a specific causal network and all nodes in its first-order neighborhood increases;

[0112] 3. The causal strength between nodes in the first-order neighborhood and nodes in the second-order neighborhood within a specific causal network decreases.

[0113] Therefore, the three features of dynamic network markers are quantified by assigning a dynamic network marker score to each node of a specific causal network based on its causal variation potential.

[0114] Specifically, for the first feature of DNM, through the first The node and all nodes in its first-order neighborhood (in a specific causal network, the node and all nodes in its first-order neighborhood) The variables connected by edges to each node constitute the first... The dynamic causal volatility intensity (DCVI) of the first-order neighborhood of each node is used for quantification, and is expressed as a symbol. Indicates; the second characteristic of DNM is through the first The quantification is based on the average of the specific transfer entropy coefficients of a node and all nodes in its first-order neighborhood over adjacent time windows, denoted by the symbol. The third characteristic of DNM is indicated by considering the first... The average of the specific transfer entropy coefficients between the first-order and second-order neighbors of a node in two adjacent time windows, denoted by the symbol . This indicates that the specific causal network is related to the first... The variables of the first-order neighborhood of each node are connected by edges to form the first-order neighborhood. The second-order neighborhood of each node.

[0115] S4. Calculate the comprehensive state index value for each time window based on the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window, and calculate the change in the comprehensive state index value between the current time window and the previous time window; determine whether the current time window is an early warning window based on the change in the comprehensive state index value between the current time window and the previous time window.

[0116] In a specific embodiment, the formula for calculating the dynamic network marker score of causal variation potential for each node in the specific causal network for each time window is as follows:

[0117] ;

[0118] in, Indicates the current time window The first specific causal network The dynamic network marker score of the causal variation potential of each node;

[0119] The formula for calculating the comprehensive status index value for each time window is as follows:

[0120] ;

[0121] in, Indicates the current time window The comprehensive state index value, where m represents the total number of variables;

[0122] The formula for calculating the change in the comprehensive status index value between the current time window and the previous time window is as follows:

[0123] ;

[0124] in, This indicates the change in the overall status index value between the current time window and the previous time window. Indicates the previous time window The comprehensive status index value.

[0125] Specifically, the dynamic network marker score for the causal variation potential of each node within each time window is calculated using the causal variation characteristics of each time window obtained above. Score.

[0126] In one example, by judging the condition Whether it is true or false determines whether the task has been completed. In a time-window-specific causal network, each node corresponds to... Calculation of scores. If the condition is not met, it indicates that the calculation is not completed, and steps S2 to S3 are repeated until the condition is met. If the condition is met, step S4 is entered to calculate the cumulative value of the scores corresponding to each node in the specific causal network as the comprehensive status indicator value of the wind turbine. Specifically, based on the scores of each node obtained through calculation within different time windows, on this basis, the cumulative value of the scores of all nodes in each time window is counted as the comprehensive status indicator (Comprehensive Status Indicator, ) value for evaluating the critical state transition of the wind turbine. Early fault warning of the wind turbine is realized based on the dynamic change of the value. When the change value of the comprehensive status indicator value of the current time window is greater than the warning threshold, it is considered that there is a significant mutation in the current time window, and the current time window is marked as a warning window; otherwise, the current time window is not a warning window. of the comprehensive status indicator value is greater than the warning threshold, it is considered that there is a significant mutation in the current time window , and the current time window is marked as a warning window; otherwise, the current time window is not a warning window.

[0127] Next, the embodiments of the present application are experimentally verified.

[0128] Taking the T01 wind turbine of a certain wind farm as an object, this wind turbine has a horizontal axis three-blade structure, a rated power of 2 MW, a rated wind speed of 12 m / s, and cut-in and cut-out wind speeds of 4 m / s and 25 m / s respectively. The rotor diameter of the wind wheel is 90 m, and its maximum rotational speed can reach 14.9 r / min. The gearbox adopts a three-stage planetary gear structure to drive an asynchronous generator, with a maximum rotational speed of 2016 r / min, a rated voltage of 690 V, and a grid-connected power frequency of 50 Hz. The tower of the wind turbine is made of steel pipe structure, and the hub height is 80 m. Data acquisition is completed by the SCADA system supporting the wind turbine, and the sampling period is 10 min. The total number of sampling points in this experiment is 2759. A sample data of 100 consecutive sampling moments is selected to construct a time window. The preprocessed time series data is processed through a sliding time window, and the window sliding step size is set to 1 sampling moment. According to the total number of sampling points, the time window length, and the window sliding step size, the total number of constructed time windows is 2660.

[0129] First, calculate the distribution of the scores of each variable in the time and variable dimensions, as shown in Figure 2 . The abscissa represents the variable number, the ordinate represents the time window number, and the vertical coordinate represents the score of the variable within the corresponding time window. The height of the column represents the score level. As shown in Figure 2It can be seen that, within most of the time window, the variables... The overall score remained at a relatively stable low-to-medium level, indicating that the system's variable contribution or anomaly intensity varied little over most periods, consistent with the normal operating status recorded in the wind turbine's operation log. According to the wind turbine's fault log, a fault occurred in window number 2629. Figure 3 It was observed that a few variables (such as variable 12) appeared near the time window number, around 2500. The bars corresponding to the scores rose significantly and formed spikes. Variable 12 is the average temperature of phase c of the high-voltage transformer, indicating that the abnormal contribution of this variable increased significantly within this time window, corresponding to a sudden change in the system state and a significant anomaly.

[0130] pass Figure 2 To observe the wind turbine under normal operating conditions, sampling times from the 100th to the 199th in the preprocessed time-series data were selected, and a 100th time window was constructed based on the set time window parameters. Within the 100th time window, the directed transfer entropy between variables was calculated using the directed transfer entropy method. This allowed for the determination of causal relationships between variables and the construction of specific causal networks among them. Figure 3 As shown. Figure 3 This is used to characterize the directional information transmission relationship between multiple variables based on directed transfer entropy within the 100th time window, i.e., the causal relationship between variables. Each circular node on the circumference represents a variable, and the number inside the node is the variable's number. The "Meaning of Number" on the right gives the one-to-one correspondence between the variable's number and the specific variable name. Figure 3 The directed edges in the network represent causal relationships between variables, with arrows pointing from the source node to the target node. This indicates that within the 100th time window, the source variable has a stronger informational contribution to the future changes of the target variable, i.e., there is a directional influence from the source node to the target node. The edges are a set of causal relationships filtered by directed transfer entropy (retaining the top 10% of variables with TE values), thus expressing the dynamic coupling structure of the original multivariate time-series data within this time window in a visual network form. By observing the distribution of incoming and outgoing edges of each node, for example, nodes representing generator phase current and high-voltage transformer phase voltage have a large number of incoming and outgoing edges, indicating that these variables are in a core position in the system's information flow. Further analysis reveals that the causal connections between electrical nodes are particularly dense, forming the backbone of the network, while temperature nodes are mostly at the end of the causal chain, which is consistent with the physical characteristic that the system's thermal dynamic response lags behind electrical changes. Specific causal networks reveal strong causal relationships, such as those from generators to high-voltage transformers. They can identify variables that play a major driving role in the window (those with more outgoing edges) and variables that are more concentratedly affected (those with more incoming edges), thus providing a basis for system state interpretation, causal link tracing, and subsequent fault mechanism analysis and early warning decision-making.

[0131] pass Figure 2 Observing the near-fault state of the wind turbine generator, the 2500th to 2600th sampling times from the preprocessed SCADA data are selected, and a 2500th time window is constructed according to the aforementioned window parameters. Within the 2500th time window, the transfer entropy between each variable is calculated using the transfer entropy method, the causal relationship between each variable is determined, and a specific causal network between each variable is constructed, such as... Figure 4 As shown. (Through) Figure 4 A wide range of dynamic coupling relationships can be observed among the thermal parameters during system operation. Significant causal relationships exist between temperature nodes at different measurement points within the same component. For example, different cooling temperature monitoring points in the gearbox form highly interconnected substructures, reflecting the transfer and balancing process of heat distribution within the component. Nodes representing average power generation have directional or reciprocal causal connections with temperature nodes in multiple cooling systems, indicating a statistically identifiable interaction mechanism between the system's output power and the thermal state of key components. Furthermore, causal links also exist between temperature nodes for generator cooling and transformer cooling, representing system-level thermal interaction paths. This specific causal network not only visually demonstrates the information flow of thermal parameters within the network structure but also directly identifies the causal relationship between temperature and the system's final output power from operational data.

[0132] Figures 3 to 4 This evolution process represents the transition of a wind turbine from a normal operating state to a critical fault state. Under normal operating conditions, the causal relationships between variables are primarily reflected in electrical and mechanical quantities, indicating normal energy conversion and transfer paths. However, under critical fault conditions, the causal relationships between temperature variables become extremely prominent, forming multiple tightly coupled thermally correlated subnetworks. Simultaneously, significant causal connections emerge between temperature nodes and system output power, suggesting that the fault may trigger thermal anomalies in the system, and changes in thermal state directly affect the system's final power generation performance. Furthermore, the number of edges in the specific causal network under fault conditions decreases. This is because the top 10% of variables with the highest TE values ​​are retained as the causal relationship set after directed transfer entropy filtering, and the 2500th time window under fault conditions has one more node (variable 24), representing the average power generation measured at the grid connection point, compared to the 100th time window under normal operating conditions. However, the connections between temperature nodes are more concentrated, indicating that the fault may weaken or interrupt some electrical causal relationships in the system, while thermodynamic causality becomes dominant. This evolutionary analysis shows that specific causal networks based on directed transfer entropy can sensitively capture fundamental changes in the operating state of a system, and the differences in specific causal networks under normal and critical fault states provide an intuitive and powerful basis for fault detection and diagnosis.

[0133] The above calculations yield the corresponding variable nodes in the specific causal network within the time window. Score, its The scores are accumulated to obtain the Comprehensive Condition Index (CSI) value for the wind turbine. The resulting CSI value trend graph over a time window is shown below. Figure 5 As shown, the horizontal axis represents the time window number, and the vertical axis represents the CSI value of the corresponding time window. The black curve represents the continuous change of the CSI value for each time window, the red dashed line represents the warning threshold (CSI>100), and the red triangle is used to identify the time window that triggers the warning threshold. This is achieved by defining the values ​​of each node within the time window. The cumulative score serves as the comprehensive status index (CSI) of the wind turbine. As shown in Figure 5, the CSI value remained at a low level and fluctuated slightly for most time windows, with only a few windows exhibiting sudden spikes. However, the overall trend remained stable without any abnormal changes exceeding the warning threshold. A significantly higher peak value appeared in time window 2519, exceeding the warning threshold and thus marked as a warning window, indicating a strong abnormal change in the system's operating status during the corresponding period. According to the wind turbine's fault log, the turbine failed in time window 2629. This demonstrates that the alarm was triggered in time window 2519, 18 hours and 20 minutes earlier than the actual fault window, effectively providing early warning of wind turbine faults.

[0134] Further reference Figure 6 As an implementation of the methods shown in the above figures, this application provides an embodiment of a wind turbine early fault warning device based on a specific causal network. This device embodiment is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0135] This application provides an early fault warning device for wind turbines based on a specific causal network, comprising:

[0136] Preprocessing module 1 is configured to acquire and preprocess the time series data of all numerical variables continuously monitored in the SCADA system of the wind turbine to obtain preprocessed time series data; and construct a data matrix of several time windows based on the preprocessed time series data using a sliding window.

[0137] The specific causal network construction module 2 is configured to calculate the directed transit entropy between any two variables in each time window based on the data matrix of each time window using conditional mutual information, and to construct a specific causal network for each time window.

[0138] Feature calculation module 3 is configured to calculate the variant causal features of the current time window based on the specific causal network of the current time window and the specific causal network of the previous time window. The variant causal features include the dynamic causal fluctuation intensity of each node in the specific causal network and all nodes in its first-order neighborhood, the average specific transfer entropy coefficient between each node and any node in its first-order neighborhood, and the average specific transfer entropy coefficient between any node in its first-order neighborhood and any node in its second-order neighborhood. Based on the variant causal features of each time window, the module calculates the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window.

[0139] The early warning module 4 is configured to calculate the comprehensive state index value of each time window based on the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window, and calculate the change value of the comprehensive state index value between the current time window and the previous time window; and determine whether the current time window is an early warning window based on the change value of the comprehensive state index value between the current time window and the previous time window.

[0140] Figure 7 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. For example... Figure 7 As shown, the electronic device of this embodiment includes a processor 701 and a memory 702; wherein the memory 702 is used to store computer execution instructions; and the processor 701 is used to execute the computer execution instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant descriptions in the foregoing method embodiments.

[0141] Alternatively, the memory 702 can be either standalone or integrated with the processor 701.

[0142] When the memory 702 is set up independently, the electronic device also includes a bus 703 for connecting the memory 702 and the processor 701.

[0143] This invention also provides a computer storage medium storing computer execution instructions, which, when executed by processor 701, implement the above method.

[0144] This invention also provides a computer program product, including a computer program that, when executed by a processor 701, implements the above-described method.

[0145] In the embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0146] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0147] Furthermore, the functional modules in the various embodiments of this invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit formed by the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0148] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor 701 to execute some steps of the methods of the various embodiments of this application.

[0149] It should be understood that the processor 701 described above can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor, or the processor 701 can be any conventional processor 701. The steps of the method disclosed in this invention can be directly manifested as the hardware processor 701 executing the steps, or as a combination of hardware and software modules within the processor 701 executing the steps.

[0150] The memory 702 may include high-speed RAM memory, and may also include non-volatile memory NVM, such as at least one disk storage device, and may also be a USB flash drive, portable hard drive, read-only memory, disk or optical disc, etc.

[0151] Bus 703 can be an Industry Standard Architecture (ISA), a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Bus 703 can be divided into address bus, data bus, control bus, etc. For ease of illustration, the bus 703 in the accompanying drawings of this application is not limited to only one bus 703 or one type of bus 703.

[0152] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0153] An exemplary storage medium is coupled to a processor 701, enabling the processor 701 to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor 701. The processor 701 and the storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor 701 and the storage medium can exist as discrete components in an electronic device or a host device.

[0154] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early fault warning of wind turbine units based on specific causal networks, characterized in that, Includes the following steps: The time-series data of all numerical variables continuously monitored in the SCADA system of the wind turbine are acquired and preprocessed to obtain preprocessed time-series data; a data matrix of several time windows is constructed using a sliding window based on the preprocessed time-series data. Based on the data matrix of each time window, the directed transit entropy between any two variables in each time window is calculated using conditional mutual information, and a specific causal network for each time window is constructed. The variational causal characteristics of the current time window are calculated based on the specific causal network of the current time window and the specific causal network of the previous time window. The variational causal characteristics include the dynamic causal fluctuation intensity of each node in the specific causal network and all nodes in its first-order neighborhood, the average specific transfer entropy coefficient between each node and any node in its first-order neighborhood, and the average specific transfer entropy coefficient between any node in its first-order neighborhood and any node in its second-order neighborhood. Based on the variational causal characteristics of each time window, the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window is calculated. The comprehensive state index value for each time window is calculated based on the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window, and the change in the comprehensive state index value between the current time window and the previous time window is calculated; based on the change in the comprehensive state index value between the current time window and the previous time window, it is determined whether the current time window is an early warning window.

2. The wind turbine early fault warning method based on specific causal networks according to claim 1, characterized in that, Based on the data matrix of each time window, the directed transit entropy between any two variables within each time window is calculated using conditional mutual information, specifically including: Let the data matrix for each time window be: ; in, The length of each time window, The total number of variables. Indicates the first The first variable within the time window The value at each moment. ; Based on the data matrix for each time window, construct sample triples for any two variables, represented as: ;in, The total number of sampling moments within each time window. , For the time lag, This indicates taking the first element from the data matrix in each time window. The variables range from the first sampling time to the second sampling time. The first vector is composed of the values ​​at each sampling time. This indicates taking the first element from the data matrix in each time window. The variable from the first From the sampling time to the... The second vector is composed of the values ​​at each sampling time. This indicates taking the first element from the data matrix in each time window. The variables range from the first sampling time to the second sampling time. The third vector is composed of the values ​​at each sampling time. In the current time window Within, given the order of nearest neighbors From the sample triplet, the first The value at the sampling time and the first sampling time The values ​​at each sampling time point are respectively used to construct the first... Joint vector at each sampling time point and the Joint vector at each sampling time point , where [] indicates splicing, , and The first vector is respectively the first The element and the first One element, and The second vector is respectively the first The element and the first One element, and The third vector is respectively the first The element and the first Each element; a joint space is constructed using the joint vector. ; In the joint space Internal search nearest neighbor radius Based on the nearest neighbor radius, respectively in the subspace , and The internal neighbor count is represented as: ； ； ； in, This represents the maximum norm distance between two vectors. Represents the count of a set. , and They represent the first Centered on each sample point in the subspace , and Falling within the nearest radius The number of other sample points in the neighborhood; The directed transitive entropy is calculated explicitly using KSG estimation based on conditional mutual information, as shown in the following equation: ； in, Indicates the current time window From the inside The variable points to the first... The directed transitive entropy of each variable, For conditional mutual information, This is the digamma function.

3. The wind turbine early fault warning method based on specific causal networks according to claim 1, characterized in that, The construction process of the specific causal network for each time window is as follows: Within each time window, using each variable as a node, starting from the first... The variable points to the first... The directed transitive entropy of the nth variable is used as the... The node points to the first The weights of the edges of each node are used to form the directed transit entropy matrix, which is the directed transit entropy matrix between any two variables. This matrix is ​​then used as the weighted directed adjacency matrix of the specific causal network for each time window, and a directed weighted graph is constructed.

4. The wind turbine early fault warning method based on specific causal networks according to claim 3, characterized in that, The formula for calculating the dynamic causal fluctuation intensity of each node and all nodes in its first-order neighborhood in the specific causal network is as follows: ； in, Indicates the first Each node in the current time window The dynamic causal fluctuation intensity of all nodes in the next first-order neighborhood set. For the first Each node in the current time window The next first-order neighborhood set, by the first It consists of nodes that are connected by edges. For the first Each node in the current time window The number of nodes in the next first-order neighborhood. For the first Each node in the current time window The average of the following values, For the first Each node in the current time window The standard deviation of the values ​​below, the first The node is the _th Each node in the current time window The next first-order neighbor node, For the first Each node in the current time window The average of the following values, For the first Each node in the current time window The standard deviation of the values ​​below; The formula for calculating the average value of the specific transfer entropy coefficient between each node in the specific causal network and any node in its first-order neighborhood is as follows: ； in, Indicates the first Each node and its position in the current time window The average value of the specific transfer entropy coefficient between each node in the next first-order neighborhood set. Indicates the previous time window From the inside The node points to the first The directed transitive entropy of each node ; Indicates the first Each node and its position in the current time window The first-order neighborhood set below The specific transfer entropy coefficient between nodes; The formula for calculating the average value of the specific transfer entropy coefficient between any node in the first-order neighborhood and any node in the second-order neighborhood of each node in the specific causal network is as follows: ； in, Indicates the first Each node in the current time window The average value of the specific transfer entropy coefficient between each node in the first-order neighborhood set and each node in the second-order neighborhood set; Indicates the first Each node in the current time window The first in the next first-order neighborhood set The node points to the first node in the second-order neighborhood set. The directed transitive entropy of each node Indicates the first The nodes in the previous time window The first-order neighborhood set under -1 The node points to the first node in the second-order neighborhood set. The directed transitive entropy of each node; Indicates the first Each node in the current time window The first-order neighborhood set below The node and its second-order neighborhood set The specific transfer entropy coefficient between nodes.

5. The wind turbine early fault warning method based on specific causal networks according to claim 4, characterized in that, The formula for calculating the dynamic network biomarker score of causal variation potential for each node in the specific causal network for each time window is as follows: ； in, Indicates the current time window The first specific causal network The dynamic network marker score of the causal variation potential of each node; The formula for calculating the comprehensive status index value for each time window is as follows: ； in, Indicates the current time window The comprehensive state index value, where m represents the total number of variables; The formula for calculating the change in the comprehensive status index value between the current time window and the previous time window is as follows: ； in, This indicates the change in the overall status index value between the current time window and the previous time window. Indicates the previous time window The comprehensive status index value.

6. The wind turbine early fault warning method based on specific causal networks according to claim 1, characterized in that, The preprocessing process includes normalization, removal of outliers using the local anomaly factor algorithm, and filling in the removed values ​​using the K-nearest neighbor algorithm.

7. A wind turbine early fault warning device based on a specific causal network, characterized in that, include: The preprocessing module is configured to acquire and preprocess the time-series data of all numerical variables continuously monitored in the SCADA system of the wind turbine to obtain preprocessed time-series data; and to construct a data matrix of several time windows using a sliding window based on the preprocessed time-series data. The specific causal network construction module is configured to calculate the directed transit entropy between any two variables in each time window using conditional mutual information based on the data matrix of each time window, and construct the specific causal network for each time window. The feature calculation module is configured to calculate the variant causal features of the current time window based on the specific causal network of the current time window and the specific causal network of the previous time window. The variant causal features include the dynamic causal fluctuation intensity of each node in the specific causal network and all nodes in its first-order neighborhood, the average value of the specific transfer entropy coefficient between each node and any node in its first-order neighborhood, and the average value of the specific transfer entropy coefficient between any node in its first-order neighborhood and any node in its second-order neighborhood. The dynamic network marker score of causal variation potential for each node in the specific causal network of each time window is calculated based on the variation causal characteristics of each time window. The early warning module is configured to calculate the comprehensive state index value of each time window based on the dynamic network marker score of the causal variation potential of each node in the specific causal network of each time window, and calculate the change value of the comprehensive state index value between the current time window and the previous time window; and determine whether the current time window is an early warning window based on the change value of the comprehensive state index value between the current time window and the previous time window.

8. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

Images