Wind turbine generator fault intelligent diagnosis method and system based on structure adaptive width learning network

Abstract

The invention provides a wind turbine generator fault intelligent diagnosis method and system based on a structure adaptive width learning network, and the method comprises the following steps: S1, carrying out the preprocessing of collected historical fault data of a wind turbine generator, and obtaining a processed data set X and a fault label Y; s2, training an initial width learning network by using the cleaned data set X and the fault tag Y as training samples to obtain an initial fault diagnosis model; s3, performing structure expansion on the initial width learning network for T times, adding a reinforcing node on the basis of the original width learning network for structure expansion each time, updating the original width learning network to obtain an updated width learning network, obtaining a width learning network after structure self-adaptive updating after the structure expansion for T times, and performing structure self-adaptive updating on the width learning network after the structure self-adaptive updating; obtaining a final fault diagnosis model; and S4, inputting a data sample of a to-be-diagnosed wind turbine generator into the final fault diagnosis model to obtain a diagnosis result. The diagnosis capability of the model can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of fault prediction and health management technology for wind turbines, and in particular to an intelligent fault diagnosis method and system for wind turbines based on a structure-adaptive width learning network. Background Technology

[0002] Wind turbines convert wind energy into electrical energy. Due to prolonged operation in extreme environments, wind turbines inevitably experience malfunctions. Therefore, accurate fault diagnosis of wind turbines is crucial for ensuring safe operation, reducing maintenance costs, and improving wind power production efficiency.

[0003] In recent years, with the widespread adoption of Supervisory Control and Data Acquisition (SCADA) systems and the rapid development of artificial intelligence technologies, particularly deep learning, big data-driven intelligent fault diagnosis methods have become a research hotspot in the wind power industry. Due to their ability to automatically extract fault features and perform end-to-end modeling, these methods significantly reduce labor costs and improve diagnostic accuracy compared to traditional fault diagnosis methods.

[0004] However, existing intelligent diagnostic methods for wind turbines generally suffer from difficulties in adjusting model structure parameters. This leads to the need for trial and error in practical applications, which is time-consuming and labor-intensive, resulting in poor model diagnostic capabilities and hindering practical application. Furthermore, SCADA data sampling intervals are typically only a few seconds to tens of seconds, resulting in extremely low sampling frequencies. This leads to serious problems of misdiagnosis and missed diagnosis in existing methods that rely on a single sampling time point for fault diagnosis. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network, which can improve the diagnostic capability of the model.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for intelligent fault diagnosis of wind turbines based on a structure-adaptive width learning network includes the following steps:

[0008] S1. Collect historical fault data of wind turbine units, clean the abnormal data in the historical fault data to obtain dataset X, and label the cleaned dataset X according to different fault states to obtain fault label Y.

[0009] S2, using dataset X and fault labels Y as training samples, trains an initial width learning network. To obtain the initial fault diagnosis model;

[0010] S3, perform T structural expansions on the initial width learning network. During each expansion, add [a certain number of elements] to the original width learning network. Each reinforcement node performs structural expansion, updating the original width learning network to obtain the updated width learning network. After T structural expansions, a structurally adaptively updated width learning network is obtained. To obtain the final fault diagnosis model;

[0011] S4 inputs the data sample of the wind turbine to be diagnosed into the final fault diagnosis model to obtain the diagnosis result.

[0012] Further, step S2 includes the following steps:

[0013] S21, Construct a series of mapping nodes to perform linear feature extraction and merging on the data in dataset X to obtain the mapping features. ;

[0014] S22, Construct a series of enhanced node pairs mapping features Further nonlinear feature extraction and merging are performed to obtain enhanced features. ;

[0015] S23, mapping features and enhanced features Merge the features to obtain the feature set. ;

[0016] S24, Construct the loss function and use ridge regression theory to train the initial width learning network. Solve the problem to obtain an initial fault diagnosis model.

[0017] Furthermore, the loss function is constructed in step S24 as follows:

[0018] ;

[0019] Where λ represents the regularization parameter to prevent overfitting.

[0020] Furthermore, in step S24, ridge regression theory is used to train the initial width learning network. The solution is as follows:

[0021] ;

[0022] Where I represents the identity matrix; the superscript T represents the matrix transpose; and the superscript –1 represents the matrix inverse.

[0023] Furthermore, in step S3, the total number of expansions T and the width are determined using a structural adaptive expansion strategy, and the number of reinforcement nodes that the learning network needs to add when performing the t-th structural expansion are learned. , t=1,2,…,T.

[0024] Furthermore, by utilizing a structural adaptive expansion strategy, the total number of expansions T and the width are determined, and the number of reinforcement nodes required for the network to perform the t-th structural expansion is learned. The steps include:

[0025] S31, let there be K opportunities to determine the number of newly added reinforced nodes during the t-th structural expansion, and let the number of newly added reinforced nodes corresponding to the K opportunities be denoted as . The number of new reinforcement nodes is determined by a multiplicative increase rule, and the training accuracy of the model after adding l reinforcement nodes in the t-th structural expansion is denoted as . The accuracy of the previous training session is denoted as ;

[0026] S32, for the first opportunity, add... A reinforced node, if Then let , This indicates that at the t-th incremental structural learning iteration, there is A reinforcement node is added to the original width learning network to obtain the updated width learning network. Then, the updated width learning network enters the (t+1)th structural expansion; if This opens the second chance. Add a reinforcement node to the original model, if Then set , This indicates that during the t-th incremental structural learning iteration, l2 reinforcement nodes are added to the original width learning network, resulting in an updated width learning network. Then, the updated width learning network enters the (t+1)-th structural expansion iteration. If it succeeds, then the next opportunity will be opened; and so on, for... If the k-th opportunity occurs for the first time When, then set , Based on the above rules, the number of reinforcing nodes added during the t-th structural expansion can be obtained:

[0027] ;

[0028] S33, for the t-th structural expansion, if all The adaptive structural expansion process is terminated, and the width learning network obtained from the (t-1)th structural expansion is selected as the final fault diagnosis model.

[0029] Furthermore, the steps for updating the width learning network include:

[0030] Feature extraction is performed using the newly added l enhancement nodes to obtain the extracted nonlinear features. ;

[0031] Will With feature set Perform a merge update to obtain the updated feature set. ;

[0032] The width learning network is updated using the block matrix pseudo-inverse algorithm to obtain the updated width learning network. :

[0033] ;

[0034] in, .

[0035] Furthermore, With feature set The merge and update should be performed according to the following formula:

[0036] ;

[0037] in, This is the updated feature set.

[0038] Further, step S4 includes the following steps:

[0039] S41, collect a full day's worth of SCADA data from the wind turbine to be diagnosed, and perform cleaning and pre-processing.

[0040] S42, input the data from each sampling moment in the SCADA data processed in step S41 into the final fault diagnosis model to obtain the diagnosis result at the current sampling moment;

[0041] S43 votes on the diagnostic results from all sampling times throughout the day, and outputs the integrated diagnostic results of wind turbine faults based on the principle of majority rule.

[0042] Furthermore, this invention provides a wind turbine fault intelligent diagnosis system based on a structure-adaptive width learning network, comprising:

[0043] The data acquisition module is configured to collect historical fault data of wind turbine units;

[0044] The initial fault diagnosis model building module is configured to build an initial fault diagnosis model based on historical fault data collected by the acquisition module, and to train the initial fault diagnosis model.

[0045] The structural adaptive expansion module is configured to update the established initial diagnostic model online using a structural adaptive expansion strategy.

[0046] The adaptive termination module is configured to terminate the structural adaptive expansion module according to the set online update termination conditions and output the final fault diagnosis model.

[0047] The integrated diagnostic module is configured to collect data samples from the wind turbine to be diagnosed and input them into the final fault diagnosis model to obtain diagnostic results.

[0048] By adopting the above technical solution, the present invention has the following beneficial effects:

[0049] 1. Unlike existing intelligent diagnostic methods for wind turbines that struggle with model parameter selection, this invention optimizes model parameters through a structural adaptive expansion strategy based on a wide learning network, effectively improving the model's diagnostic capabilities.

[0050] 2. Unlike existing intelligent diagnostic methods for wind turbines that require iterative training for model parameter tuning, this invention uses a block matrix pseudo-inverse operation rule for online incremental updates, avoiding a lengthy iteration process and greatly improving modeling efficiency.

[0051] 3. Unlike existing intelligent diagnostic methods for wind turbines that rely on independent SCADA data at a single point in time, this invention utilizes an integrated diagnostic strategy based on daily data to vote on diagnostic results, thus avoiding missed diagnoses and misdiagnoses caused by data fluctuations and improving diagnostic accuracy. Attached Figure Description

[0052] Figure 1 This is a flowchart of a preferred embodiment of the present invention, which describes an intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network.

[0053] Figure 2 This is a diagram showing the original SCADA data in an embodiment of the present invention. Figure 2 (a) is a graph showing the relationship between wind speed and power. Figure 2 (b) A graph showing the change in the main bearing temperature 1 at test point 1.

[0054] Figure 3 For the purposes of this embodiment of the invention Figure 2 The image shown is a display of the SCADA data after cleaning and preprocessing. Figure 3 (a) is a graph showing the relationship between wind speed and power after cleaning and pretreatment. Figure 3 (b) is a diagram showing the temperature change of the main bearing 1 after cleaning and pretreatment.

[0055] Figure 4 This diagram illustrates the adaptive expansion process of the structure in an embodiment of the present invention.

[0056] Figure 5 This is a block diagram of a wind turbine fault intelligent diagnosis system based on a structure-adaptive width learning network, which is a preferred embodiment of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] In the description of this invention, it should be understood that the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0059] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0060] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0061] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0062] Please see Figure 1 A preferred embodiment of the present invention provides a method for intelligent fault diagnosis of wind turbines based on a structure-adaptive width learning network, comprising the following steps:

[0063] S1. Collect historical fault data of wind turbine units, clean the abnormal data in the historical fault data to obtain dataset X, and label the cleaned dataset X according to different fault states to obtain fault labels Y.

[0064] In this embodiment, SCADA data of the wind turbine under different fault modes is collected as historical fault data through a Supervisory Control and Data Acquisition (SCADA) system. SCADA data includes temperature parameters (ambient temperature, generator temperature, bearing temperature, etc.), vibration parameters (nacelle vibration, bearing vibration, etc.), velocity parameters (wind speed, bearing speed, generator speed, etc.), and electrical parameters (generator power, voltage, current, etc.). Taking the fault diagnosis of the wind turbine main bearing as an example, SCADA sample data of the main bearing under different fault modes is collected, including 27 state parameters such as temperature parameters (ambient temperature, generator temperature, bearing temperature, etc.), vibration parameters (nacelle vibration, etc.), and electrical parameters (generator power, voltage, current, etc.), with a sampling interval of 7 seconds. The collected fault modes and sample sizes are as follows: 100 days of normal operation, 97 days of main bearing wear, 146 days of main bearing cracking, and 82 days of main bearing breakage. The data collected in the first half of the days is used as the training set to train the model, and the data collected in the second half of the days is used as the test set to test the model.

[0065] In this implementation, abnormal data in the historical fault data is cleaned and preprocessed to remove invalid and abnormal data (such as zero-power accumulation data, constant-power power-limiting data, outlier abnormal data, etc.). Figure 2 As shown), to improve data availability, the specific steps are as follows:

[0066] To address the accumulation of zero-power data, the wind turbine status monitoring variables in the SCADA system were used to clean the zero-power data. The SCADA system monitors seven wind turbine statuses: 0: Initialization, 1: Shutdown Process, 2: Shutdown, 3: Standby, 4: Startup, 5: Grid-connected Generation, and 9: Maintenance. Statuses 0, 1, 2, 3, 4, and 9 are all causes of zero-power data accumulation. Therefore, all monitoring data except for status 5 were removed to clean up the zero-power data accumulation.

[0067] For constant power curtailment data, the available state variables in the SCADA system are used to clean the data. The available state of the wind speed-power ratio data monitored by the SCADA system is divided into two states: "0: unavailable" and "1: available". By removing data with an available state of 0, constant power curtailment data in the wind speed-power graph can be effectively cleaned up.

[0068] For outlier data, the 3σ principle is used for cleaning. First, the mean μ and standard value σ of each variable are calculated, and the data distributed in the range of... Data that falls outside the 3σ range is considered normal data, while data that falls outside the 3σ range is considered abnormal data and is removed.

[0069] In this embodiment, the cleaned SCADA data ( Figure 3 As a dataset X, the dataset X is labeled according to different fault states of the main bearing to obtain fault labels Y.

[0070] S2, using the cleaned dataset X and fault labels Y as training samples, trains the initial width learning network. An initial fault diagnosis model is obtained.

[0071] In this embodiment, let the training samples be... ,in, , represents a data matrix consisting of N samples, each sample containing L variables (for time-domain signals, L represents the signal length), and R represents the real number domain; Let represent the label matrix of N samples, where the superscript C represents the number of classes. Assume the BLN mapping nodes have n groups, and each group contains... There are m mapping nodes and m groups of reinforced nodes, each group containing There are several reinforced nodes. Step S2 then includes the following steps:

[0072] S21, Construct a series of mapping nodes to perform linear feature extraction and merging on the data in dataset X to obtain the mapping features. .

[0073] Specifically, firstly, a series of mapping nodes are constructed to extract features from the dataset X, extracting a series of linear features. :

[0074] ;

[0075] Among them, z i Represents the i-th group of linear features; and This represents the weight matrix and bias of the randomly generated i-th group of mapping nodes.

[0076] Subsequently, a series of linear features were extracted. Merge the samples to obtain the mapping features of X. :

[0077] ;

[0078] In this embodiment, , .

[0079] S22, Construct a series of enhanced node pairs mapping features Further nonlinear feature extraction and merging are performed to obtain enhanced features. .

[0080] Specifically, in mapping features Based on this, further nonlinear feature extraction is performed through enhanced nodes to transform it into a series of enhanced features. :

[0081] ;

[0082] in, Represents the enhanced features of the j-th group; , representing the nonlinear Sigmoid activation function; and Let represent the weight matrix and bias of the randomly generated j-th group of reinforcement nodes, respectively.

[0083] Enhancement features of each group Merge to obtain enhanced features :

[0084] ;

[0085] In this embodiment, , .

[0086] S23, mapping features and enhanced features Merge the features to obtain the feature set. In this embodiment, the mapping features are mapped according to the following formula. and enhanced features Merge the features to obtain the feature set. :

[0087] ;

[0088] S24, based on the feature set A loss function is constructed by connecting it to the fault label Y, and ridge regression theory is used to learn the initial width of the network. Solve the problem to obtain an initial fault diagnosis model.

[0089] In this embodiment, the feature set Connected to fault label Y via a linear connection:

[0090]

[0091] Then construct the loss function:

[0092]

[0093] Where λ represents the regularization parameter to prevent overfitting.

[0094] Using ridge regression theory to learn the initial width of the network To solve this problem, we need to apply ridge regression theory. It can be approximated as:

[0095] ;

[0096] Where I represents the identity matrix; the superscript T denotes the matrix transpose; and the superscript -1 denotes the matrix inverse. In this embodiment, let... It is approximately equal to 0.

[0097] The characteristic matrix A is calculated according to the following formula. m The false inverse:

[0098] ;

[0099] The superscript + indicates the pseudo-inverse of the matrix.

[0100] S3, perform T structural expansions on the initial width learning network. During each expansion, add [a certain number of elements] to the original width learning network. Each reinforcement node performs structural expansion, updating the original width learning network to obtain the updated width learning network. After T structural expansions, a structurally adaptively updated width learning network is obtained. This leads to the final fault diagnosis model.

[0101] In this embodiment, a structural adaptive expansion strategy is used to determine the total number of expansions T and the number of reinforcement nodes that the width learning network needs to add when performing the t-th structural expansion. For t=1,2,…,T, the specific steps include:

[0102] S31, let there be K opportunities to determine the number of newly added reinforced nodes during the t-th structural expansion, and let the number of newly added reinforced nodes corresponding to the K opportunities be denoted as . The number of new reinforcement nodes is determined by a multiplicative increase rule, and the training accuracy of the model after adding l reinforcement nodes in the t-th structural expansion is denoted as . The accuracy of the previous training session is denoted as In this embodiment, let , ;

[0103] S32, for the first opportunity, add... A reinforced node, if Then let , This indicates that at the t-th incremental structural learning iteration, there is A reinforcement node is added to the original width learning network to obtain the updated width learning network. Then, the updated width learning network enters the (t+1)th structural expansion; if This opens the second chance. Add a reinforcement node to the original model, if Then set , This indicates that during the t-th incremental structural learning iteration, l2 reinforcement nodes are added to the original width learning network, resulting in an updated width learning network. Then, the updated width learning network enters the (t+1)-th structural expansion iteration. If it succeeds, then the next opportunity will be opened; and so on, for... If the k-th opportunity occurs for the first time When, then set , Based on the above rules, the number of reinforcing nodes added during the t-th structural expansion can be obtained:

[0104] ;

[0105] S33, for the t-th structural expansion, if all The adaptive structural expansion process is terminated, and the width learning network obtained from the (t-1)th structural expansion is selected as the final fault diagnosis model.

[0106] In this embodiment, the step of updating the width learning network includes:

[0107] Utilizing the newly added Feature extraction is performed on each enhanced node to obtain the extracted nonlinear features. Specifically, suppose a new set of reinforcement nodes is added to the width learning model, and let the number of new reinforcement nodes be denoted as . The newly added enhancement features are ,but It can be calculated using the following formula:

[0108] ;

[0109] in, and This represents the weight matrix and bias of the newly added enhanced features.

[0110] Will With feature set Perform a merge update to obtain the updated feature set. In this embodiment, With feature set The merge and update should be performed according to the following formula:

[0111] ;

[0112] in, For the updated feature set, .

[0113] The width learning network is updated using the block matrix pseudo-inverse algorithm to obtain the updated width learning network. Specifically, according to the operational rules of the pseudo-inverse of a block matrix, Available It is obtained by updating based on the above, that is:

[0114] ;

[0115] in, ;

[0116] In the formula, 1 represents a matrix of all 1s; 0 represents a zero matrix;

[0117] Calculated Then, the width learning network is updated online according to the following formula:

[0118] .

[0119] S4, input the data sample of the wind turbine to be diagnosed into the final fault diagnosis model to obtain the diagnosis result. In this embodiment, integrated diagnosis is performed on a daily basis, specifically including the following steps:

[0120] S41, Collect SCADA data of the wind turbine to be diagnosed for a whole day, and clean and preprocess it; In this embodiment, the test set is regarded as the sample data of the wind turbine to be diagnosed, and the data is cleaned and preprocessed. The processing procedure is detailed in step S1.

[0121] S42, input the data from each sampling moment in the SCADA data processed in step S41 into the final fault diagnosis model to obtain the diagnosis result at the current sampling moment;

[0122] S43 votes on the diagnostic results from all sampling times throughout the day, and outputs the integrated diagnostic results of wind turbine faults based on the principle of majority rule.

[0123] Please see also Figure 5 This invention also provides a wind turbine fault intelligent diagnosis system based on a structure-adaptive width learning network, comprising:

[0124] The data acquisition module is configured to collect historical fault data of wind turbine units;

[0125] The initial fault diagnosis model building module is configured to build an initial fault diagnosis model based on historical fault data collected by the acquisition module, and to train the initial fault diagnosis model.

[0126] The structural adaptive expansion module is configured to update the established initial diagnostic model online using a structural adaptive expansion strategy.

[0127] The adaptive termination module is configured to terminate the adaptive structural expansion module according to a set online update termination condition and output the final fault diagnosis model. In this embodiment, the online update termination condition is: for the t-th structural expansion, if all This terminates the adaptive expansion process of the structure.

[0128] The integrated diagnostic module is configured to collect data samples from the wind turbine to be diagnosed and input them into the final fault diagnosis model to obtain diagnostic results.

[0129] The technical effects of the present invention will be described in detail below based on experimental results.

[0130] 1. Operating conditions:

[0131] It features an Intel Xeon Platinum 8255C CPU (2.50 GHz) and 128 GB of RAM, with the software platform MatlabR2019b.

[0132] 2. Results Analysis:

[0133] Figure 4 The process of adaptive expansion and updating of the structure in this invention is demonstrated. It can be seen that after initial model training, the model undergoes 18 adaptive expansion operations to obtain the final diagnostic model. Throughout the process, the number of nodes added is selected using an adaptive expansion method, and the termination point can be adaptively determined without human intervention or selection. Simultaneously, it can be observed that the training accuracy gradually increases with structural expansion, indicating that this invention provides positive benefits and improves modeling efficiency in the fault diagnosis modeling process.

[0134] The integrated diagnostic method of this invention, measured in days, was validated using a test set. The final test accuracy was 78.70%, while the diagnostic accuracy of the traditional independent diagnostic method was 69.81%. It can be seen that the integrated diagnostic method measured in days can significantly improve diagnostic accuracy. This is because the integrated diagnostic method measured in days uses a voting process, which can effectively filter out missed diagnoses and misdiagnoses caused by data fluctuations, thereby improving diagnostic accuracy.

[0135] Unlike existing intelligent diagnostic methods for wind turbines that struggle with model parameter selection, this invention optimizes model parameters through a structural adaptive expansion strategy based on a wide learning network, effectively improving the model's diagnostic capabilities.

[0136] Unlike existing intelligent diagnostic methods for wind turbines that require iterative training for model parameter tuning, this invention uses a block matrix pseudo-inverse operation rule for online incremental updates, avoiding a lengthy iteration process and greatly improving modeling efficiency.

[0137] Unlike existing intelligent diagnostic methods for wind turbines that rely on independent SCADA data at a single point in time, this invention utilizes an integrated diagnostic strategy based on daily data to vote on diagnostic results, thus avoiding missed diagnoses and misdiagnoses caused by data fluctuations and improving diagnostic accuracy.

[0138] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0142] The above description is a detailed description of the preferred embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modifications made under the technical spirit of the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for intelligent fault diagnosis of wind turbine units based on a structure-adaptive width learning network, characterized in that, Includes the following steps: S1. Collect historical fault data of wind turbine units, clean the abnormal data in the historical fault data to obtain dataset X, and label the cleaned dataset X according to different fault states to obtain fault label Y. S2, using the cleaned dataset X and fault labels Y as training samples, train the initial width learning network. To obtain the initial fault diagnosis model; S3, perform T structural expansions on the initial width learning network. During each expansion, add [a certain number of elements] to the original width learning network. Each reinforcement node performs structural expansion, updating the original width learning network to obtain the updated width learning network. After T structural expansions, a structurally adaptively updated width learning network is obtained. To obtain the final fault diagnosis model; S4 inputs the data sample of the wind turbine to be diagnosed into the final fault diagnosis model to obtain the diagnosis result.

2. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 1, characterized in that, Step S2 includes the following steps: S21, Construct a series of mapping nodes to perform linear feature extraction and merging on the data in dataset X to obtain the mapping features. ; S22, Construct a series of enhanced node pairs mapping features Further nonlinear feature extraction and merging are performed to obtain enhanced features. ; S23, mapping features and enhanced features Merge the features to obtain the feature set. ; S24, Construct the loss function and use ridge regression theory to train the initial width learning network. Solve the problem to obtain an initial fault diagnosis model.

3. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 2, characterized in that, The loss function constructed in step S24 is as follows: ； Where λ represents the regularization parameter to prevent overfitting.

4. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 2, characterized in that, In step S24, ridge regression theory is used to train the initial width learning network. The solution is as follows: ； Where I represents the identity matrix; the superscript T represents the matrix transpose; and the superscript –1 represents the matrix inverse.

5. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 1, characterized in that, In step S3, the total number of expansions T and the width are determined using a structural adaptive expansion strategy, and the number of reinforcement nodes that the learning network needs to add during the t-th structural expansion are learned. , t=1,2,…,T.

6. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 5, characterized in that, Using a structural adaptive expansion strategy, the total number of expansions T and the network width are determined, and the number of reinforcement nodes required for the t-th structural expansion is learned. The steps include: S31, let there be K opportunities to determine the number of newly added reinforced nodes during the t-th structural expansion, and let the number of newly added reinforced nodes corresponding to the K opportunities be denoted as . The number of new reinforcement nodes is determined by a multiplicative increase rule, and the training accuracy of the model after adding l reinforcement nodes in the t-th structural expansion is denoted as . The accuracy of the previous training session is denoted as ; S32, for the first opportunity, add... A reinforced node, if Then let , This indicates that at the t-th incremental structural learning iteration, there is A reinforcement node is added to the original width learning network to obtain the updated width learning network. Then, the updated width learning network enters the (t+1)th structural expansion; if This opens the second chance. Add a reinforcement node to the original model, if Then set , This indicates that during the t-th incremental structural learning iteration, l2 reinforcement nodes are added to the original width learning network, resulting in an updated width learning network. Then, the updated width learning network enters the (t+1)-th structural expansion iteration. If it succeeds, then the next opportunity will be opened; and so on, for... If the k-th opportunity occurs for the first time When, then set , Based on the above rules, the number of reinforcing nodes added during the t-th structural expansion can be obtained: ； S33, for the t-th structural expansion, if all The adaptive structural expansion process is terminated, and the width learning network obtained from the (t-1)th structural expansion is selected as the final fault diagnosis model.

7. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 1, characterized in that, The steps for updating a width-learning network include: Feature extraction is performed using the newly added l enhancement nodes to obtain the extracted nonlinear features. ; Will With feature set Perform a merge update to obtain the updated feature set. ; The width learning network is updated using the block matrix pseudo-inverse algorithm to obtain the updated width learning network. : ； in, .

8. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 7, characterized in that, Will With feature set The merge and update should be performed according to the following formula: ； in, This is the updated feature set.

9. The intelligent fault diagnosis method for wind turbines based on a structure-adaptive width learning network as described in claim 1, characterized in that, Step S4 includes the following steps: S41, collect a full day's worth of SCADA data from the wind turbine to be diagnosed, and perform cleaning and pre-processing. S42, input the data from each sampling moment in the SCADA data processed in step S41 into the final fault diagnosis model to obtain the diagnosis result at the current sampling moment; S43 votes on the diagnostic results from all sampling times throughout the day, and outputs the integrated diagnostic results of wind turbine faults based on the principle of majority rule.

10. A wind turbine fault intelligent diagnosis system based on a structure-adaptive width learning network, characterized in that, include: The data acquisition module is configured to collect historical fault data of wind turbine units; The initial fault diagnosis model building module is configured to build an initial fault diagnosis model based on historical fault data collected by the acquisition module, and to train the initial fault diagnosis model. The structural adaptive expansion module is configured to update the established initial diagnostic model online using a structural adaptive expansion strategy. The adaptive termination module is configured to terminate the structural adaptive expansion module according to the set online update termination conditions and output the final fault diagnosis model. The integrated diagnostic module is configured to collect data samples from the wind turbine to be diagnosed and input them into the final fault diagnosis model to obtain diagnostic results.

Images