Data-driven methods, systems, equipment and media for wind turbine fault diagnosis and prediction

Abstract

This invention discloses a data-driven method, system, device, and medium for wind turbine fault diagnosis and prediction, comprising: acquiring a baseline sequence and a test sequence; calculating a fault measurement sequence of the test sequence based on the baseline sequence; performing two clustering operations on the fault measurement sequence of the test sequence and obtaining the fault type of the wind turbine based on the clustering results; calculating a statistical feature value sequence of the fault measurement sequence; and inputting the statistical feature value sequence into a pre-trained TCN network to obtain the operating status of the wind turbine in the next time period. The method, system, device, and medium have the characteristics of high accuracy and good real-time performance.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine fault diagnosis and prediction, and relates to a data-driven wind turbine fault diagnosis and prediction method, system, equipment and medium. Background Technology

[0002] Existing wind turbine fault diagnosis and prediction methods can be broadly categorized into model-based analysis methods and data-driven analysis methods. Model-based analysis methods rely heavily on accurate modeling of the wind turbine system. However, as the scale and complexity of wind turbine systems increase, the difficulty of modeling rises sharply, and the effectiveness of model diagnosis and prediction becomes increasingly limited. Data-driven analysis methods, on the other hand, focus on fault feature extraction. For the same set of data, different fault feature extraction algorithms can yield vastly different conclusions. Currently, the large-scale application of SCADA systems in wind turbine generators has made real-time monitoring of various state variables in large wind farms possible. Massive amounts of data can be used to determine overall or partial faults in wind turbines without a detailed mechanistic model of the turbine itself.

[0003] Existing data-driven diagnostic and predictive methods all have to make trade-offs between accuracy and real-time performance. Algorithms with excellent accuracy often have high time complexity, while algorithms with good real-time performance often struggle to achieve satisfactory accuracy. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a data-driven method, system, device and medium for wind turbine fault diagnosis and prediction. This method, system, device and medium have the characteristics of high accuracy and good real-time performance.

[0005] To achieve the above objectives, this invention discloses a data-driven method for wind turbine fault diagnosis and prediction, comprising:

[0006] Obtain the baseline sequence and the sequence to be tested;

[0007] The fault measurement sequence of the test sequence is calculated based on the benchmark sequence;

[0008] The fault measurement sequence of the test sequence is clustered twice, and the fault type of the wind turbine is obtained based on the clustering results;

[0009] Calculate the statistical characteristic value sequence of the fault measurement sequence;

[0010] The statistical feature value sequence is input into the pre-trained TCN network to obtain the operating status of the wind turbine in the next time period.

[0011] The specific process for calculating the fault measurement sequence of the test sequence based on the benchmark sequence is as follows:

[0012] Calculate the Pearson correlation coefficient Pcc between each pair of time slice subsequences of length m on the test sequence and the reference sequence. Select the time slice subsequence in the test sequence that corresponds to the maximum Pearson correlation coefficient Pcc, and use this time slice subsequence as the fault measurement sequence of the test sequence.

[0013] For sequence M R =[r1,r2,…,r m ] and sequence M x =[x1,x2,…,x m The Pearson correlation coefficient Pcc(M) between the two is... R M X m) is:

[0014]

[0015] When the length of the sequence to be tested is l, the length of the fault measurement sequence is l-m+1.

[0016] The fault measurement sequence of the test sequence is clustered twice using the unsupervised K-Means method.

[0017] A sliding window is used to sample the fault metric sequence FM to obtain a sequence of statistical feature values.

[0018] This invention discloses a data-driven wind turbine fault diagnosis and prediction system, comprising:

[0019] The acquisition module is used to acquire the baseline sequence and the sequence to be tested;

[0020] The first calculation module is used to calculate the fault measurement sequence of the sequence under test based on the benchmark sequence;

[0021] The diagnostic module is used to perform two clusters on the fault measurement sequence of the test sequence and obtain the fault type of the wind turbine based on the clustering results.

[0022] The second calculation module is used to calculate the statistical feature value sequence of the fault measurement sequence;

[0023] The prediction module is used to input the statistical feature value sequence into the pre-trained TCN network to obtain the operating status of the wind turbine in the next time period.

[0024] The present invention discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the data-driven wind turbine fault diagnosis and prediction method.

[0025] The present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the data-driven wind turbine fault diagnosis and prediction method.

[0026] The present invention has the following beneficial effects:

[0027] In practical operation, the data-driven wind turbine fault diagnosis and prediction method, system, equipment, and medium described in this invention determine the fault type of the wind turbine based on a fault measurement sequence and employing a two-stage clustering approach. Furthermore, the statistical feature value sequence of the fault measurement sequence is input into a pre-trained TCN network to predict the wind turbine operating status for the next time period, thereby reducing computational load and improving real-time performance. Moreover, this invention utilizes neural networks and clustering methods for fault diagnosis and prediction, resulting in high computational accuracy. Attached Figure Description

[0028] Figure 1 This is a schematic diagram for fault feature extraction.

[0029] Figure 2 This is a schematic diagram for fault diagnosis;

[0030] Figure 3 This is a schematic diagram for fault prediction.

[0031] Figure 4 This is a schematic diagram for continuous prediction.

[0032] Figure 5 A diagram showing the operational status of the wind turbine;

[0033] Figure 6 A diagram showing the classification of wind turbine fault types;

[0034] Figure 7 Verification diagram for Fault_0 fault type;

[0035] Figure 8 Verification diagram for Fault_1 fault type;

[0036] Figure 9 Verification diagram for Fault_2 fault type;

[0037] Figure 10 A schematic diagram of the training dataset for the TCN network;

[0038] Figure 11 This is a verification diagram for fault prediction. Detailed Implementation

[0039] To enable those skilled in the art to better understand the present invention, 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 merely some embodiments of the present invention, not all embodiments, and are not intended to limit the scope of the present invention. Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion regarding the concepts disclosed in the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0040] The accompanying drawings show structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not drawn 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.

[0041] Example 1

[0042] The data-driven wind turbine fault diagnosis and prediction method of the present invention includes the following steps:

[0043] 1) Fan fault diagnosis;

[0044] 11) Fault Feature Extraction Based on Time Slices

[0045] The SCADA system selects a segment of time-series sample data representing normal operating conditions as the benchmark sequence Benckmark. The Pearson correlation coefficient Pcc is calculated between each pair of time slice subsequences of length m and the benchmark sequence Benckmark. The time slice subsequence in the test sequence corresponding to the maximum Pcc is selected and used as the fault measurement sequence (FM) of the test sequence Test Data. When the length of the test sequence Test Data is l, the length of the fault measurement sequence is l-m+1.

[0046] The calculation process for the Pearson correlation coefficient is as follows:

[0047] For sequence M R =[r1,r2,…,r m ] and sequence M x = [x1,x2,…,x m The Pearson correlation coefficient Pcc(M) between the two is... RM X m) is:

[0048]

[0049] From the above equation, it can be seen that the value of Pearson correlation Pcc always falls within the interval [0,1], and when M R With M x The closer the distance is, the closer the Pearson correlation Pcc value is to 1; conversely, when M... R With M x The greater the distance, the closer the Pearson correlation Pcc value is to 0.

[0050] The fault metric sequence FM is then:

[0051] FM=[f_1,f_2,\dots,f_{n-m+1}]

[0052] in

[0053] f j =max{Pcc(M R M x ,m),i=1,2,…,n-m+1}

[0054] The Fault Measurement Sequence (FM) contains the fault characteristics of each time slice in the sequence under test, and its calculation is illustrated as follows: Figure 1 As shown;

[0055] 12) Fault diagnosis and prediction;

[0056] 121) Fault diagnosis;

[0057] like Figure 2 As shown, after obtaining the fault metric sequence FM of the test sequence, a clustering algorithm (taking unsupervised K-Means as an example) is introduced to sort the disordered fault metric sequence by class. Based on the sorting result, a fault diagnosis result is given to represent the working and operating status of the wind turbine equipment. According to the characteristics of the Pearson correlation Pcc, the closer the Pearson correlation is to 0, the worse the sequence similarity. In this invention, this represents the greater the difference from normal operating conditions, i.e., the closer to fault conditions. Figure 2 Based on Class_0, the fault categories are clustered again to obtain the classification of fault types.

[0058] 122) Fault Prediction

[0059] It should be noted that fault prediction must first consider the timeliness of the prediction, as the accuracy of the prediction algorithm decreases as the prediction time increases. Therefore, in high-frequency sampled streaming data, prediction of sampling points is not very meaningful.

[0060] Based on the aforementioned fault feature extraction and fault diagnosis, this invention employs a sliding window to sample the fault metric sequence (FM) to obtain a statistical feature value sequence. Here, the statistical feature values ​​(Mean-cluster Centers) represent the curve of the fault metric sequence FM, thereby extending the prediction time. For the statistical feature value sequence, it is input into the trained TCN network to obtain the statistical feature value corresponding to the next sliding window. Through iterative prediction, continuous prediction of future statistical feature values ​​is achieved. Figure 4 As shown, this serves as the basis for fault prediction.

[0061] Example 2

[0062] First, based on the equipment characteristics, a stator A-phase output current waveform of the wind turbine generator with an amplitude of the equipment's rated value and a frequency of 50Hz (power frequency) is selected as the reference sequence. For the input test sequence, after obtaining the corresponding fault measurement sequence using this invention, unsupervised K-Means clustering is used to divide it. When the number of K-Means clusters K = 4, the operating state division of the motor equipment is obtained, such as... Figure 5 As shown, where, Figure 5 In the diagram, the horizontal axis represents the magnitude of the fault metric, and the vertical axis represents the density distribution of the corresponding category. The higher the density, the more points there are at that location, and vice versa. Furthermore, unsupervised K-Means clustering is applied again to the fault category Class_0, with the number of clusters K=3, thus obtaining the fault type classification.

[0063] for Figure 6 For each fault type in the list, verify it. For example... Figure 7 As shown, Fault_0 indicates significant high-order harmonic interference in the current waveform, or a clear three-term asymmetry. This type of fault often originates from bearing eccentricity or abnormal vibration in the wind turbine equipment, and is a serious fault. Upon discovery, the machine should be stopped immediately for inspection and repair to avoid greater losses and damage.

[0064] like Figure 8 As shown, the current waveform in the Fault_1 type exhibits obvious low-frequency oscillations. After comparison, it was found that the frequency of this current is smaller than that of the reference current. This type of fault often originates from the mass imbalance of the wind turbine rotor blades and the failure of the motor equipment, which is often a precursor to major faults.

[0065] like Figure 9 As shown, the current waveform in Fault_2 type exhibits relatively obvious fixed-frequency oscillation, and its current frequency is also slightly lower than the baseline data. If left unchecked, it will soon evolve into Fault_1 type, and maintenance personnel should pay attention to this.

[0066] Based on fault diagnosis, a temporal convolutional neural network (TCN) is introduced for fault prediction. This neural network requires training with historical data before use. This invention selects multiple evolutionary data points from normal to fault conditions from the generator's stator A-phase current data, and uses the aforementioned feature extraction algorithm to obtain a fault measurement sequence. The statistical features of the fault measurement sequence are then calculated, and this data forms the training dataset for the TCN network. Figure 10 As shown.

[0067] For a trained TCN network, a set of data that gradually fails is selected for validation, such as... Figure 11 As shown.

[0068] from Figure 11 As can be seen, the vertical axis represents the statistical distribution of changes over time. The actual data clearly shows a trend of evolution from normal operating conditions to fault conditions. Inputting the data to the right of the dashed line into the TCN network for prediction yields the predicted point line. Clearly, the trend of the predicted results closely matches the actual trend, both demonstrating a rapid tendency for the equipment to fail, presenting satisfactory prediction results.

[0069] Example 3

[0070] The data-driven wind turbine fault diagnosis and prediction system of the present invention includes:

[0071] The acquisition module is used to acquire the baseline sequence and the sequence to be tested;

[0072] The first calculation module is used to calculate the fault measurement sequence of the sequence under test based on the benchmark sequence;

[0073] The diagnostic module is used to perform two clusters on the fault measurement sequence of the test sequence and obtain the fault type of the wind turbine based on the clustering results.

[0074] The second calculation module is used to calculate the statistical feature value sequence of the fault measurement sequence;

[0075] The prediction module is used to input the statistical feature value sequence into the pre-trained TCN network to obtain the operating status of the wind turbine in the next time period.

[0076] Example 4

[0077] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of a data-driven wind turbine fault diagnosis and prediction method. The memory may include main memory, such as high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device. The processor, network interface, and memory are interconnected via an internal bus, which may be an industry-standard architecture bus, a peripheral component interconnection standard bus, an extended industry-standard architecture bus, etc. The bus may be categorized as an address bus, a data bus, a control bus, etc. The memory stores the program; specifically, the program may include program code, which includes computer operation instructions. The memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0078] Example 5

[0079] A computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the data-driven wind turbine fault diagnosis and prediction method. Specifically, the computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0080] 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.

[0081] 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 1A device that provides the functions specified in one or more boxes.

[0082] 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.

[0083] 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.

[0084] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A data-driven method for wind turbine fault diagnosis and prediction, characterized in that, include: Obtain the baseline sequence and the sequence to be tested. Among them, a segment of data representing normal operating conditions is selected from the time series sample data provided by the SCADA system as the baseline sequence. The fault measurement sequence of the test sequence is calculated based on the benchmark sequence; The fault measurement sequence of the test sequence is clustered twice, and the fault type of the wind turbine is obtained based on the clustering results; Calculate the statistical characteristic value sequence of the fault measurement sequence; The statistical feature value sequence is input into the pre-trained TCN network to obtain the operating status of the wind turbine in the next time period; The specific process for calculating the fault measurement sequence of the test sequence based on the benchmark sequence is as follows: Calculate the Pearson correlation coefficient Pcc between each pair of time slice subsequences of length m on the test sequence and the reference sequence. Select the time slice subsequence in the test sequence that corresponds to the maximum Pearson correlation coefficient Pcc, and use this time slice subsequence as the fault measurement sequence of the test sequence.

2. The data-driven wind turbine fault diagnosis and prediction method according to claim 1, characterized in that, For sequence M R =[r1,r2,…,r m ] and sequence M x =[x1,x2,…,x m The Pearson correlation coefficient Pcc(M) between the two is... R M X m) is:

3. The data-driven wind turbine fault diagnosis and prediction method according to claim 1, characterized in that, When the length of the sequence to be tested is l, the length of the fault measurement sequence is l-m+1.

4. The data-driven wind turbine fault diagnosis and prediction method according to claim 1, characterized in that, The fault measurement sequence of the test sequence is clustered twice using the unsupervised K-Means method.

5. The data-driven wind turbine fault diagnosis and prediction method according to claim 1, characterized in that, A sliding window is used to sample the fault metric sequence FM to obtain a sequence of statistical feature values.

6. A data-driven wind turbine fault diagnosis and prediction system, characterized in that, include: The acquisition module is used to acquire the baseline sequence and the sequence to be tested. Among them, the time series sample data provided by the SCADA system is used to select a segment of data that represents normal operating conditions as the baseline sequence. The first calculation module is used to calculate the fault measurement sequence of the sequence under test based on the benchmark sequence; The diagnostic module is used to perform two clusters on the fault measurement sequence of the test sequence and obtain the fault type of the wind turbine based on the clustering results. The second calculation module is used to calculate the statistical feature value sequence of the fault measurement sequence; The prediction module is used to input the statistical feature value sequence into the pre-trained TCN network to obtain the operating status of the wind turbine in the next time period; The specific process for calculating the fault measurement sequence of the test sequence based on the benchmark sequence is as follows: Calculate the Pearson correlation coefficient Pcc between each pair of time slice subsequences of length m on the test sequence and the reference sequence. Select the time slice subsequence in the test sequence that corresponds to the maximum Pearson correlation coefficient Pcc, and use this time slice subsequence as the fault measurement sequence of the test sequence.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the data-driven wind turbine fault diagnosis and prediction method as described in any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the data-driven wind turbine fault diagnosis and prediction method as described in any one of claims 1-5.

Images