Blade damage identification method, device and equipment based on unsupervised learning and medium

Abstract

The invention provides a blade damage identification method and device based on unsupervised learning, equipment and a medium, and relates to the technical field of wind power generation. Based on the to-be-detected multi-source data, the identification model is adopted to predict the state of the wind power blade; wherein the recognition model is used for determining a training positive sample data set based on historical detection multi-source data; and based on the training positive sample data set, the initial identification model is trained, so that the accuracy of blade failure identification is improved, and personal errors are reduced at the same time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power generation, in particular to a blade damage identification method and device based on unsupervised learning, equipment and medium. BACKGROUND

[0002] At present, blade failure identification is mainly performed by analyzing images and sensor data through artificial intelligence algorithms. These algorithms are constructed or optimized by using supervised learning based on positive and negative sample data sets. However, due to the lack of failure data in the wind power industry, there are more normal data in the positive samples, and there are less blade failure data in the negative samples, which makes the supervised learning prone to overfitting, resulting in inaccurate identification. SUMMARY

[0003] Therefore, the purpose of the present application is to provide a blade damage identification method and device based on unsupervised learning to improve the accuracy of identifying blade failure.

[0004] In a first aspect, the present application provides a blade damage identification method based on unsupervised learning, comprising: obtaining the to-be-detected multi-source data of the wind turbine blade; predicting the state of the wind turbine blade based on the to-be-detected multi-source data using an identification model; wherein the identification model is determined based on historical detection multi-source data to determine a training positive sample data set; and the initial identification model is trained based on the training positive sample data set to obtain.

[0005] Optionally, after obtaining the to-be-detected multi-source data of the wind turbine blade, the method further comprises: performing outlier rejection on the to-be-detected multi-source data to obtain first to-be-detected multi-source data; performing missing value processing on the first to-be-detected multi-source data to obtain second to-be-detected multi-source data; performing standardization conversion on the second to-be-detected multi-source data to obtain standardized to-be-detected multi-source data.

[0006] Optionally, predicting the state of the wind turbine blade based on the to-be-detected multi-source data using an identification model comprises: determining the first Euclidean distance between the to-be-detected multi-source data and the training positive sample data in the training positive sample data set based on the to-be-detected multi-source data; determining the k smallest first Euclidean distances of the to-be-detected multi-source data based on the first Euclidean distances; determining the first average Euclidean distance value of the training positive sample data set based on each k smallest first Euclidean distance; determining the state of the wind turbine blade based on the first average Euclidean distance value and the threshold value in the identification model.

[0007] Optionally, the state of the wind turbine blade is determined based on the first average Euclidean distance value and a threshold value in the identification model, including: If the first average Euclidean distance value is not greater than the threshold value, it is determined that the state of the wind turbine blade is a normal state; wherein the normal state is that the wind turbine blade has no failure risk; If the first average Euclidean distance value is greater than the threshold value, it is determined that the state of the wind turbine blade is a failure or damage state; wherein the failure or damage state is that the wind turbine blade has a crack and wear failure risk.

[0008] Optionally, the wind turbine blade damage identification method based on unsupervised learning provided by the present application further comprises: acquiring historical detection multi-source data; standardizing the historical detection multi-source data to obtain standardized historical detection multi-source data; and taking the normal state data in the standardized historical detection multi-source data as a training positive sample data set.

[0009] Optionally, the initial identification model is trained based on the training positive sample data set, including: determining, based on the training positive sample data set, each second Euclidean distance between each training positive sample data in the training positive sample data set and other training positive sample data in the training positive sample data set; determining, based on each second Euclidean distance, m minimum second Euclidean distances corresponding to each training positive sample data; determining, based on each of the m minimum second Euclidean distances, a second average Euclidean distance value of the training positive sample data set; determining, based on the second average Euclidean distance value, a threshold value corresponding to the training positive sample data set, and obtaining an identification model based on the threshold value.

[0010] Optionally, the wind turbine blade damage identification method based on unsupervised learning provided by the present application further comprises: selecting a preset training positive sample data from the training positive sample data set as a validation sample data; optimizing the threshold value based on the validation sample data to obtain an optimized identification model.

[0011] In a second aspect, the present application provides a wind turbine blade damage identification device based on unsupervised learning, comprising: a data acquisition module for acquiring wind turbine blade detection multi-source data; a data processing module for predicting the state of the wind turbine blade based on the detection multi-source data using an identification model; wherein the identification model is obtained by determining a training positive sample data set based on historical detection multi-source data, and training an initial identification model based on the training positive sample data set.

[0012] Thirdly, this application provides an electronic device including 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 aforementioned blade damage identification method based on unsupervised learning.

[0013] Fourthly, this application provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, provide the aforementioned blade damage identification method based on unsupervised learning.

[0014] This invention provides a blade damage identification method, device, equipment, and medium based on unsupervised learning. It acquires multi-source data to be detected; based on the multi-source data to be detected, an identification model is used to predict the state of wind turbine blades; wherein, the identification model is determined based on historical multi-source data to determine a training positive sample dataset; and the initial identification model is trained based on the training positive sample dataset, so as to improve the accuracy of blade failure identification while reducing human error.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating the determination of the identification model provided in an embodiment of the present invention is shown; Figure 2 A flowchart of a blade damage identification method based on unsupervised learning provided by an embodiment of the present invention is shown; Figure 3 A schematic diagram of the structure of a blade damage identification method based on unsupervised learning provided in an embodiment of the present invention is shown. Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present invention is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] Currently, the identification of blade failures in the wind power industry mainly relies on manual inspection, instrument testing, drone inspection, and blade vibration monitoring. Among these, manual inspection is inefficient, highly susceptible to subjective factors, prone to missed detections due to human error, and poses safety hazards. For large-scale wind farms, the inspection cycle is long, making real-time monitoring difficult. Instrument testing methods, such as ultrasonic testing and infrared detection, are also effective. Ultrasonic testing can detect internal blade defects such as cracks, but it requires contact testing, which is inconvenient. Infrared detection determines damage by detecting the temperature distribution on the blade surface; however, its accuracy is easily affected by factors such as ambient temperature. Drone inspection utilizes drones or fixed cameras to inspect wind turbine blades. Image acquisition: Drones can capture images of blades from different angles, suitable for inspections of large wind farms; fixed cameras can be installed near wind turbines for real-time monitoring of blades; blade vibration monitoring involves installing vibration sensors at the blade root to collect blade vibration frequencies, identifying the blade's natural frequency, and analyzing changes in the natural frequency to identify blade failures. The machine learning algorithms used in these methods are all built or optimized using supervised learning methods based on positive and negative sample datasets. However, due to the scarcity of failure data in the wind power industry, with a large number of positive normal samples and a small number of negative blade failure samples, supervised learning methods are prone to overfitting, leading to inaccurate identification. Therefore, this application provides a blade damage identification method based on unsupervised learning. This method acquires multi-source data to be detected; based on this multi-source data, an identification model is used to predict the state of wind turbine blades; the identification model is determined by using historical multi-source data to determine a training positive sample dataset; and the initial identification model is trained using this training positive sample dataset to obtain the identification model, thereby improving the accuracy of blade failure identification while reducing human error.

[0020] This application provides a method for determining a recognition model, see below. Figure 1 As shown in the embodiments of this application, the general flow of the method for determining the recognition model is as follows: Step 110: Obtain historical multi-source detection data.

[0021] In this embodiment of the application, the multi-source data for detection includes physical characteristic data of wind turbine blades such as vibration signals, strain signals, temperature signals and / or acoustic signature signals. Among them, the physical characteristic data collected by various sensors installed on the wind turbine blades can be used as historical multi-source data for detection.

[0022] Step 120: Standardize the historical multi-source detection data to obtain standardized historical multi-source detection data; and use the normal state data in the standardized historical multi-source detection data as the training positive sample dataset.

[0023] To eliminate the impact of low recognition accuracy caused by dimensional interference, in this embodiment, historical detection multi-source data is standardized to obtain standardized historical detection multi-source data. Standardization unifies features of different dimensions to the same scale and simultaneously completes outlier processing and feature dimensionality reduction. The standardized historical detection multi-source data can be obtained in the following ways: outlier removal is performed on the historical detection multi-source data to obtain first historical detection multi-source data; missing value processing is performed on the first historical detection multi-source data to obtain second historical detection multi-source data; and standardization transformation is performed on the second historical detection multi-source data to obtain standardized historical detection multi-source data.

[0024] Specifically, 3 The criteria remove outliers such as peak values ​​of instantaneous false alarms from sensors or low values ​​of faults from historical multi-source detection data to obtain the first set of historical multi-source detection data. For example, the mean value is calculated for vibration data. with standard deviation Remove those exceeding [ , The values ​​are within a certain range to reduce noise interference with the baseline of the recognition model; For missing values ​​in the first historical multi-source data caused by sensor failure or transmission interruption, the nearest time interpolation method suitable for time series data or the mean imputation method of similar samples suitable for static feature data can be used to process the missing values ​​and obtain the second historical multi-source data. The second set of historical multi-source detection data is subjected to Z-score standardization (standard deviation standardization) for unified scaling. The transformed historical multi-source detection data has a mean of 0 and a standard deviation of 1, thus completely eliminating dimensional differences and ensuring that the features of the multi-source detection data have equal weights in the distance calculation of the recognition model. The Z-score standardization formula is as follows:

[0025] In the formula, For standardized historical multi-source detection data, For the second historical detection of multi-source data, The mean, denoted as , where both the mean and standard deviation can be determined based on all historical multi-source data.

[0026] Since the recognition model provided in this application embodiment is based on the positive sample dataset (normal state dataset) to train the model and construct the feature distribution baseline of normal data (i.e., normal samples should be clustered in a specific region of the feature space, while failed samples will be shifted to that region due to feature abnormalities), the normal state data in the standardized historical detection multi-source data is used as the training positive sample dataset.

[0027] Step 130: Based on the training positive sample dataset, determine the second Euclidean distances between each training positive sample data and other training positive sample data in the training positive sample dataset; based on each second Euclidean distance, determine the m minimum Euclidean distances corresponding to each training positive sample data; based on each m minimum second Euclidean distances, determine the second average Euclidean distance value of the training positive sample dataset; based on the second average Euclidean distance value, determine the threshold corresponding to the training positive sample dataset, and based on the threshold, obtain the recognition model.

[0028] In this embodiment of the application, the specific process of training the initial recognition model based on the training positive sample dataset is as follows: First, for each training positive sample in the training positive sample dataset, calculate its second Euclidean distance with all other training positive samples in the same dataset. The second Euclidean distance can be calculated using the following formula:

[0029] In the formula, Let n be the Euclidean distance between two training positive sample data, and n be the feature dimension of the training positive sample data. Secondly, for each training positive sample data, select the m smallest second Euclidean distances, and calculate the second average Euclidean distance value corresponding to the m smallest second Euclidean distances, that is, the K nearest neighbor average distance; Then, the second average Euclidean distance value corresponding to all training sample data is statistically analyzed, and 1.5 to 2 times the value is taken as the threshold T (i.e. the feature distribution baseline of normal data) to obtain the recognition model.

[0030] It should be noted that the threshold T is the quantization boundary and execution standard of the feature distribution baseline of normal data in the identification model, reflecting the clustering density and range of multi-source data when the wind turbine blade is in a normal state; and the identification model is a model trained by using massive normal state data through KNN unsupervised reconstruction, and a model that constructs the feature distribution baseline of normal data.

[0031] Step 140: Select preset training positive sample data from the training positive sample dataset as validation sample data; based on the validation sample data, optimize the threshold to obtain the optimized recognition model.

[0032] In order to determine the recognition accuracy of the recognition model, in this embodiment of the application, a preset training positive sample data is selected from the training positive sample dataset as the verification sample data; based on the verification sample data, the threshold is optimized to obtain the optimized recognition model.

[0033] Specifically, 40% of the training positive sample data in the training positive sample dataset is selected as the validation sample data. Based on the validation sample data, the validation sample data is divided into n subsets. The threshold in the recognition model is verified using n-1 subsets, and the remaining subset is used as the boundary sample detection set. If the validation sample coverage of a certain subset is less than 99%, the threshold is fine-tuned according to the rule that for every 1% decrease in coverage, the threshold is increased by 0.1 times (for example, from 1.5 times the mean to 1.6 times), until the coverage of all cross-validation subsets is ≥99%. Finally, the mean of the optimal thresholds of all cross-validation rounds is taken as the optimized threshold T, and the optimized recognition model is obtained.

[0034] The identification model provided in this application reduces the interference of sensor noise on the identification model through outlier removal, missing value processing, and feature standardization in the preprocessing stage; the threshold T is cross-validated using training positive samples, which can adapt to the normal data distribution differences of different wind fields and different blade models; by constructing a baseline with only a massive number of positive samples, without relying on scarce negative samples for unsupervised mode and without the need for manual labeling of negative samples, human error is reduced and the stability of the algorithm in real-world scenarios is improved.

[0035] This application provides a blade damage identification method based on unsupervised learning, see below. Figure 2 As shown in the embodiments of this application, the general flow of the blade damage identification method based on unsupervised learning is as follows: Step 210: Obtain multi-source data to be tested for wind turbine blades.

[0036] In this embodiment of the application, physical feature data collected by various sensors installed on the wind turbine blades can be used as multi-source data to be detected. The multi-source data to be detected can be a feature vector of a single real-time data or a time-series data.

[0037] In this embodiment of the application, after acquiring the multi-source data to be detected for wind turbine blades, the multi-source data to be detected is preprocessed. The specific preprocessing process is as follows: outlier removal is performed on the multi-source data to be detected to obtain the first multi-source data to be detected; missing value processing is performed on the first multi-source data to be detected to obtain the second multi-source data to be detected; and standardization transformation is performed on the second multi-source data to be detected to obtain standardized multi-source data to be detected.

[0038] Specifically, 3 The criterion is to remove extreme outliers such as peak values ​​of instantaneous false alarms from sensors or low values ​​of faults from the multi-source data to be detected, thus obtaining the first set of multi-source data to be detected. For example, the mean value is calculated for vibration data. with standard deviation Remove those exceeding [ , The range of values; For missing values ​​in the first multi-source data to be detected due to sensor failure or transmission interruption, the nearest time interpolation method suitable for time series data or the mean of similar samples filling method suitable for static feature data can be used to process the missing values ​​and obtain the second multi-source data to be detected. The second set of multi-source data to be tested is subjected to Z-score standardization (standard deviation standardization) for uniform scaling. The mean of the transformed multi-source data is 0 and the standard deviation is 1, thus completely eliminating the difference in dimensions. The Z-score standardization formula is as follows:

[0039] In the formula, For the standardized multi-source data to be tested, For the second set of multi-source data to be detected, The mean, denoted as , where both the mean and standard deviation can be determined based on all multi-source data to be tested.

[0040] Step 220: Based on the multi-source data to be detected, use an identification model to predict the state of the wind turbine blades; wherein, the identification model is obtained by determining the training positive sample dataset based on historical multi-source data; and by training the initial identification model based on the training positive sample dataset.

[0041] In this embodiment, the state of a wind turbine blade is predicted using an identification model based on the multi-source data to be detected. This includes: determining each first Euclidean distance between the multi-source data to be detected and the training positive sample data in the training positive sample dataset; determining k minimum Euclidean distances between the multi-source data to be detected based on each first Euclidean distance; determining a first average Euclidean distance value for the training positive sample dataset based on each of the k minimum Euclidean distances; comparing the first average Euclidean distance value with a threshold in the identification model; if the first average Euclidean distance value is not greater than the threshold, the wind turbine blade is determined to be in a normal state; wherein, a normal state means that the wind turbine blade has no risk of failure; if the first average Euclidean distance value is greater than the threshold, the wind turbine blade is determined to be in a failed or damaged state; wherein, a failed or damaged state means that the wind turbine blade has a risk of cracking and wear failure.

[0042] Specifically, the process of predicting the state of wind turbine blades using a recognition model based on the multi-source data to be detected is as follows: First, for each data point in the multi-source data to be detected, calculate its first Euclidean distance to all training positive sample data in the training positive sample dataset. The first Euclidean distance can be calculated using the following formula:

[0043] In the formula, is the Euclidean distance between the data in the multi-source data to be detected and the training sample data in the training positive sample set, and n is the feature dimension of the multi-source data to be detected and the training positive sample data; Secondly, for each set of data from the multi-source dataset to be detected, select the k smallest first Euclidean distances and calculate the first average Euclidean distance value d-avg corresponding to the k smallest first Euclidean distances; Then, the first average Euclidean distance value d-avg is compared with the threshold T in the recognition model. If the first average Euclidean distance value d-avg ≤ threshold T, the wind turbine blade is determined to be in a normal state, indicating that the characteristics of the sample to be tested are consistent with the distribution of normal samples, and the blade has no risk of failure. If the first average Euclidean distance value d-avg > threshold T, the wind turbine blade is determined to be in a failed or damaged state, indicating that the characteristics of the sample to be tested deviate from the normal baseline, and the blade may have failure problems such as cracks or wear.

[0044] Furthermore, based on the first average Euclidean distance value d-avg and the threshold T in the recognition model, the ratio of the first average Euclidean distance value d-avg to the threshold T in the recognition model is determined to establish a confidence level. This confidence level reflects the relative degree to which the multi-source data to be detected deviates from the normal boundary. For example, if the ratio is ≤1.2, the multi-source data to be detected is within the normal range; if the ratio is >1.2, the multi-source data to be detected has entered the abnormal range. The larger the ratio, the higher the confidence level of the abnormal data. This provides maintenance personnel with decision-making references for tracking low-confidence anomalies or repairing high-confidence anomalies.

[0045] This application provides a blade damage identification device based on unsupervised learning, see below. Figure 3 As shown, the blade damage identification device based on unsupervised learning provided in this application includes: The data acquisition module 310 is used to acquire multi-source data to be detected from the wind turbine blades; The data processing module 320 is used to predict the state of wind turbine blades based on the multi-source data to be detected using an identification model. The identification model is obtained by determining a training positive sample dataset based on historical multi-source data and training the initial identification model based on the training positive sample dataset.

[0046] In an optional embodiment, the data processing module 320 is used to: Outlier removal is performed on the multi-source data to be detected to obtain the first multi-source data to be detected; missing value processing is performed on the first multi-source data to be detected to obtain the second multi-source data to be detected; the second multi-source data to be detected is standardized to obtain the standardized multi-source data to be detected.

[0047] In an optional embodiment, the data processing module 320 is used to: Based on the multi-source data to be detected, determine each first Euclidean distance between the multi-source data to be detected and the training positive sample data in the training positive sample dataset; based on each first Euclidean distance, determine k minimum first Euclidean distances of the multi-source data to be detected; based on each of the k minimum first Euclidean distances, determine the first average Euclidean distance value of the training positive sample dataset; based on the first average Euclidean distance value and the threshold in the recognition model, determine the state of the wind turbine blade.

[0048] In an optional embodiment, the data processing module 320 is used to: If the first average Euclidean distance value is not greater than the threshold, the wind turbine blade is determined to be in a normal state; where the normal state means that the wind turbine blade has no risk of failure. If the first average Euclidean distance value is greater than the threshold, the wind turbine blade is determined to be in a failed or damaged state; where the failed or damaged state means that the wind turbine blade has the risk of failure due to cracks and wear.

[0049] In an optional embodiment, the unsupervised learning-based blade damage identification device further includes a model training module 330, used for: Acquire historical multi-source detection data; standardize the historical multi-source detection data to obtain standardized historical multi-source detection data; and use the normal state data in the standardized historical multi-source detection data as the training positive sample dataset.

[0050] In an optional embodiment, the model training module 330 is used for: Based on the training positive sample dataset, determine the second Euclidean distances between each training positive sample data and other training positive sample data in the training positive sample dataset; based on each second Euclidean distance, determine the m minimum second Euclidean distances corresponding to each training positive sample data; based on each of the m minimum second Euclidean distances, determine the second average Euclidean distance value of the training positive sample dataset; based on the second average Euclidean distance value, determine the threshold corresponding to the training positive sample dataset, and based on the threshold, obtain the recognition model.

[0051] In an optional embodiment, the model training module 330 is used for: Select a preset training positive sample data as validation sample data from the training positive sample dataset; optimize the threshold based on the validation sample data to obtain the optimized recognition model.

[0052] It should be noted that the principle of the blade damage identification device based on unsupervised learning provided in this application embodiment to solve the technical problem is similar to the blade damage identification method based on unsupervised learning provided in this application embodiment. Therefore, the implementation of the blade damage identification device based on unsupervised learning provided in this application embodiment can refer to the implementation of the blade damage identification method based on unsupervised learning provided in this application embodiment, and the repeated parts will not be described again.

[0053] After introducing the unsupervised learning-based blade damage identification method and apparatus provided in the embodiments of this application, the electronic equipment provided in the embodiments of this application will be briefly introduced next.

[0054] See Figure 4 As shown, the electronic device 500 provided in this application embodiment includes at least a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the processor 501 executes the computer program, it implements the blade damage identification method based on unsupervised learning provided in this application embodiment.

[0055] The electronic device 500 provided in this application embodiment may further include a bus 503 connecting different components (including processor 501 and memory 502). The bus 503 represents one or more types of bus structures, including memory bus, peripheral bus, local area bus, etc.

[0056] Memory 502 may include a readable storage medium in the form of volatile memory, such as random access memory (RAM) 5021 and / or cache memory 5022, and may further include read-only memory (ROM) 5023. Memory 502 may also include a program tool 5025 having a set (at least one) of program modules 5024, including but not limited to an operating subsystem, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0057] Processor 501 can be a single processing element or a collective term for multiple processing elements. For example, processor 501 can be a central processing unit (CPU) or one or more integrated circuits configured to implement the unsupervised learning-based blade damage identification method provided in the embodiments of this application. Specifically, processor 501 can be a general-purpose processor, including but not limited to CPUs, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0058] Electronic device 500 can communicate with one or more external devices 504 (e.g., keyboard, remote control, etc.), and also with one or more devices that enable a user to interact with electronic device 500 (e.g., mobile phone, computer, etc.), and / or with devices that enable electronic device 500 to communicate with one or more other electronic devices 500 (e.g., router, modem, etc.). This communication can be performed through input / output (I / O) interface 505. Furthermore, electronic device 500 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 506. Figure 4As shown, network adapter 506 communicates with other modules of electronic device 500 via bus 503. It should be understood that, although... Figure 4 As not shown, other hardware and / or software modules may be used in conjunction with the electronic device 500, including but not limited to microcode, device drivers, redundant processors, external disk drive arrays, Redundant Arrays of Independent Disks (RAID) subsystems, tape drives, and data backup storage subsystems.

[0059] It should be noted that, Figure 4 The electronic device 500 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0060] The following describes the computer-readable storage medium provided in the embodiments of this application. The computer-readable storage medium provided in the embodiments of this application stores computer instructions, which, when executed by a processor, implement the unsupervised learning-based blade damage identification method provided in the embodiments of this application. Specifically, the computer instructions can be built into or installed in the processor, so that the processor can implement the unsupervised learning-based blade damage identification method provided in the embodiments of this application by executing the built-in or installed computer instructions.

[0061] In addition, the unsupervised learning-based blade damage identification method provided in this application embodiment can also be implemented as a computer program product. The computer program product includes program code, which implements the unsupervised learning-based blade damage identification method provided in this application embodiment when running on a processor.

[0062] The computer program product provided in this application embodiment may employ one or more computer-readable storage media, which may be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination thereof. Specifically, more specific examples (a non-exhaustive list) of computer-readable storage media include electrical connections with one or more wires, portable disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0063] The computer program product provided in this application embodiment can be a CD-ROM and include program code, and can also run on electronic devices such as computers. However, the computer program product provided in this application embodiment is not limited thereto. In this application embodiment, the computer-readable storage medium can be any tangible medium that contains or stores program code, which can be used by or in conjunction with an instruction execution system, device, or apparatus.

[0064] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0065] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0066] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0067] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A blade damage identification method based on unsupervised learning, characterized in that, include: Acquire multi-source data to be tested for wind turbine blades; Based on the multi-source data to be detected, an identification model is used to predict the state of wind turbine blades; wherein, the identification model is obtained by determining a training positive sample dataset based on historical multi-source data; and by training the initial identification model based on the training positive sample dataset.

2. The blade damage identification method based on unsupervised learning according to claim 1, characterized in that, After acquiring the multi-source data to be detected for the wind turbine blades, the method further includes: Outlier removal is performed on the multi-source data to be detected to obtain the first multi-source data to be detected; Missing values ​​are processed on the first multi-source data to be detected to obtain the second multi-source data to be detected; The second multi-source data to be detected is standardized to obtain the standardized multi-source data to be detected.

3. The blade damage identification method based on unsupervised learning according to claim 1, characterized in that, Based on the multi-source data to be detected, a recognition model is used to predict the state of wind turbine blades, including: Based on the multi-source data to be detected, determine each first Euclidean distance between the multi-source data to be detected and the training positive sample data in the training positive sample dataset; Based on each of the first Euclidean distances, determine the k smallest first Euclidean distances of the multi-source data to be detected; Based on each of the k minimum first Euclidean distances, determine the first average Euclidean distance value of the training positive sample dataset; The state of the wind turbine blade is determined based on the first average Euclidean distance value and the threshold in the recognition model.

4. The blade damage identification method based on unsupervised learning according to claim 3, characterized in that, Determining the state of the wind turbine blade based on the first average Euclidean distance value and the threshold in the recognition model includes: If the first average Euclidean distance value is not greater than the threshold, then the wind turbine blade is determined to be in a normal state; wherein, the normal state means that the wind turbine blade has no risk of failure. If the first average Euclidean distance value is greater than the threshold, the wind turbine blade is determined to be in a state of failure or damage; wherein, the state of failure or damage is that the wind turbine blade has the risk of failure due to cracks and wear.

5. The blade damage identification method based on unsupervised learning according to any one of claims 1 to 4, characterized in that, Also includes: Acquire historical multi-source detection data; The historical detection multi-source data is standardized to obtain standardized historical detection multi-source data; The normal state data from the standardized historical multi-source detection data will be used as the training positive sample dataset.

6. The blade damage identification method based on unsupervised learning according to claim 5, characterized in that, Based on the training positive sample dataset, the initial recognition model is trained, including: Based on the training positive sample dataset, determine each second Euclidean distance between each training positive sample data in the training positive sample dataset and other training positive sample data in the training positive sample dataset; Based on each of the second Euclidean distances, determine m minimum second Euclidean distances corresponding to each of the training positive sample data; Based on each of the m minimum second Euclidean distances, determine the second average Euclidean distance value of the training positive sample dataset; Based on the second average Euclidean distance value, a threshold corresponding to the training positive sample dataset is determined, and based on the threshold, the recognition model is obtained.

7. The blade damage identification method based on unsupervised learning according to claim 6, characterized in that, Also includes: Select preset training positive sample data from the training positive sample dataset as validation sample data; Based on the verification sample data, the threshold is optimized to obtain the optimized recognition model.

8. A blade damage identification device based on unsupervised learning, characterized in that, include: The data acquisition module is used to acquire multi-source data to be tested from wind turbine blades; The data processing module is used to predict the state of wind turbine blades based on the multi-source data to be detected using an identification model; wherein the identification model is obtained by determining a training positive sample dataset based on historical multi-source data; and by training the initial identification model based on the training positive sample dataset.

9. An electronic device, characterized in that, It includes 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 unsupervised learning-based blade damage identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the unsupervised learning-based blade damage identification method as described in any one of claims 1 to 7.

Images