Wind turbine generator fault prediction method and system

Abstract

The embodiment of the invention provides a wind turbine generator fault prediction method and system, and belongs to the technical field of wind power. The method comprises the steps of collecting infrared imaging information of corresponding monitoring positions based on infrared monitoring devices arranged at the monitoring positions in a target wind turbine generator; performing target region segmentation on the infrared imaging information, and associating a target region subjected to association segmentation with a corresponding target component to obtain associated data; calling a pre-trained fault prediction model based on the associated data to obtain a fault prediction result of the corresponding target component; and matching and triggering a corresponding early warning strategy based on the fault prediction result. According to the scheme, the accuracy and the response speed of fault prediction are effectively improved, full-automatic monitoring and early warning of the equipment state are achieved, the shutdown risk caused by equipment faults is reduced, and the reliability and the operation and maintenance efficiency of the wind turbine generator are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a wind turbine generator set fault prediction method and a wind turbine generator set fault prediction system. Background Art

[0002] At present, wind turbines in the wind power industry often suffer from fatigue damage or loosening failures due to vibration on various rotating parts and bolted connections. Existing fault monitoring systems monitor the status of equipment by detecting parameters such as vibration and temperature of components. This requires engineering and technical personnel to manually analyze the data according to relevant standards to determine the fault problem. However, these systems have the limitation of being unable to predict faults in advance. The monitoring system will only send out an alarm signal after the rotating parts such as gearboxes and shafts have been obviously damaged or loosened. At this point, the damaged parts usually have reached the scrap standard and are difficult to continue to use, resulting in increased maintenance costs and may cause major economic losses such as wind turbine shutdown.

[0003] Currently, surveillance cameras have been installed in the cabins of wind turbines for remote monitoring, but these devices are mainly used to observe the overall operating conditions and cannot provide accurate fault prediction functions. In addition, after years of development, the wind power industry has accumulated a large amount of fault data, but the existing technology has not effectively used this data for intelligent analysis and prediction. Therefore, the existing technology has the following defects: 1) It is impossible to predict failures in advance: The existing monitoring system will only issue an alarm after the components are damaged or loose, and it is impossible to provide predictive warnings in the early stages of failure, causing the damage to expand to the point where it cannot be repaired.

[0004] 2) Reliance on manual analysis, low efficiency: Fault judgment relies on engineering and technical personnel to compare and analyze various parameters, which not only increases manpower input, but is also easily affected by subjective factors, making it difficult to guarantee accuracy and efficiency.

[0005] 3) Lack of effective use of historical fault data: Although the wind power industry has accumulated a large amount of historical data, the existing monitoring system fails to use this data for intelligent analysis, resulting in a waste of data resources.

[0006] In view of the above problems, a new wind turbine fault prediction scheme needs to be proposed. Summary of the invention

[0007] The purpose of the embodiments of the present invention is to provide a method and system for predicting faults of wind turbines, so as to at least solve the problems of low accuracy and insufficient efficiency of existing wind turbine fault predictions.

[0008] In order to achieve the above-mentioned purpose, the first aspect of the present invention provides a wind turbine fault prediction method, the method comprising: based on an infrared monitoring device set at each monitoring position in the target wind turbine, collecting infrared imaging information of the corresponding monitoring position; performing target area segmentation on the infrared imaging information, and associating the segmented target area with the corresponding target component to obtain associated data; calling a pre-trained fault prediction model based on the associated data to obtain a fault prediction result of the corresponding target component; matching and triggering a corresponding early warning strategy based on the fault prediction result.

[0009] Optionally, performing target area segmentation on the infrared imaging information includes: performing feature segmentation in the infrared image to identify valid component areas; performing contour recognition on each component area, and performing corresponding component area segmentation based on the recognized contour to obtain a segmented area of ​​each target component.

[0010] Optionally, the feature segmentation in the infrared image and the identification of the valid component area include: determining the acquisition field of view based on the acquisition parameter information of the current infrared image, and identifying all target components in the current acquisition field of view based on the determined acquisition field of view; determining the component positioning relationship based on the relative position relationship and size ratio relationship of each target component based on the identification of all target components; calibrating each target component in the current infrared image based on the component positioning relationship as the corresponding valid component area, and cutting the ineffective component area.

[0011] Optionally, the acquisition parameter information includes: focal length information and viewing angle information; the calibration of each target component in the current infrared image based on the component positioning relationship includes: identifying each target component in the current infrared image based on the YOLO algorithm, and taking the target component with the highest recognition confidence as the reference component; constructing a coordinate system based on the reference component, and identifying the size information of the reference component; based on the component positioning relationship, based on the relative position relationship and size ratio relationship with the reference component, calibrating each target component.

[0012] Optionally, the method also includes performing fault prediction model training; the training rules of the fault prediction model are: collecting historical infrared images of each component in a fault state, and dividing each target component in each historical infrared image; performing fault type and fault level labeling on the divided target components, and using the labeled infrared images as training samples; based on the training samples, model training is performed in a pre-constructed neural network to obtain a fault prediction model.

[0013] Optionally, the training samples include a training set and a test set; the model training is performed in a pre-constructed neural network based on the training samples to obtain a fault prediction model, including: taking the fault type and fault level as the target, and the infrared images in the training set as input parameters, performing model training in a pre-constructed neural network to obtain an initial model; and verifying the initial model based on the test set, and using the verified initial model as the fault prediction model.

[0014] Optionally, matching and triggering corresponding early warning strategies based on fault prediction results include: if the fault type of the fault prediction result triggers a preset critical fault type, triggering a shutdown operation and pushing an alarm message to the supervisory end; if the fault type of the fault result does not trigger a preset critical fault type, matching an early warning plan of a corresponding level based on the fault level of the fault result; wherein early warning plans of corresponding levels include: a continuous monitoring plan, which continuously monitors corresponding target components and does not push alarm information; an abnormal reminder plan, which only reminds of abnormal conditions and does not push alarm information; and a fault alarm plan, which pushes alarm information to the supervisory end.

[0015] Optionally, the push rule for pushing alarm information to the supervision end is: determine whether the wired communication is normal; if the wired communication is normal, push the alarm information based on the wired communication; if the wired communication is abnormal, push the alarm information based on the wireless communication.

[0016] The second aspect of the present invention provides a wind turbine fault prediction system, the system comprising: a collection unit, which is used to collect infrared imaging information of corresponding monitoring positions based on infrared monitoring devices set at various monitoring positions in the target wind turbine; a segmentation unit, which is used to perform target area segmentation on the infrared imaging information, and associate the segmented target area with the corresponding target component to obtain associated data; a prediction unit, which is used to call a pre-trained fault prediction model based on the associated data to obtain a fault prediction result of the corresponding target component; and an early warning unit, which is used to match and trigger a corresponding early warning strategy based on the fault prediction result.

[0017] On the other hand, the present invention provides a computer-readable storage medium, on which instructions are stored, which, when executed on a computer, enable the computer to execute the above-mentioned wind turbine fault prediction method.

[0018] Through the above technical solution, the solution of the present invention collects infrared imaging information at corresponding positions by setting up infrared monitoring devices at various monitoring positions in the wind turbine. The system can segment the target area in real time and associate it with related components. Combined with the pre-trained fault prediction model, the state of the target component is intelligently predicted, and the corresponding early warning strategy is matched and triggered when the fault first occurs. This method effectively improves the accuracy and response speed of fault prediction, realizes fully automatic monitoring and early warning of equipment status, reduces the risk of downtime due to equipment failure, and significantly improves the reliability and operation and maintenance efficiency of wind turbines.

[0019] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings: Figure 1 is a flow chart of the steps of a wind turbine fault prediction method provided by an embodiment of the present invention; Figure 2 It is a fault intelligent prediction logic diagram provided by an embodiment of the present invention; Figure 3 It is a system structure diagram of a wind turbine fault prediction system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0021] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0022] Figure 1 FIG. 1 is a flow chart of a method for predicting wind turbine faults provided by an embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a method for predicting faults of a wind turbine generator set, the method comprising: Step S10: Based on the infrared monitoring devices arranged at the respective monitoring positions in the target wind turbine generator system, infrared imaging information of the corresponding monitoring positions is collected.

[0023] Specifically, based on the infrared monitoring devices installed at each monitoring position in the target wind turbine, the system can accurately collect infrared imaging information at each monitoring position to achieve comprehensive thermal imaging monitoring. The infrared monitoring devices are installed near the key components of the wind turbine, such as gearboxes, shafts, bearings, and bolted joints, which are often prone to fatigue damage due to long-term vibration and load changes. Through the infrared monitoring devices, the surface temperature of these components can be monitored in real time, their thermal imaging data can be recorded, and any abnormal temperature changes can be quickly captured.

[0024] The thermal imaging information collected by the infrared monitoring device is mainly used to reflect the temperature changes and heat distribution of each component of the equipment. By analyzing the thermal imaging data, the system can identify whether there is abnormal energy dissipation in each component. Usually, fatigue damage or loose failure of components will manifest as a significant increase in local temperature, and these temperature anomalies can be reflected in real time through infrared thermal imaging. The infrared monitoring device can provide high-precision thermal imaging data to ensure the accuracy and timeliness of fault prediction. The monitoring process is fully automated and does not require manual intervention, thereby reducing the workload and cost of manual monitoring.

[0025] Step S20: performing target area segmentation on the infrared imaging information, and associating the segmented target area with the corresponding target component to obtain associated data.

[0026] Specifically, the target area segmentation of the infrared imaging information includes: performing feature segmentation in the infrared image to identify effective component areas; performing contour recognition on each component area, and performing corresponding component area segmentation based on the recognized contour to obtain the segmented area of ​​each target component.

[0027] Furthermore, the feature segmentation in the infrared image and the identification of the valid component area include: determining the acquisition field of view based on the acquisition parameter information of the current infrared image, and identifying all target components in the current acquisition field of view based on the determined acquisition field of view; determining the component positioning relationship based on the relative position relationship and size ratio relationship of each target component based on the identification of all target components; calibrating each target component in the current infrared image based on the component positioning relationship as the corresponding valid component area, and cutting the ineffective component area.

[0028] Specifically, the acquisition parameter information includes: focal length information and viewing angle information; the calibration of each target component in the current infrared image based on the component positioning relationship includes: identifying each target component in the current infrared image based on the YOLO algorithm, and taking the target component with the highest recognition confidence as the reference component; constructing a coordinate system based on the reference component, and identifying the size information of the reference component; based on the component positioning relationship, based on the relative position relationship and size ratio relationship with the reference component, calibrating each target component.

[0029] In the embodiment of the present invention, during the feature segmentation of the infrared image, the system first needs to determine the acquisition field of view based on the acquisition parameter information of the current infrared image. These acquisition parameter information include focal length information and viewing angle information, which are the basis for determining all target components within the acquisition field of view. By using these acquisition parameters, the system can accurately determine the position and characteristics of each target component in the infrared image and identify the effective component area. This step is crucial for the subsequent segmentation and calibration process, because only through accurate component identification and feature extraction can the reliability and accuracy of the fault prediction function of the entire system be ensured.

[0030] Furthermore, after determining the acquisition field of view, the system will identify all target components within the current acquisition field of view, and determine the relative position relationship and size ratio relationship of each target component based on these identifications. In this way, the system can construct the positioning relationship of the components in the current infrared image, and then effectively calibrate the target components. Specifically, the system will calibrate each target component in the infrared image, treat these calibrated component areas as valid component areas, and cut other irrelevant areas as non-valid component areas to reduce noise and interference information in the image.

[0031] Furthermore, during the feature segmentation process, the system will also use the YOLO algorithm to identify each target component in the current infrared image. YOLO (You Only Look Once) is a target detection algorithm based on a convolutional neural network that can quickly identify multiple targets in an image in a single detection. The application of the YOLO algorithm enables the system to quickly and accurately identify the position and shape of each target component in the infrared image, and use the target component with the highest confidence as the reference component. The selection of the reference component is crucial for building a coordinate system and calibrating other target components because it provides a standard reference point.

[0032] Furthermore, based on the reference component, the system will construct a coordinate system and identify and measure the dimensional information of the reference component. By accurately measuring the dimensions of the reference component, the system can provide a reference for the calibration of other target components. Next, the system calibrates other target components based on the positioning relationship of the components, combined with the relative position relationship and size ratio relationship of the reference components. This calibration process ensures the accurate positioning of each target component and generates segmented areas for each component. These segmented areas will be used as input data for the subsequent fault prediction model to evaluate the status of the target component.

[0033] By segmenting and calibrating the target area, the system can obtain detailed temperature distribution information for each target component. Infrared imaging information mainly reflects the temperature of the component surface. When fatigue damage or loose failure occurs in the component, it usually manifests as an abnormal increase in local temperature. By segmenting and identifying each target component in the infrared image, the system can accurately identify the components with abnormal temperatures and record these temperature anomalies as characteristics of potential failures.

[0034] Based on the solution of the present invention, the segmentation process of the target area ensures that the input data of the fault prediction model has high accuracy, thereby improving the reliability and effectiveness of the prediction results. By performing target area segmentation on infrared imaging information and associating the target area with the corresponding components, the system can obtain accurate associated data. Combining the YOLO algorithm and the fault prediction model trained by deep learning, the system can perform intelligent predictions on each target component and match the corresponding early warning strategy based on the prediction results. This method significantly improves the accuracy and response speed of fault prediction, realizes fully automatic monitoring and early warning of equipment status, reduces the risk of downtime caused by equipment failure, and greatly improves the reliability and operation and maintenance efficiency of wind turbines.

[0035] Step S30: calling a pre-trained fault prediction model based on the associated data to obtain a fault prediction result of the corresponding target component.

[0036] Specifically, the method also includes executing fault prediction model training; the training rules of the fault prediction model are: collecting historical infrared images of each component in a fault state, and dividing each target component in each historical infrared image; performing fault type and fault level labeling on the divided target components, and using the labeled infrared images as training samples; based on the training samples, model training is performed in a pre-constructed neural network to obtain a fault prediction model.

[0037] Furthermore, the training samples include a training set and a test set; the model training is performed in a pre-constructed neural network based on the training samples to obtain a fault prediction model, including: taking the fault type and fault level as the target, and the infrared images in the training set as input parameters, performing model training in the pre-constructed neural network to obtain an initial model; and verifying the initial model based on the test set, and using the verified initial model as the fault prediction model.

[0038] In an embodiment of the present invention, in order to build a reliable fault prediction model, the system needs to train the model through a large amount of historical infrared image data. First, historical infrared images of various components of the wind turbine in a faulty state are collected, and each target component is divided in these historical images to ensure that the temperature information and state of each component can be accurately calibrated and recorded. Subsequently, the system labels the fault type and fault level of the divided target components. These labeled information reflects the status of different components in different fault scenarios, including the severity and specific type of the fault. The labeled infrared images are used as training samples to build a fault prediction model.

[0039] During the model training process, the system inputs these annotated infrared images into the pre-built neural network for training. The neural network continuously adjusts its weights and parameters to learn the characteristic patterns of each target component under different fault types and levels, so that the model can identify similar fault characteristics. The goal of the entire model training process is to obtain a model that can accurately predict the fault type and severity of each component of the wind turbine. In order to ensure the generalization ability and prediction accuracy of the model, the system divides all training samples into training sets and test sets, where the training set is used for neural network training, and the test set is used to verify and validate the initial model obtained through training.

[0040] Specifically, the system first uses the infrared images in the training set as input parameters, and takes the fault type and fault level as the target, and performs model training in the pre-built neural network to obtain an initial model. The performance of the initial model depends on the effective learning ability of the neural network for various fault characteristics. Therefore, the training process requires a lot of iterations and optimizations to ensure that the model can accurately capture the temperature characteristics and abnormal patterns of each component under fault conditions. After obtaining the initial model, the system will perform verification on the initial model based on the test set, and verify its prediction accuracy and reliability by inputting data that is not involved in the training in the test set into the model. Only when the performance of the initial model on the test set meets the expected standards will the model be determined as the final fault prediction model.

[0041] Through the above training process, the fault prediction model obtained has the ability to detect and accurately judge the faults of various components of wind turbines at an early stage. When the system is running, the infrared images collected in real time will be input into the fault prediction model. The model will analyze the status of the target component according to the characteristic patterns learned during the training process to determine whether there is a fault and the type and severity of the fault. If a potential fault risk is detected, the system will trigger the corresponding early warning mechanism based on the prediction results so that the operation and maintenance personnel can take timely measures to avoid further deterioration of the fault.

[0042] By training the fault prediction model, the system can accurately predict the fault type and fault level of key components of wind turbines. The labeling and classification of historical infrared images during the training process enables the model to have a strong fault feature recognition capability; and the neural network-based model training method ensures that the model can still make accurate judgments when faced with complex fault modes. The introduction of this training method not only improves the prediction accuracy of the model, but also significantly enhances the system's overall monitoring and early warning capabilities for wind turbine faults.

[0043] Step S40: Match and trigger corresponding early warning strategies based on the fault prediction results.

[0044] Specifically, if the fault type of the fault prediction result triggers the preset key fault type, the shutdown operation is triggered and the alarm information is pushed to the supervision end; if the fault type of the fault result does not trigger the preset key fault type, the warning plan of the corresponding level is matched based on the fault level of the fault result; among which, the warning plans of the corresponding levels include: continuous monitoring plan, which continuously monitors the corresponding target components and does not push alarm information; abnormal reminder plan, which only reminds of abnormal status and does not push alarm information; fault alarm plan, which pushes alarm information to the supervision end.

[0045] Furthermore, the push rule for pushing alarm information to the supervisory end is: determine whether the wired communication is normal; if the wired communication is normal, push the alarm information based on the wired communication; if the wired communication is abnormal, push the alarm information based on the wireless communication.

[0046] In an embodiment of the present invention, after the fault prediction model detects a fault in the target component of the wind turbine, the system will match the corresponding early warning strategy according to the type and level of the fault prediction result. When the fault type of the fault prediction result triggers the preset key fault type, the system will immediately trigger a shutdown operation to prevent the further deterioration of the fault from causing serious damage to the wind turbine. At the same time, the system will also push alarm information to the supervision end to remind the operation and maintenance personnel to perform maintenance and repairs in a timely manner. This timely shutdown operation can effectively avoid safety accidents caused by component damage and ensure the safe operation of the wind turbine.

[0047] If the preset critical fault type is not triggered, the system will match the corresponding warning plan based on the fault level of the fault prediction result. Among them, the corresponding level of warning plans include three types: 1) Continuous monitoring solution: When the fault level is low and has little impact on the operation of the wind turbine, the system will choose the continuous monitoring solution. At this time, the system will continuously monitor the status changes of the corresponding target components, but will not push alarm information to the supervisory end. This solution is suitable for minor faults or situations that do not temporarily affect the operation of the equipment. It aims to observe the development trend of the fault through continuous monitoring to determine whether further maintenance measures are needed.

[0048] 2) Abnormal reminder scheme: When the fault level is at a medium level, there is a certain potential risk but it has not reached a serious level, the system will select the abnormal reminder scheme. This scheme only reminds the operation and maintenance personnel of the abnormal state, but does not push formal alarm information. This reminder can help the operation and maintenance personnel understand the potential problems of the wind turbine in advance, so that they can be prepared and take quick measures when the fault worsens.

[0049] 3) Fault alarm scheme: When the fault level is high and may affect the normal operation of the wind turbine, the system will select the fault alarm scheme and push the alarm information to the supervisory end. The purpose of this scheme is to promptly remind the operation and maintenance personnel to carry out maintenance when the fault may cause significant losses, so as to prevent the fault from further deteriorating.

[0050] In the process of pushing alarm information to the supervisory end, the system will determine whether the wired communication is normal according to the push rules. If the wired communication is normal, the system will prioritize the push of alarm information based on wired communication; if the wired communication is abnormal, the system will automatically switch to wireless communication to push the alarm information, ensuring that the alarm information can reach the supervisory end in time. The design of this dual communication mechanism greatly improves the reliability of the system in complex environments and ensures the timely transmission of alarm information.

[0051] Through this multi-level early warning strategy, the system can flexibly select the corresponding treatment plan according to different fault types and fault levels, ensuring that the wind turbine can receive the most appropriate treatment when a fault occurs. Shutdown protection for critical fault types can effectively prevent major accidents, while continuous monitoring and abnormal reminder solutions for minor faults can help operation and maintenance personnel discover potential problems in advance and reduce equipment maintenance costs.

[0052] In a possible implementation, Figure 2, a fault intelligent prediction logic is proposed. The system collects the operating data of each component through infrared thermal imaging. If data anomalies are detected, the system will compare and analyze the abnormal data with the fault database to identify potential fault characteristics. Next, the intelligent main control system compares and analyzes the results to make predictions and determine whether there are fault trends or hidden dangers. If the prediction results indicate that further measures are needed, the system will enter the main control action stage to achieve protection and maintenance of the equipment. On the contrary, if no data anomalies are detected or no further action is required, the process ends directly. This process can effectively achieve fault warning and active protection of wind turbines through multi-step screening and judgment.

[0053] Figure 3 FIG. 1 is a system structure diagram of a wind turbine fault prediction system provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides a wind turbine fault prediction system, the system comprising: a collection unit, for collecting infrared imaging information of corresponding monitoring positions based on infrared monitoring devices arranged at each monitoring position in the target wind turbine; a segmentation unit, for performing target area segmentation on the infrared imaging information, and associating the segmented target area with the corresponding target component to obtain associated data; a prediction unit, for calling a pre-trained fault prediction model based on the associated data to obtain a fault prediction result of the corresponding target component; and an early warning unit, for matching and triggering a corresponding early warning strategy based on the fault prediction result.

[0054] The embodiment of the present invention further provides a computer-readable storage medium, on which instructions are stored, which, when executed on a computer, enable the computer to execute the above-mentioned wind turbine fault prediction method.

[0055] Those skilled in the art will understand that all or part of the steps in the method for implementing the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0056] The optional embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, the technical scheme of the embodiments of the present invention can be subjected to a variety of simple modifications, and these simple modifications all belong to the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0057] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A method for predicting wind turbine faults, characterized in that: The method comprises: Based on the infrared monitoring devices set at each monitoring position in the target wind turbine, infrared imaging information of the corresponding monitoring position is collected; Performing target area segmentation on the infrared imaging information, and associating the segmented target area with the corresponding target component to obtain associated data; Based on the associated data, a pre-trained fault prediction model is called to obtain a fault prediction result of a corresponding target component; Match and trigger corresponding early warning strategies based on fault prediction results.

2. The method according to claim 1, characterized in that The performing target area segmentation on the infrared imaging information comprises: Perform feature segmentation in infrared images to identify valid component areas; The contour of each component area is recognized, and the corresponding component area is segmented based on the recognized contour to obtain the segmented area of ​​each target component.

3. The method according to claim 2, characterized in that The step of performing feature segmentation in the infrared image and identifying effective component areas includes: Determine the acquisition field of view based on the acquisition parameter information of the current infrared image, and identify all target components within the current acquisition field of view based on the determined acquisition field of view; Based on the identification of all target components, the relative position relationship and size ratio relationship of each target component are determined to determine the component positioning relationship; Based on the component positioning relationship, each target component is calibrated in the current infrared image as the corresponding valid component area, and the ineffective component area is cut.

4. The method according to claim 3, characterized in that The acquisition parameter information includes: Focal length information and viewing angle information; The calibration of each target component in the current infrared image based on the component positioning relationship includes: Based on the YOLO algorithm, each target component is identified in the current infrared image, and the target component with the highest recognition confidence is used as the reference component; Construct a coordinate system based on the reference component and identify the size information of the reference component; Based on the component positioning relationship, and based on the relative position relationship and size ratio relationship with the reference component, each target component is calibrated.

5. The method according to claim 1, characterized in that The method also includes performing fault prediction model training; The training rules of the fault prediction model are: Collect historical infrared images of each component in a fault state, and divide each target component in each historical infrared image; Perform fault type and fault level labeling on the divided target components, and use the labeled infrared images as training samples; Model training is performed in a pre-built neural network based on the training samples to obtain a fault prediction model.

6. The method according to claim 5, characterized in that The training samples include a training set and a test set; The performing model training in a pre-built neural network based on the training samples to obtain a fault prediction model includes: Taking the fault type and fault level as the target and the infrared images in the training set as the input parameters, the model is trained in the pre-built neural network to obtain the initial model; The initial model is verified based on the test set, and the initial model that passes the verification is used as the fault prediction model.

7. The method according to claim 1, characterized in that Match and trigger corresponding warning strategies based on fault prediction results, including: If the fault type of the fault prediction result triggers the preset critical fault type, the shutdown operation is triggered and the alarm information is pushed to the supervisory end; If the fault type of the fault result does not trigger the preset critical fault type, the fault level of the fault result is matched with the corresponding level of warning scheme; wherein, The corresponding levels of early warning plans include: Continuous monitoring solution: continuously monitors the corresponding target components without sending alarm information; The abnormal reminder solution only reminds you of abnormal status and does not push alarm information; Fault alarm solution, push alarm information to the supervision end.

8. The method according to claim 7, characterized in that The push rules for pushing alarm information to the supervisor are: Determine whether wired communication is normal; If the wired communication is normal, the alarm information is pushed based on the wired communication; If the wired communication is abnormal, the alarm information is pushed based on wireless communication.

9. A wind turbine fault prediction system, characterized in that: The system comprises: A collection unit, used to collect infrared imaging information of corresponding monitoring positions based on infrared monitoring devices arranged at each monitoring position in the target wind turbine; A segmentation unit, used to perform target area segmentation on the infrared imaging information, and associate the segmented target area with the corresponding target component to obtain associated data; A prediction unit, configured to call a pre-trained fault prediction model based on the associated data to obtain a fault prediction result of a corresponding target component; The early warning unit is used to match and trigger corresponding early warning strategies based on the fault prediction results.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the wind turbine fault prediction method according to any one of claims 1 to 8.

Images