Fault intelligent monitoring and rapid handling method and device for wind power plant fan

Abstract

The invention provides a fault intelligent monitoring and rapid handling method and device for a wind power plant fan. The method comprises the steps of obtaining multi-dimensional operation state data of a target fan; preprocessing the multi-dimensional operation state data at an edge computing node to generate a high-dimensional feature vector; inputting the high-dimensional feature vector into a first intelligent analysis model deployed at a cloud end to output real-time fault early warning information and component fault positioning information for a target fan, and automatically matching and generating disposal information from a preset disposal scheme library; and after the disposal operation is executed based on the disposal information, automatically collecting operation data after the target fan is restarted, and comparing the operation data with a preset verification reference to generate a quantized disposal effect verification result. According to the method, the diagnosis and prediction dual intelligent model is deployed, so that real-time early warning and accurate component positioning can be performed on the current fault, the future fault probability can be predicted, and the operation and maintenance strategy is converted from passive response to active planning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power generators, in particular to a method and device for intelligent monitoring and rapid disposal of faults of wind turbines in a wind farm. BACKGROUND

[0002] Predictive operation and maintenance of wind power generators (referred to as "wind turbines" for short) is a key link to improve the operation efficiency of wind farms and reduce operation and maintenance costs. Among them, the wind turbine fault early warning technology collects and analyzes the operation data of the wind turbine in real time, uses mechanism models or data-driven algorithms (such as machine learning algorithms) to predict the probability of a specific fault occurring in the wind turbine in the future period of time, thereby providing decision support for formulating a scientific operation and maintenance strategy. Effective fault early warning can significantly reduce unplanned downtime and avoid major equipment losses, thereby improving the economic benefits of the whole life cycle of wind power generation.

[0003] In the prior art, the implementation of wind turbine fault early warning is usually to integrate various early warning algorithms in a monolithic application program, and then deploy the program on the local computer of the wind farm booster station or the server of the remote centralized monitoring center. This centralized monolithic architecture and localized deployment mode exposes a series of problems in practice: 1. Algorithm iteration and management difficulty: The research and optimization of fault early warning algorithms is a continuous process, and new algorithms are constantly being developed, and existing algorithms also need to be quickly iterated according to field data. In the existing mode, the updating, testing and verification of these algorithms are usually completed by the research and development team of the headquarters or regional headquarters, and sometimes the local operation and maintenance personnel of the wind farm need to participate. Because the application program is monolithic and is deployed in multiple geographic locations, new and old algorithm versions coexist, forming an "algorithm fragmentation" phenomenon, making it difficult to manage, verify and use uniformly and efficiently.

[0004] 2. High deployment and maintenance cost and low efficiency: Whenever an early warning algorithm needs to be updated or added, the developer must repeat the code integration, compilation, testing and deployment work in each locally deployed application program. This process not only consumes time and effort, greatly reducing the efficiency of development and deployment, but also brings high human coordination and project management costs, which cannot meet the requirements of modern wind power operation for agility and intelligence. SUMMARY

[0005] The present application provides a method and device for intelligent monitoring and rapid disposal of faults of wind turbines in a wind farm, to solve the defects of algorithm management confusion, low iteration update efficiency and high deployment cost of the wind turbine fault early warning system in the prior art due to the architecture and deployment mode, and to provide a method that can conveniently and efficiently develop, test, deploy and uniformly manage fault early warning algorithms.

[0006] According to the first aspect of the present application, a method for intelligent monitoring and rapid disposal of faults of wind turbines in a wind farm is provided, comprising: acquiring multi-dimensional operating state data of a target wind turbine, including at least mechanical, electrical and environmental dimensions; preprocessing the multi-dimensional operating state data at an edge computing node to generate a high-dimensional feature vector representing the real-time operating state of the target wind turbine; inputting the high-dimensional feature vector into a first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault positioning information for the target wind turbine, wherein the first intelligent analysis model is trained based on historical fault data and normal operating data; According to the real-time fault warning information and the component fault positioning information, automatically matching and generating disposal information from a pre-set disposal scheme library; inputting the high-dimensional feature vector and historical operating data into a second intelligent prediction model deployed in the cloud to output the fault probability of the target wind turbine in a future pre-set period; After performing a disposal operation based on the disposal information, automatically collecting operating data of the target wind turbine after restarting and comparing with a pre-set verification benchmark to generate a quantitative disposal effect verification result.

[0007] According to an embodiment of the present application, the step of acquiring multi-dimensional operating state data of a target wind turbine, including at least mechanical, electrical and environmental dimensions, specifically comprises: acquiring vibration signals, acoustic signals, electrical signals and environmental parameters through sensing devices deployed on key components of the target wind turbine, wherein the sensing devices include at least high-frequency vibration sensors, voiceprint sensors, current and voltage sensors and environmental sensors; Based on the acquired vibration signals, acoustic signals, electrical signals and environmental parameters, the multi-dimensional operating state data is generated; And, the step of preprocessing the multi-dimensional operating state data at an edge computing node to generate a high-dimensional feature vector representing the real-time operating state of the target wind turbine, specifically comprises: Performing wavelet threshold denoising and normalization processing on the multi-dimensional operating state data, and extracting time domain, frequency domain and cepstrum features to construct the high-dimensional feature vector.

[0008] Specifically, the embodiment provides an implementation of obtaining multi-dimensional operation state data of a target fan, at least including mechanical, electrical and environmental three dimensions, and constructs an intelligent sensing front end which guarantees data quality from the source and completes efficient processing at the edge. It not only ensures the comprehensiveness and accuracy of input data, but also significantly improves the usability and information density of data through denoising, normalization and feature extraction, and optimizes the data transmission efficiency of the system.

[0009] According to an embodiment of the application, the step of inputting the high-dimensional feature vector into the first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault positioning information specifically includes: The computing power of the edge computing node is not less than 2TOPS, the sampling frequency of the vibration signal collected by the edge computing node is greater than or equal to 10kHz, and the dimension of the generated high-dimensional feature vector is 128; The first intelligent analysis model deployed in the cloud, which internally contains a historical fault sample library, stores high-dimensional feature vector samples corresponding to multiple core component fault types; The output real-time fault warning information specifically includes: comparing each dimension value of the real-time input 128-dimensional high-dimensional feature vector with preset "emergency fault" feature threshold and "early fault" feature threshold one by one, when any dimension value exceeds the "emergency fault" threshold, outputting "emergency fault" warning; when no dimension exceeds the "emergency fault" threshold but at least one dimension exceeds the "early fault" threshold, outputting "early fault" warning; The output component fault positioning information specifically includes: after outputting "emergency fault" or "early fault" warning, extracting a feature subset related to the over-limit dimension from the 128-dimensional high-dimensional feature vector, and extracting the feature template of each component from the preset component fault feature template library of the core component; by calculating the cosine similarity between the feature subset and each feature template, when a certain similarity calculation result is greater than or equal to a preset value, the component corresponding to the template with the highest cosine similarity is output as the final fault positioning result; the core component at least includes a gear box, a generator and a main bearing.

[0010] Specifically, the embodiment provides an implementation of inputting the high-dimensional feature vector into the first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault positioning information. Through detailed limitation of key links such as computing power, sampling rate, feature dimension, warning logic and positioning algorithm, they work together to finally realize earlier and more accurate discovery of fan faults, more scientific and more hierarchical warning, and faster and more accurate positioning and disposal, thereby comprehensively improving the intelligent level, safety and economic benefit of wind farm operation and maintenance.

[0011] According to an embodiment of the present application, the first intelligent analysis model is a hybrid network model combining a convolutional neural network and a long short-term memory network. The step of outputting the real-time fault warning information for the target wind turbine specifically comprises: The high-dimensional feature vector is compared with a preset multi-level feature threshold to determine the current state of the target wind turbine, which is one of "emergency failure", "early failure" or "normal operation"; The step of outputting the real-time fault warning information and the component fault positioning information for the target wind turbine specifically comprises: In the case of "emergency failure" or "early failure", the cosine similarity between the fault-related feature subset in the high-dimensional feature vector and the historical component fault feature template is calculated, and the component with a similarity higher than a preset positioning threshold is determined as the fault positioning result.

[0012] Specifically, the embodiment provides an implementation of outputting real-time fault warning information for the target wind turbine. By introducing the CNN-LSTM hybrid model, multi-level state threshold judgment and fault positioning based on cosine similarity, the method has the advantages of high precision, hierarchical and accurate positioning. It builds a complete closed loop from data collection to intelligent analysis, hierarchical warning and accurate positioning, which is the core technology of realizing intelligent, refined and predictive operation and maintenance of wind farms, and has high practical value and economic benefits.

[0013] According to an embodiment of the present application, the method further comprises: A full-amount traceability database covering the whole life cycle of the wind farm is constructed, wherein the full-amount traceability database comprises a first wind turbine file corresponding to the wind turbine one by one, and a second wind turbine file corresponding to the wind turbine fault one by one. After the treatment effect verification result is generated, the full-process data related to the current fault event is automatically filed to the full-amount traceability database, wherein the full-process data comprises original multi-dimensional operating state data triggering the warning, generated high-dimensional feature vector, warning and positioning results of the first intelligent analysis model, historical prediction records of the second intelligent prediction model, matched treatment information, personnel operation records and on-site photos of each treatment link, and quantitative comparison data and final verification result of treatment effect verification.

[0014] Specifically, the embodiment provides an implementation of constructing a full-amount traceability database covering the whole life cycle of a wind farm, and each fault handling process is converted into valuable knowledge assets by constructing a dynamic learning and self-optimizing data closed-loop system. This not only greatly enhances the performance and accuracy of the intelligent early warning model, but also provides a solid data foundation for realizing the whole life cycle health management of the wind turbine, optimizing the operation and maintenance process and depositing expert experience, and is a key step for realizing the transition from "digitalization" to "intelligence" and "wisdom" of the wind farm station.

[0015] According to an embodiment of the present application, the step of automatically matching and generating treatment information from the preset treatment scheme library according to the real-time fault early warning information and the component fault positioning information specifically comprises: Based on the fault level of the real-time fault early warning information and the fault component of the component fault positioning information, searching in a treatment scheme library containing at least N sets of standardized treatment schemes, N being a positive integer greater than or equal to 2; According to the optimization principle of the shortest treatment time and the lowest treatment cost, the optimal treatment scheme is matched from the searched applicable schemes; Generate structured treatment information containing fault details, fault cause analysis, optimal treatment scheme operation steps, required spare parts list and safety precautions, and push the information to different permission operation and maintenance personnel terminals according to the fault level through the Internet of Things gateway.

[0016] Specifically, the embodiment provides an implementation of automatically matching and generating treatment information from a preset treatment scheme library, which solves the technical problems of slow response, decision-making relying on artificial experience, uncontrollable cost, low information transmission efficiency and other technical problems in traditional fault handling by introducing multi-dimensional search, optimization decision based on economic principle, structured information output and hierarchical information distribution, and finally realizes the core goal of "fast, accurate, economic and safe" of wind farm fault handling.

[0017] According to an embodiment of the present application, the step of generating a quantitative treatment effect verification result specifically comprises: After the treatment operation is completed and the target wind turbine is restarted, the running data is continuously collected for at least a first preset time; A double-layer verification system is constructed to compare the collected running data with a verification benchmark; The double-layer verification system comprises: The first layer of basic parameter verification: comparing the basic running parameters after the wind turbine is restarted with the standard running parameter range of the wind turbine under the current working condition to determine whether there is abnormal fluctuation, the basic running parameters at least including speed, power and temperature; Second-level specific indicator verification: The specific performance indicators after the faulty component is restarted are compared with the historical normal values ​​of the faulty component before the fault occurred to determine whether the specific indicators have recovered to the normal range. The specific performance indicators include at least the vibration characteristic frequency and the electrical harmonic content. If both the first-level basic parameter verification and the second-level special indicator verification pass, a verification result of "successful disposal" is generated.

[0018] Specifically, this embodiment provides an implementation method for generating quantitative verification results of the treatment effect. Through a carefully designed, data-driven, two-layer verification model, it solves the core pain point of "how to confirm that the repair is really done" in fault treatment. It transforms the traditional, vague, and subjective post-repair evaluation into a precise, efficient, objective, and comprehensive automated verification process, bringing a qualitative improvement to the intelligence and reliability of the entire wind farm operation and maintenance system.

[0019] According to one embodiment of the present invention, it further includes: When the first intelligent analysis model generates an early warning of "emergency fault" or "early fault", the generation and hierarchical push of handling information are completed. The "emergency fault" information is pushed to the on-site operation and maintenance personnel, team leaders and wind farm managers at the same time, and the "early fault" information is pushed to the on-site operation and maintenance personnel and team leaders. The maintenance personnel receive the handling information through a mobile terminal and scan or click to confirm the core progress nodes in the handling process in real time. The background automatically records the timestamp and geographical location information of each node. The core progress nodes include at least "instruction received", "on-site in place", "spare parts replacement", "handling completed" and "application for trial operation". The steps for real-time scanning or clicking to confirm key progress nodes in the processing process specifically include: Each standardized handling solution in the handling solution library has a QR code embedded. When the operation and maintenance personnel perform the real-time scanning or click to confirm, they need to scan the corresponding QR code. The mobile terminal will automatically display the graphic or video operation guide for this step and record the operation completion status. After the maintenance personnel submit an "application for trial operation", the system remotely unlocks and starts the wind turbine, automatically executing the two-layer verification system; If the verification fails, a secondary diagnostic report will be automatically generated, analyzing the cause and recommending supplementary treatment measures; If the verification is successful, a complete acceptance report is automatically generated, including all process node records, operation photos, call recordings, and a quantitative verification report. This report is then pushed to the wind farm manager for electronic approval, and all data is packaged, encrypted, and stored in the full traceability database.

[0020] Specifically, this embodiment provides a method for generating and hierarchically pushing disposal information, as well as for real-time scanning or clicking to confirm key progress nodes in the disposal process. This solves the core pain points in traditional wind power operation and maintenance, such as poor information transmission, lack of transparency, inconsistent standards, difficulty in evaluating effects, and difficulty in tracing data. Ultimately, it achieves comprehensive technical effects such as improving operation and maintenance efficiency, ensuring disposal quality, reducing management costs, and laying a solid foundation for long-term data-driven optimization.

[0021] According to one embodiment of the present invention, the second intelligent prediction model is a gradient boosting tree model; The step of outputting the failure probability of the target wind turbine within a future preset period specifically includes: By integrating historical operating data, fault data, maintenance records, and meteorological data of the target wind turbine and other wind turbines at the same site, a predictive feature set is constructed. The gradient boosting tree model is used to analyze the predicted feature set, calculate and output the probability values ​​of various preset faults occurring in the target wind turbine within a second preset time period; The method further includes: Through an incremental learning mechanism, newly added operational and fault data are automatically included in the training set at a preset training cycle to retrain the gradient boosting tree model, thereby dynamically optimizing the prediction accuracy of the gradient boosting tree model.

[0022] Specifically, this embodiment provides an implementation method for outputting the failure probability of the target wind turbine in a future preset period. It not only solves the problem of "whether it can be predicted", but also solves a series of key engineering problems such as "how to accurately predict, how to effectively utilize the prediction results, and how to make the prediction capability never outdated". It provides strong technical support for realizing truly intelligent and refined operation and maintenance management of wind farms.

[0023] According to a second aspect of the present invention, a fault intelligent monitoring and rapid handling device for wind turbines in a wind farm includes: The data acquisition module is used to acquire multi-dimensional operating status data of the target wind turbine, including at least three dimensions: mechanical, electrical, and environmental. The vector generation module is used to preprocess the multi-dimensional operating status data at the edge computing node to generate a high-dimensional feature vector representing the real-time operating status of the target wind turbine. The information output module is used to input the high-dimensional feature vector into the first intelligent analysis model deployed in the cloud, so as to output real-time fault warning information and component fault location information for the target wind turbine. The first intelligent analysis model is trained based on historical fault data and normal operation data. The information matching module is used to automatically match and generate handling information from a preset handling solution library based on the real-time fault warning information and component fault location information; The probability output module is used to input the high-dimensional feature vector and historical operating data into the second intelligent prediction model deployed in the cloud, so as to output the failure probability of the target wind turbine in the future preset period. The treatment verification module is used to automatically collect the operating data of the target wind turbine after restarting after performing treatment operations based on the treatment information, and compare it with the preset verification benchmark to generate a quantitative treatment effect verification result.

[0024] The above-mentioned one or more technical solutions of the present invention have at least one of the following technical effects: The present invention provides a method and device for intelligent monitoring and rapid handling of faults in wind turbines in wind farms. Through a cloud-edge collaborative architecture, data preprocessing is performed at the edge, which realizes efficient real-time response to the status of wind turbines, reduces data transmission pressure, and by deploying a dual intelligent model of diagnosis and prediction, the method can not only provide real-time early warning and accurate component location for current faults, but also predict the probability of future faults, realizing the transformation of operation and maintenance strategy from "passive response" to "proactive planning".

[0025] Furthermore, the present invention also has the following technical effects: Efficient and real-time fault response: Utilizing a cloud-edge collaborative architecture, data preprocessing is performed at the edge and intelligent analysis is conducted in the cloud, balancing the needs for low-latency response and high-precision diagnosis.

[0026] Intelligent operation and maintenance that takes into account both the present and the future: By deploying dual models of diagnosis and prediction, it can provide real-time early warning and precise location of sudden failures, as well as trend prediction of the long-term health status of wind turbines, supporting the transformation from passive maintenance to predictive maintenance.

[0027] Automated and standardized handling processes: By automatically matching handling solutions, the fault response process is automated and standardized, reducing human intervention and errors, and greatly improving operation and maintenance efficiency and security.

[0028] Data-driven continuous optimization capability: By introducing a quantitative verification process for treatment effects, a complete feedback loop of "monitoring-analysis-treatment-verification" is constructed, enabling treatment plans and intelligent models to be continuously iterated and optimized based on real effect data, thus possessing self-evolution capabilities. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating the intelligent monitoring and rapid handling method for wind turbine faults in wind farms provided by the present invention.

[0031] Figure 2 This is a schematic diagram of the intelligent fault monitoring and rapid response device for wind turbines in wind farms provided by the present invention.

[0032] Figure label: 10. Data acquisition module; 20. Vector generation module; 30. Information output module; 40. Information matching module; 50. Probability output module; 60. Disposal verification module. Detailed Implementation

[0033] 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, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] The present invention will now be described in detail with reference to the accompanying drawings. The specific operation methods in the method embodiments can also be applied to the device embodiments or system embodiments. In the description of the present invention, unless otherwise stated, "at least one" includes one or more. "Multiple" refers to two or more. For example, at least one of A, B, and C includes: A existing alone, B existing alone, A and B existing simultaneously, A and C existing simultaneously, B and C existing simultaneously, and A, B, and C existing simultaneously. In the present invention, " / " means "or". For example, A / B can mean A or B. "And / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone.

[0035] The present invention will now be described in detail with reference to specific embodiments.

[0036] In some specific embodiments of the present invention, such as Figure 1As shown, this solution provides a method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms, including: Step S100: Obtain multi-dimensional operating status data of the target wind turbine, including at least three dimensions: mechanical, electrical, and environmental.

[0037] Specifically, the above steps ensure the comprehensiveness of the data input. By integrating data from different dimensions, the overall operating status of the wind turbine can be obtained more accurately, providing a high-quality, multi-perspective data foundation for subsequent intelligent analysis and avoiding misjudgments or omissions caused by a single data source.

[0038] Step S200: Preprocess the multi-dimensional operating status data at the edge computing node to generate a high-dimensional feature vector representing the real-time operating status of the target wind turbine.

[0039] Specifically, the above steps, performed at the edge close to the data source, can significantly reduce data transmission bandwidth and latency, enabling near real-time responses to wind turbine status. Simultaneously, it alleviates the computational burden on the cloud data center and improves the overall system efficiency.

[0040] Step S300: Input the high-dimensional feature vector into the first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault location information for the target wind turbine. The first intelligent analysis model is trained based on historical fault data and normal operation data.

[0041] Specifically, the above steps enable real-time diagnosis and precise location: utilizing a powerful model deployed in the cloud and trained on a large amount of historical data, real-time, high-precision early warnings of faults are achieved. More importantly, it can not only determine "whether a fault will occur," but also identify "which component may fail," greatly shortening fault diagnosis time and providing clear guidance for rapid response.

[0042] Step S400: Based on real-time fault warning information and component fault location information, automatically match and generate handling information from the preset handling solution library.

[0043] Specifically, the above steps achieve an automated closed loop from "monitoring" to "decision-making".

[0044] Understandably, this step solidifies operational knowledge into the solution library, enabling rapid and standardized responses to faults, reducing improper handling caused by delays or lack of experience in manual decision-making, and improving the efficiency and quality of operations and maintenance.

[0045] Step S500: Input the high-dimensional feature vector and historical operating data into the second intelligent prediction model deployed in the cloud to output the failure probability of the target wind turbine in the future preset period.

[0046] Specifically, unlike the real-time early warning of the first model, the second model focuses on predicting long-term health trends. This allows the operation and maintenance strategy to shift from "passive response" to "proactive planning."

[0047] In a possible implementation, spare parts procurement and downtime maintenance plans can be arranged in advance based on the high probability of future failures, thereby minimizing unplanned downtime and reducing operation and maintenance costs.

[0048] Step S600: After performing the disposal operation based on the disposal information, automatically collect the operating data of the target wind turbine after restarting and compare it with the preset verification benchmark to generate a quantitative verification result of the disposal effect.

[0049] Specifically, the above steps construct a complete and quantifiable feedback loop.

[0050] Furthermore, not only were the actions taken, but the effectiveness of those actions was also objectively and quantitatively evaluated. This verification result not only confirms whether the fault has been truly resolved, but also allows the data to feed back into the "action plan library" and the two intelligent models, enabling the entire system to possess the ability to learn and continuously evolve.

[0051] In some possible embodiments of the present invention, the step of obtaining multi-dimensional operating status data of the target wind turbine, including at least three dimensions: mechanical, electrical, and environmental, specifically includes: Step S110: Acquire vibration signals, acoustic signals, electrical signals and environmental parameters by means of sensing devices deployed on key components of the target wind turbine. The sensing devices include at least a high-frequency vibration sensor, an acoustic fingerprint sensor, a current and voltage sensor and an environmental sensor.

[0052] Step S120: Based on the collected vibration signals, acoustic signals, electrical signals and environmental parameters, generate multi-dimensional operating status data.

[0053] Specifically, the above steps improve the comprehensiveness and accuracy of fault monitoring by using sensing devices on key components of the target wind turbine. Compared with monitoring methods that rely on a single type of data, the multi-dimensional and multi-modal data acquisition strategy of this invention can construct a more complete and three-dimensional health profile of the wind turbine.

[0054] Vibration and acoustic signals can potentially detect early physical degradation or anomalies in critical mechanical components such as gearboxes and bearings.

[0055] Electrical signals may directly reflect the operating conditions of electrical systems such as generators and converters.

[0056] Environmental parameters (such as temperature and humidity, refer to "Wind Farm Environmental Visibility and Freezing Period Determination" in the knowledge base) may provide key contextual information for analyzing faults (such as icing and overheating).

[0057] Understandably, by integrating this information, the system can distinguish between different types of faults, improve the accuracy of fault identification, and capture related faults that traditional methods may overlook, thereby achieving more comprehensive monitoring.

[0058] The steps include: preprocessing multi-dimensional operational status data at edge computing nodes to generate high-dimensional feature vectors representing the real-time operational status of the target wind turbine; specifically, these steps include: Step S210: Perform wavelet threshold denoising and normalization on the multi-dimensional operating status data, and extract time domain, frequency domain and cepstral features to construct a high-dimensional feature vector.

[0059] Specifically, this embodiment provides an implementation method for generating high-dimensional feature vectors that characterize the real-time operating status of a target wind turbine, thereby enhancing the signal-to-noise ratio and usability of the original data.

[0060] Furthermore, the above steps provide an effective solution to the problem of poor original signal quality when wind turbines operate in harsh environments. In detail: Regarding wavelet threshold denoising: During wind turbine operation, the signals collected by sensors (especially vibration and acoustic signals) inevitably become contaminated with a large amount of environmental noise and irrelevant interference. Denoising processing can effectively filter out noise, highlighting weak feature signals related to the fault, and greatly improving the signal-to-noise ratio. This allows subsequent feature extraction and fault diagnosis models to be analyzed based on "cleaner" data, thereby reducing false alarms and false negatives.

[0061] Regarding normalization: multi-dimensional data from different sensors have different physical units and numerical ranges. Normalization eliminates the influence of data units, mapping all data to a uniform scale. This prevents certain dimensions from dominating during subsequent model training due to excessively large values, ensuring that features in each dimension are evaluated fairly, thereby improving the stability and reliability of the entire diagnostic model.

[0062] Understandably, this invention brings dual benefits by transforming massive amounts of raw time-series data into structured feature vectors with higher information density.

[0063] Regarding reducing data transmission burden: The raw data volume, such as high-frequency vibration data, is enormous; directly uploading it to the cloud would consume significant network bandwidth and increase latency. Performing feature extraction at edge nodes and transmitting only lightweight feature vectors to the upper layer greatly reduces the pressure on the communication network and improves the system's response speed.

[0064] Regarding improving the efficiency and depth of fault diagnosis: directly analyzing raw data places high demands on the model and is inefficient. However, the extracted time-domain, frequency-domain, and cepstral features are validated indicators highly correlated with equipment health. These features, forming a "high-dimensional feature vector," are more easily understood and processed by machine learning algorithms. This strengthens the representational information of the fault, enabling higher-level fault diagnosis and early warning models to identify fault modes faster and more accurately, achieving an effective transformation from data to information.

[0065] In some possible embodiments of the present invention, the step of inputting a high-dimensional feature vector into a first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault location information specifically includes: Step S310: The computing power of the edge computing node is not less than 2 TOPS, the sampling frequency of the vibration signal collected by the edge computing node is greater than or equal to 10kHz, and the dimension of the generated high-dimensional feature vector is 128.

[0066] Step S320: The first intelligent analysis model deployed in the cloud contains a historical fault sample library, which stores high-dimensional feature vector samples corresponding to various core component fault types.

[0067] Specifically, the above steps achieve high-precision and high-timeliness front-end data processing.

[0068] Furthermore, the high sampling frequency (≥10kHz) ensures the ability to capture high-frequency, weak vibration signals generated by core wind turbine components (such as bearings and gears) during early failures, which is the foundation for early warning. Meanwhile, powerful edge computing power (≥2TOPS) enables real-time processing of massive amounts of raw vibration data and the extraction of complex 128-dimensional features. Performing this process at the edge, close to the wind turbine, significantly reduces the amount of data that needs to be uploaded to the cloud, lowers network bandwidth pressure and data transmission latency, and ensures a rapid response in the future.

[0069] Step S330: Output real-time fault warning information. Specifically, compare the values ​​of each dimension of the real-time input 128-dimensional high-dimensional feature vector with the preset "emergency fault" feature threshold and "early fault" feature threshold one by one. When any dimension value exceeds the "emergency fault" threshold, output an "emergency fault" warning. When no dimension exceeds the "emergency fault" threshold but at least one dimension exceeds the "early fault" threshold, output an "early fault" warning.

[0070] Specifically, the above steps enable tiered early warning systems and improve the scientific nature of operation and maintenance decisions.

[0071] Furthermore, this dual-threshold design changes the traditional binary judgment mode of "faulty / no fault". It can clearly distinguish the urgency of the fault, thereby achieving refined hierarchical early warning.

[0072] In a possible implementation, "early failure" warnings provide the operations and maintenance team with ample response time, allowing maintenance work to be incorporated into planned maintenance, thereby optimizing spare parts inventory, rationally allocating manpower, and significantly reducing operations and maintenance costs and power generation losses caused by unplanned outages.

[0073] In a possible implementation, an "emergency failure" warning can trigger an immediate response, preventing catastrophic damage to equipment, ensuring the safety of equipment and personnel, and avoiding greater economic losses.

[0074] Step S340: Output component fault location information. Specifically, after outputting an "emergency fault" or "early fault" warning, extract the feature subset related to the exceeded dimension from the 128-dimensional high-dimensional feature vector, and extract the feature templates of each component from the preset component fault feature template library of core components; calculate the cosine similarity between the feature subset and each feature template, and when a certain similarity calculation result is greater than or equal to a preset value, output the component corresponding to the template with the highest cosine similarity as the final fault location result; the core components include at least the gearbox, generator, and main bearing.

[0075] Specifically, the above steps enable accurate location of faulty components, significantly improving handling efficiency.

[0076] Furthermore, this invention provides an efficient and accurate location method. By comparing only a subset of features related to the abnormal signal, the computational complexity is reduced. Utilizing a cosine similarity matching feature template library, the root cause of the fault can be quickly and accurately located to a specific core component. This allows maintenance personnel to go directly to the fault point, greatly shortening on-site troubleshooting and diagnosis time, achieving "rapid handling," and improving the accuracy and first-time success rate of maintenance.

[0077] Understandably, the above-described capture mechanism of this invention provides a complete intelligent diagnostic process of "data acquisition - feature extraction - hierarchical early warning - precise positioning". It combines the real-time performance of the edge with the powerful analysis and storage capabilities of the cloud, forming a highly efficient cloud-edge collaborative architecture. By continuously accumulating and enriching the historical fault sample library and feature template library in the cloud, the diagnostic capabilities of the model can be continuously optimized and iterated, enabling the entire fault monitoring and handling system to possess self-learning and adaptive capabilities.

[0078] In some possible embodiments of the present invention, the first intelligent analysis model is a hybrid network model that combines a convolutional neural network and a long short-term memory network.

[0079] Specifically, the combination of these two models represents an advanced approach to processing industrial time-series data. CNNs excel at automatically extracting key local and spatial features from raw sensor data (such as vibration and temperature sequences), effectively capturing subtle anomaly patterns in the early stages of a fault. LSTMs, on the other hand, are specifically designed to analyze the dependencies and evolution trends of these features over time. This combination allows the model to not only identify "what" the anomaly is (through CNN features) but also understand "how the anomaly develops" (through LSTM time-series analysis), thereby generating a highly condensed "high-dimensional feature vector" that lays a solid foundation for subsequent accurate judgment. Compared to traditional statistical methods or single models, this significantly improves the depth and accuracy of fault identification.

[0080] The steps for outputting real-time fault warning information for the target wind turbine specifically include: Step S410: Compare the high-dimensional feature vector with the preset multi-level feature threshold to determine the current state of the target wind turbine. The current state is one of "emergency fault", "early fault" or "normal operation".

[0081] Specifically, this hierarchical mechanism described above is key to achieving predictable operations and maintenance. It changes the traditional binary approach of "faulty / fault-free".

[0082] Furthermore, for "early failures," the system can issue early warnings, giving the maintenance team ample time to plan maintenance schedules and prepare spare parts, enabling maintenance to be carried out without affecting the power generation schedule, effectively reducing the number of unplanned outages and maintenance costs.

[0083] In the event of an "emergency failure," the system will trigger the highest level of alarm, requiring immediate action (such as shutdown) to prevent catastrophic damage to the equipment and ensure the fundamental safety of the equipment and the site.

[0084] The steps for outputting real-time fault warning information and component fault location information for the target wind turbine specifically include: Step S420: When the status is "emergency fault" or "early fault", calculate the cosine similarity between the fault-related feature subset in the high-dimensional feature vector and the historical component fault feature template, and determine the component with a similarity higher than the preset location threshold as the fault location result.

[0085] Specifically, by comparing the above steps with historical fault templates (e.g., "data fingerprints" of specific faults such as gearbox bearing wear and generator coil overheating), the system can identify which specific component has a problem with a very high probability.

[0086] Furthermore, by using cosine similarity as a metric, the focus is on the similarity of fault feature patterns rather than the magnitude of absolute values, which enables the method to have good identification capabilities for similar faults of different severity levels.

[0087] Furthermore, this feature significantly reduces on-site fault diagnosis time. Maintenance personnel no longer need to conduct extensive troubleshooting; they can directly inspect and repair the identified components, significantly improving maintenance efficiency, reducing downtime, and ultimately enhancing the wind farm's total lifecycle return on investment.

[0088] Understandably, the above-mentioned solution of the present invention achieves an intelligent leap from "fuzzy early warning" to "hierarchical and precise positioning", which greatly improves the accuracy and operability of wind turbine fault early warning.

[0089] In some possible embodiments of the present invention, it further includes: Step S700: Construct a full traceability database covering the entire life cycle of a wind farm. The full traceability database includes a first wind turbine file corresponding to each wind turbine and a second wind turbine file corresponding to each wind turbine failure.

[0090] Specifically, the above scheme can be broken down as follows: First, database construction: A "full traceability database" needs to be established. This database has a clear structure and contains two core archives: First wind turbine file: Corresponding one-to-one with each physical wind turbine, recording its static information and long-term dynamic health status.

[0091] Second wind turbine file: Corresponding to each wind turbine failure event, recording the details of the failure.

[0092] Secondly, the timing of data archiving: Data archiving is performed automatically after the "results of the effectiveness verification" are generated, which forms a complete closed loop from "early warning - action - verification - archiving".

[0093] Then, the comprehensiveness of the archived data: The archived data must be "full-process data," extremely detailed, covering every stage of the failure event. For example, the source of the early warning: the original multi-dimensional operational status data that triggers the early warning.

[0094] For example, the model analysis process includes: the generated high-dimensional feature vectors, the early warning and positioning results of the intelligent model, and historical prediction records.

[0095] For example, decision-making and execution: matching disposal information, personnel operation records at each stage, and on-site photos.

[0096] For example, effectiveness evaluation: quantitative comparison data and final verification results for verifying the effectiveness of the treatment.

[0097] Understandably, the above steps, by building a full traceability database and realizing automatic archiving of data throughout the entire process, have brought about significant technical effects, elevating wind turbine fault management from a passive "response and handling" level to a proactive "learning and optimization" level.

[0098] Step S800: After the verification results of the handling effect are generated, the full-process data related to this fault event will be automatically archived into the full-scale traceability database. The full-process data includes: the original multi-dimensional operation status data that triggered the warning, the generated high-dimensional feature vector, the warning and location results of the first intelligent analysis model, the historical prediction records of the second intelligent prediction model, the matched handling information, the personnel operation records and on-site photos of each handling link, as well as the quantitative comparison data and final verification results of the handling effect verification.

[0099] Specifically, the archived end-to-end data, especially the complete chain of "raw data -> model prediction -> handling plan -> final result," provides extremely valuable supervised learning samples for machine learning and artificial intelligence models. By learning from massive amounts of historical fault cases, the first intelligent analysis model (early warning and location model) and the second intelligent prediction model can be continuously optimized, improving the accuracy of early warnings and the precision of fault location.

[0100] Furthermore, by archiving each failure's data into the corresponding "first wind turbine file," a full lifecycle health profile for each wind turbine can be created. This not only helps predict future failure trends but also provides precise data support for wind turbine upgrades, overhauls, spare parts strategies, and even decommissioning decisions, thereby maximizing the asset's full lifecycle value.

[0101] In some possible embodiments of the present invention, the step of automatically matching and generating handling information from a preset handling solution library based on real-time fault warning information and component fault location information specifically includes: Step S430: Based on the fault level of the real-time fault warning information and the fault location information of the component, search in the handling plan library containing at least N standardized handling plans, where N is a positive integer greater than or equal to 2.

[0102] Specifically, compared to a vague fault description, this method can more accurately pinpoint the root cause of the problem and quickly filter out directly relevant and applicable solutions from a large solution library containing N solutions. This avoids the errors and delays that may be caused by manual judgment, and improves the accuracy and efficiency of solution matching.

[0103] Step S440: Based on the preset optimization principles of minimizing processing time and minimizing processing cost, the optimal processing solution is matched from the retrieved applicable solutions.

[0104] Specifically, the above steps go beyond simply finding a "usable" solution; they aim to find an "optimal" one. In wind farm operation, downtime (handling time) and maintenance costs (handling expenses) are key economic indicators. The effectiveness of this technology lies in its ability to automatically weigh and decide between economic efficiency and timeliness, bringing direct economic benefits to wind farm operation and achieving intelligent and optimized handling solutions.

[0105] Step S450 generates structured handling information including fault details, fault cause analysis, optimal handling solution operation steps, required spare parts list and safety precautions, and pushes it to the terminals of operation and maintenance personnel with different permissions in a hierarchical manner through the IoT gateway according to the fault level.

[0106] Specifically, this structured information package provides maintenance personnel with a "one-stop" guide. It not only tells "what to do" (operational steps), but also explains "why" (cause analysis), "what to use" (spare parts list), and "what to pay attention to" (safety precautions). This significantly reduces reliance on the personal experience of maintenance personnel, minimizes secondary troubleshooting or operational errors caused by incomplete information, and improves the success rate of single fault repairs and the standardization and safety of on-site operations.

[0107] Furthermore, through the IoT gateway, the system pushes information to "operation and maintenance personnel with different permissions" in a tiered manner based on the "fault level," thus establishing an intelligent tiered response mechanism.

[0108] In a possible implementation, for low-level faults, information may only be pushed to junior on-site maintenance personnel, avoiding information overload for management personnel.

[0109] In a possible implementation, for high-level, complex faults, information can be simultaneously pushed to senior engineers on site, spare parts warehouse managers, and remote technical experts, enabling rapid multi-party collaboration.

[0110] This invention ensures that the most suitable person receives the most needed information at the first time through a hierarchical push mechanism, avoiding information mismatch and response delay, and greatly optimizing the management process and overall response speed of the operation and maintenance team.

[0111] Understandably, when a wind turbine malfunctions, it is possible to automatically, accurately, and economically generate a complete solution that matches the personnel's permissions and efficiently deliver it to the most suitable personnel, thereby greatly improving the efficiency of fault handling, reducing operating costs, and ensuring operational safety.

[0112] In some possible embodiments of the present invention, the step of generating quantitative verification results of treatment effects specifically includes: Step S450: After the disposal operation is completed and the target wind turbine is restarted, continuously collect operating data for at least the first preset duration; Step S460: Construct a two-layer verification system by comparing the collected operational data with the verification benchmark.

[0113] Specifically, the above steps involve constructing a "dual-layer verification system" to combine the macroscopic overall machine operating status with the microscopic performance indicators of faulty components for judgment.

[0114] Furthermore, traditional verification methods may rely solely on experience or observation of power recovery, which is highly one-sided. For example, after repairing a bearing failure, the wind turbine may temporarily resume power generation, but the abnormal vibration of the bearing may persist, quickly leading to secondary damage. This solution ensures the wind turbine "can operate" through a first layer (basic parameters) and ensures the root cause of the failure is completely eliminated through a second layer (specific indicators). Only when both are met simultaneously can the problem be considered "successfully resolved," greatly eliminating the possibility of false repairs and making the verification conclusions highly reliable.

[0115] Step S461, the two-layer verification system includes: First-level basic parameter verification: Compare the basic operating parameters after the fan restart with the standard operating parameter range of the fan under the current operating conditions to determine whether there are abnormal fluctuations. The basic operating parameters include at least speed, power and temperature. The second layer of specific indicator verification: compare the specific performance indicators of the faulty component after restarting with the historical normal values ​​of the faulty component before the fault occurred to determine whether the specific indicators have returned to the normal range. The specific performance indicators include at least the vibration characteristic frequency and the electrical harmonic content. Step S470: If both the first-level basic parameter verification and the second-level special indicator verification pass, a verification result of "successful disposal" is generated.

[0116] Specifically, the above steps completely change the previous model that relied on manual experience and subjective judgment of "whether it is repaired". It transforms the verification process into a data-driven, benchmark-based, repeatable, and standardized procedure. Regardless of the technician, the wind farm, or the turbine, the same set of quantitative standards is used for evaluation, ensuring the consistency and objectivity of the verification work and providing a unified quality benchmark for the operation and maintenance management of wind farms.

[0117] Furthermore, this "whole-part" combined verification approach is very comprehensive. It not only confirms that the fault point itself has been repaired, but also confirms that the repair operation did not cause any new negative impact on the overall performance of the wind turbine.

[0118] Understandably, an objective, quantitative, comprehensive and highly reliable closed-loop verification mechanism for wind turbine fault handling effect has been constructed, which significantly improves the quality and efficiency of fault handling and effectively avoids secondary faults caused by insufficient verification.

[0119] In some possible embodiments of the present invention, it further includes: In step S900, when the first intelligent analysis model generates an early warning of "emergency fault" or "early fault", the generation and hierarchical push of handling information are completed. The "emergency fault" information is pushed to the on-site operation and maintenance personnel, team leaders and wind farm managers at the same time, and the "early fault" information is pushed to the on-site operation and maintenance personnel and team leaders.

[0120] Specifically, by distinguishing between "urgent faults" and "early-stage faults," the system can achieve tiered information delivery for handling. Information of high urgency will be notified to management (wind farm managers) to ensure that management can grasp major risks in a timely manner; while routine early warnings will only be pushed to the execution level (operation and maintenance personnel, team leaders) to avoid information interference to management.

[0121] In step S1000, maintenance personnel receive the handling information through a mobile terminal and scan or click to confirm the core progress nodes in the handling process in real time. The background automatically records the timestamp and geographical location information of each node. The core progress nodes include at least "instruction received", "on-site positioning", "spare parts replacement", "handling completed" and "application for trial operation".

[0122] Specifically, by using mobile terminals, QR code scanning / click confirmation, and automatic recording of timestamps and geolocation, offline operation and maintenance workflows are digitized and standardized. Managers can monitor the completion status of key progress nodes such as "instruction reception" and "on-site deployment" in real time and objectively.

[0123] The steps for real-time scanning or clicking to confirm key progress nodes during the processing include: Step S1010: Each standardized disposal solution in the disposal solution library has an embedded QR code. When performing real-time scanning or clicking confirmation, the operation and maintenance personnel need to scan the corresponding QR code. The mobile terminal will automatically display the graphic or video operation guide for this step and record the operation completion status. In step S1020, after the maintenance personnel submit the "application for trial operation", the system remotely unlocks and starts the wind turbine, automatically executing the two-layer verification system; Step S1030: If the verification fails, a secondary diagnostic report is automatically generated to analyze the cause and recommend supplementary treatment measures. Step S1040: If the verification is successful, a complete acceptance report is automatically generated, including all process node records, operation photos, call recordings, and quantitative verification reports. The report is then pushed to the wind farm manager for electronic approval, and all data is packaged, encrypted, and stored in the full traceability database.

[0124] Specifically, after the maintenance personnel complete the handling and "apply for trial operation," the system takes over the subsequent verification work. It can automatically execute a "two-layer verification system" to objectively evaluate the maintenance effect. If the verification fails, the system can also automatically generate a secondary diagnostic report and recommend supplementary measures.

[0125] Furthermore, the system can automatically aggregate all data during the disposal process—including process node records, on-site photos, call recordings, quantitative verification reports, etc.—to generate a comprehensive and tamper-proof electronic acceptance report, which is then encrypted and stored in the "full traceability database."

[0126] In some possible embodiments of the present invention, the second intelligent prediction model is a gradient boosting tree model; The steps for outputting the failure probability of the target wind turbine within a preset future period specifically include: Step S510: Integrate historical operating data, fault data, maintenance records, and meteorological data of the target wind turbine and other wind turbines at the same site to construct a predictive feature set; Step S520: Analyze the predicted feature set using the gradient boosting tree model, calculate and output the probability values ​​of various preset faults occurring in the target wind turbine within the second preset time period.

[0127] Specifically, by incorporating data from other wind turbines at the same site, the model no longer analyzes a single turbine in isolation. It can learn common issues across the entire wind farm, such as the prevalent risk of blade icing in specific seasons, the aging patterns of similar batches of components, and the combined effects of regional extreme weather. This allows predictions to transcend the limitations of a single turbine, providing a more comprehensive perspective and more accurate assessments.

[0128] The method also includes: Step S1100: Through the incremental learning mechanism, newly added operation and fault data are automatically included in the training set at a preset training period to retrain the gradient boosting tree model in order to dynamically optimize the prediction accuracy of the gradient boosting tree model.

[0129] Specifically, this embodiment provides an implementation method for outputting the failure probability of a target wind turbine within a preset future period. By introducing a series of specific and interconnected technical features, it achieves significant technical improvements and beneficial effects. Its core contribution lies in transforming the broad concept of "intelligent prediction" into a concrete, efficient, and self-evolving engineering solution.

[0130] In some specific embodiments of the present invention, such as Figure 2 As shown, this solution provides an intelligent monitoring and rapid fault handling device for wind turbines in wind farms, including: Data acquisition module 10 is used to acquire multi-dimensional operating status data of the target wind turbine, including at least three dimensions: mechanical, electrical, and environmental. The vector generation module 20 is used to preprocess multi-dimensional operating status data at the edge computing node to generate high-dimensional feature vectors that characterize the real-time operating status of the target wind turbine. The information output module 30 is used to input high-dimensional feature vectors into the first intelligent analysis model deployed in the cloud, so as to output real-time fault warning information and component fault location information for the target wind turbine. The first intelligent analysis model is trained based on historical fault data and normal operation data. The information matching module 40 is used to automatically match and generate handling information from a preset handling solution library based on real-time fault warning information and component fault location information. The probability output module 50 is used to input high-dimensional feature vectors and historical operating data into the second intelligent prediction model deployed in the cloud, so as to output the failure probability of the target wind turbine in the future preset period. The treatment verification module 60 is used to automatically collect the operating data of the target wind turbine after it is restarted after the treatment operation is performed based on the treatment information, and compare it with the preset verification benchmark to generate a quantitative treatment effect verification result.

[0131] Optionally, the steps for obtaining multi-dimensional operational status data of the target wind turbine, including at least three dimensions: mechanical, electrical, and environmental, specifically include: Vibration signals, acoustic signals, electrical signals and environmental parameters are acquired by sensing devices deployed on key components of the target wind turbine. The sensing devices include at least high-frequency vibration sensors, acoustic fingerprint sensors, current and voltage sensors and environmental sensors. Based on the collected vibration signals, acoustic signals, electrical signals and environmental parameters, multi-dimensional operating status data is generated; The steps include: preprocessing multi-dimensional operational status data at edge computing nodes to generate high-dimensional feature vectors representing the real-time operational status of the target wind turbine; specifically, these steps include: Wavelet thresholding and normalization are performed on multi-dimensional operational status data, and time-domain, frequency-domain, and cepstral features are extracted to construct high-dimensional feature vectors.

[0132] Specifically, this embodiment provides an implementation method for acquiring multi-dimensional operational status data of a target wind turbine, including at least three dimensions: mechanical, electrical, and environmental. It constructs an intelligent sensing front-end that ensures data quality from the source and performs efficient processing at the edge. This not only ensures the comprehensiveness and accuracy of the input data but also significantly improves data usability and information density through noise reduction, normalization, and feature extraction, while simultaneously optimizing the system's data transmission efficiency.

[0133] Optionally, the step of inputting the high-dimensional feature vector into the first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault location information specifically includes: The computing power of the edge computing node is no less than 2 TOPS, the sampling frequency of the vibration signal collected by the edge computing node is greater than or equal to 10kHz, and the dimension of the generated high-dimensional feature vector is 128. The first intelligent analysis model deployed in the cloud contains a historical fault sample library, which stores high-dimensional feature vector samples corresponding to various core component fault types. The system outputs real-time fault warning information by comparing the values ​​of each dimension of the real-time input 128-dimensional high-dimensional feature vector with the preset "emergency fault" feature threshold and "early fault" feature threshold one by one. When any dimension value exceeds the "emergency fault" threshold, an "emergency fault" warning is output. When no dimension exceeds the "emergency fault" threshold but at least one dimension exceeds the "early fault" threshold, an "early fault" warning is output. The output component fault location information is as follows: after outputting an "emergency fault" or "early fault" warning, a feature subset related to the exceeded dimension is extracted from the 128-dimensional high-dimensional feature vector, and feature templates of each component are extracted from the preset component fault feature template library of core components; by calculating the cosine similarity between the feature subset and each feature template, when a certain similarity calculation result is greater than or equal to a preset value, the component corresponding to the template with the highest cosine similarity is output as the final fault location result; the core components include at least the gearbox, generator, and main bearing.

[0134] Specifically, this embodiment provides an implementation method that inputs high-dimensional feature vectors into a first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault location information. By specifying key aspects such as computing power, sampling rate, feature dimension, warning logic, and location algorithm in detail, the combined effect of these factors ultimately achieves earlier and more accurate detection of wind turbine faults, more scientific and hierarchical early warning, and faster and more accurate location and handling, thereby comprehensively improving the intelligence level, safety, and economic benefits of wind farm operation and maintenance.

[0135] Optionally, the first intelligent analysis model is a hybrid network model combining a convolutional neural network and a long short-term memory network; The steps for outputting real-time fault warning information for the target wind turbine specifically include: The high-dimensional feature vector is compared with a preset multi-level feature threshold to determine the current state of the target wind turbine, which is one of "emergency failure", "early failure" or "normal operation". The steps for outputting real-time fault warning information and component fault location information for the target wind turbine specifically include: When the status is "emergency fault" or "early fault", the cosine similarity between the fault-related feature subset in the high-dimensional feature vector and the historical component fault feature template is calculated, and the component with a similarity higher than the preset location threshold is identified as the fault location result.

[0136] Specifically, this embodiment provides an implementation method for outputting real-time fault early warning information for target wind turbines. By introducing three specific technical means—a CNN-LSTM hybrid model, multi-level state threshold judgment, and fault location based on cosine similarity—the protected method possesses significant technical advantages in terms of high precision, hierarchical classification, and accurate positioning. It constructs a complete closed loop from data acquisition to intelligent analysis, hierarchical early warning, and precise positioning, representing the core technology for realizing intelligent, refined, and predictive operation and maintenance of wind farms, and possesses extremely high practical value and economic benefits.

[0137] Optionally, it also includes: Construct a full traceability database covering the entire life cycle of a wind farm. The full traceability database includes a first wind turbine file corresponding to each wind turbine and a second wind turbine file corresponding to each wind turbine failure. After the results of the handling effect verification are generated, all process data related to this fault event will be automatically archived into the full traceability database. The full process data includes: the original multi-dimensional operation status data that triggered the warning, the generated high-dimensional feature vector, the warning and location results of the first intelligent analysis model, the historical prediction records of the second intelligent prediction model, the matched handling information, the personnel operation records and on-site photos of each handling link, as well as the quantitative comparison data and final verification results of the handling effect verification.

[0138] Specifically, this embodiment provides an implementation method for constructing a comprehensive traceability database covering the entire lifecycle of a wind farm. By building a dynamic learning and self-optimizing data closed-loop system, each fault handling process is transformed into valuable knowledge assets. This not only greatly enhances the performance and accuracy of the intelligent early warning model, but also provides a solid data foundation for realizing full lifecycle health management of wind turbines, optimizing operation and maintenance processes, and accumulating expert experience. It is a key step in realizing the transformation of wind farms from "digitalization" to "intelligence" and "smartization."

[0139] Optionally, the step of automatically matching and generating handling information from a preset handling solution library based on real-time fault warning information and component fault location information specifically includes: Based on the fault level of real-time fault warning information and the fault location information of the component, the faulty component is searched in the handling plan library containing at least N standardized handling plans, where N is a positive integer greater than or equal to 2. Based on the preset optimization principles of minimizing processing time and reducing processing cost, the optimal processing solution is matched from the available solutions. The system generates structured handling information that includes fault details, fault cause analysis, optimal handling solution operation steps, required spare parts list, and safety precautions. This information is then pushed to the terminals of maintenance personnel with different permissions through an IoT gateway, based on the fault level.

[0140] Specifically, this embodiment provides an implementation method that automatically matches and generates disposal information from a preset disposal solution library. By introducing multi-dimensional retrieval, economically optimized decision-making, structured information output, and hierarchical information distribution, it solves the technical problems in traditional fault handling, such as slow response, decision-making relying on human experience, uncontrollable costs, and low information transmission efficiency. Ultimately, it achieves the core goal of "fast, accurate, economical, and safe" fault handling in wind farms.

[0141] Optionally, the step of generating quantitative verification results of the treatment effect specifically includes: After the handling operation is completed and the target wind turbine is restarted, continuous collection of operational data will continue for at least the first preset duration; A two-layer verification system is constructed to compare the collected operational data with the verification benchmark; The two-layer verification system includes: First-level basic parameter verification: Compare the basic operating parameters after the fan restart with the standard operating parameter range of the fan under the current operating conditions to determine whether there are abnormal fluctuations. The basic operating parameters include at least speed, power and temperature. The second layer of specific indicator verification: compare the specific performance indicators of the faulty component after restarting with the historical normal values ​​of the faulty component before the fault occurred to determine whether the specific indicators have returned to the normal range. The specific performance indicators include at least the vibration characteristic frequency and the electrical harmonic content. If both the first-level basic parameter verification and the second-level special indicator verification pass, a verification result of "successful disposal" is generated.

[0142] Specifically, this embodiment provides an implementation method for generating quantitative verification results of the treatment effect. Through a carefully designed, data-driven, two-layer verification model, it solves the core pain point of "how to confirm that the repair is really done" in fault treatment. It transforms the traditional, vague, and subjective post-repair evaluation into a precise, efficient, objective, and comprehensive automated verification process, bringing a qualitative improvement to the intelligence and reliability of the entire wind farm operation and maintenance system.

[0143] Optionally, it also includes: When the first intelligent analysis model generates an early warning of "emergency fault" or "early fault", it completes the generation and hierarchical push of handling information. The "emergency fault" information is pushed to the on-site operation and maintenance personnel, team leaders and wind farm managers at the same time, while the "early fault" information is pushed to the on-site operation and maintenance personnel and team leaders. Maintenance personnel receive handling information via mobile terminals and scan or click to confirm key progress nodes in the handling process in real time. The background automatically records the timestamp and geographical location information of each node. Key progress nodes include at least "instruction received", "on-site positioning", "spare parts replacement", "handling completed" and "application for trial operation". The steps for real-time scanning or clicking to confirm key progress nodes during the processing include: Each standardized handling solution in the handling solution library has a QR code embedded. When performing real-time scanning or clicking confirmation, the operation and maintenance personnel need to scan the corresponding QR code. The mobile terminal will automatically display the graphic or video operation guide for that step and record the operation completion status. After the maintenance personnel submit an "application for trial operation", the system remotely unlocks and starts the wind turbine, automatically executing a two-layer verification system; If the verification fails, a secondary diagnostic report will be automatically generated, analyzing the cause and recommending supplementary treatment measures; If the verification is successful, a complete acceptance report is automatically generated, including all process node records, operation photos, call recordings, and a quantitative verification report. This report is then pushed to the wind farm manager for electronic approval, and all data is packaged, encrypted, and stored in the full traceability database.

[0144] Specifically, this embodiment provides a method for generating and hierarchically pushing disposal information, as well as for real-time scanning or clicking to confirm key progress nodes in the disposal process. This solves the core pain points in traditional wind power operation and maintenance, such as poor information transmission, lack of transparency, inconsistent standards, difficulty in evaluating effects, and difficulty in tracing data. Ultimately, it achieves comprehensive technical effects such as improving operation and maintenance efficiency, ensuring disposal quality, reducing management costs, and laying a solid foundation for long-term data-driven optimization.

[0145] Optionally, the second intelligent prediction model is a gradient boosting tree model; The steps for outputting the failure probability of the target wind turbine within a preset future period specifically include: Integrate historical operating data, fault data, maintenance records, and meteorological data of the target wind turbine and other wind turbines at the same site to construct a predictive feature set; The gradient boosting tree model is used to analyze the predicted feature set, calculate and output the probability values ​​of various preset faults occurring in the target wind turbine within the second preset time period; The method also includes: Through an incremental learning mechanism, newly added operational and fault data are automatically incorporated into the training set at a preset training cycle to retrain the gradient boosting tree model, thereby dynamically optimizing the prediction accuracy of the gradient boosting tree model.

[0146] Specifically, this embodiment provides an implementation method for outputting the failure probability of a target wind turbine within a preset future period. It not only solves the problem of "whether it can be predicted", but also effectively solves a series of key engineering problems such as "how to accurately predict, how to effectively utilize the prediction results, and how to ensure that the prediction capability never becomes outdated". This provides strong technical support for realizing truly intelligent and refined operation and maintenance management of wind farms.

[0147] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0148] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent monitoring and rapid fault handling of wind turbines in wind farms, characterized in that, include: Acquire multi-dimensional operational status data of the target wind turbine, including at least three dimensions: mechanical, electrical, and environmental. The multi-dimensional operating status data is preprocessed at the edge computing node to generate a high-dimensional feature vector characterizing the real-time operating status of the target wind turbine; The high-dimensional feature vector is input into the first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault location information for the target wind turbine. The first intelligent analysis model is trained based on historical fault data and normal operation data. Based on the real-time fault warning information and the component fault location information, the system automatically matches and generates handling information from the preset handling solution library; The high-dimensional feature vector and historical operating data are input into the second intelligent prediction model deployed in the cloud to output the failure probability of the target wind turbine in a future preset period; After performing the disposal operation based on the disposal information, the operating data of the target wind turbine after restarting is automatically collected and compared with the preset verification benchmark to generate a quantitative verification result of the disposal effect.

2. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to claim 1, characterized in that, The steps for obtaining multi-dimensional operational status data of the target wind turbine, including at least mechanical, electrical, and environmental dimensions, specifically include: Vibration signals, acoustic signals, electrical signals and environmental parameters are acquired by sensing devices deployed on key components of the target wind turbine, wherein the sensing devices include at least a high-frequency vibration sensor, an acoustic fingerprint sensor, a current and voltage sensor and an environmental sensor. Based on the collected vibration signals, acoustic signals, electrical signals, and environmental parameters, the multi-dimensional operating status data is generated. The step of preprocessing the multi-dimensional operating status data at the edge computing node to generate a high-dimensional feature vector representing the real-time operating status of the target wind turbine specifically includes: Wavelet threshold denoising and normalization are performed on the multi-dimensional operating status data, and time-domain, frequency-domain and cepstral features are extracted to construct the high-dimensional feature vector.

3. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to claim 2, characterized in that, The step of inputting the high-dimensional feature vector into the first intelligent analysis model deployed in the cloud to output real-time fault warning information and component fault location information specifically includes: The computing power of the edge computing node is not less than 2 TOPS, the sampling frequency of the vibration signal collected by the edge computing node is greater than or equal to 10kHz, and the dimension of the generated high-dimensional feature vector is 128. The first intelligent analysis model deployed in the cloud contains a historical fault sample library, which stores high-dimensional feature vector samples corresponding to various core component fault types. The real-time fault warning information is output by comparing the values ​​of each dimension of the 128-dimensional high-dimensional feature vector that is input in real time with the preset "emergency fault" feature threshold and "early fault" feature threshold one by one. When any dimension value exceeds the "emergency fault" threshold, an "emergency fault" warning is output; when no dimension exceeds the "emergency fault" threshold but at least one dimension exceeds the "early fault" threshold, an "early fault" warning is output. The output component fault location information specifically involves: after outputting an "emergency fault" or "early fault" warning, extracting a subset of features related to the exceeded dimension from the 128-dimensional high-dimensional feature vector, and extracting feature templates for each component from a preset core component component fault feature template library; calculating the cosine similarity between the feature subset and each feature template, and when a certain similarity calculation result is greater than or equal to a preset value, outputting the component corresponding to the template with the highest cosine similarity as the final fault location result; the core components include at least a gearbox, a generator, and a main bearing.

4. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to claim 1, characterized in that, The first intelligent analysis model is a hybrid network model that combines a convolutional neural network and a long short-term memory network; The step of outputting real-time fault warning information for the target wind turbine specifically includes: The high-dimensional feature vector is compared with a preset multi-level feature threshold to determine the current state of the target wind turbine, which is one of "emergency failure", "early failure" or "normal operation". The step of outputting real-time fault warning information and component fault location information for the target wind turbine specifically includes: When the status is "emergency fault" or "early fault", the cosine similarity between the fault-related feature subset in the high-dimensional feature vector and the historical component fault feature template is calculated, and the component with a similarity higher than the preset location threshold is identified as the fault location result.

5. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to any one of claims 2 to 4, characterized in that, Also includes: Construct a full traceability database covering the entire life cycle of a wind farm, wherein the full traceability database includes a first wind turbine file corresponding one-to-one with the wind turbine, and a second wind turbine file corresponding one-to-one with wind turbine failures; After the verification results of the handling effect are generated, the full-process data related to this fault event will be automatically archived into the full-scale traceability database. The full-process data includes: the original multi-dimensional operation status data that triggered the warning, the generated high-dimensional feature vector, the warning and location results of the first intelligent analysis model, the historical prediction records of the second intelligent prediction model, the matched handling information, the personnel operation records and on-site photos of each handling link, and the quantitative comparison data and final verification results of the handling effect verification.

6. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to claim 5, characterized in that, The step of automatically matching and generating handling information from a preset handling solution library based on the real-time fault warning information and component fault location information specifically includes: Based on the fault level of the real-time fault warning information and the faulty component of the component fault location information, a search is performed in a solution library containing at least N standardized solution solutions, where N is a positive integer greater than or equal to 2. Based on the preset optimization principles of minimizing processing time and reducing processing cost, the optimal processing solution is matched from the available solutions. The system generates structured handling information that includes fault details, fault cause analysis, optimal handling solution operation steps, required spare parts list, and safety precautions. This information is then pushed to the terminals of maintenance personnel with different permissions through an IoT gateway, based on the fault level.

7. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to claim 6, characterized in that, The step of generating quantitative verification results of the treatment effect specifically includes: After the handling operation is completed and the target wind turbine is restarted, continuous collection of operational data will continue for at least the first preset duration; A two-layer verification system is constructed to compare the collected operational data with the verification benchmark; The two-layer verification system includes: First-level basic parameter verification: The basic operating parameters of the fan after restarting are compared with the standard operating parameter range of the fan under the current operating conditions to determine whether there are abnormal fluctuations. The basic operating parameters include at least speed, power and temperature. Second-level specific indicator verification: The specific performance indicators after the faulty component is restarted are compared with the historical normal values ​​of the faulty component before the fault occurred to determine whether the specific indicators have recovered to the normal range. The specific performance indicators include at least the vibration characteristic frequency and the electrical harmonic content. If both the first-level basic parameter verification and the second-level special indicator verification pass, a verification result of "successful handling" is generated.

8. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to claim 7, characterized in that, Also includes: When the first intelligent analysis model generates an "emergency fault" or "early fault" warning, the generation and hierarchical push of handling information are completed. The "emergency fault" information is pushed to the on-site operation and maintenance personnel, team leaders and wind farm managers at the same time, and the "early fault" information is pushed to the on-site operation and maintenance personnel and team leaders. The maintenance personnel receive the handling information through a mobile terminal and scan or click to confirm the core progress nodes in the handling process in real time, recording the timestamp and geographical location information of each node. The core progress nodes include at least "instruction received", "on-site in place", "spare parts replacement", "handling completed" and "application for trial operation". The steps for real-time scanning or clicking to confirm key progress nodes in the processing process specifically include: Each standardized handling solution in the handling solution library has a QR code embedded. When the operation and maintenance personnel perform the real-time scanning or click to confirm, they need to scan the corresponding QR code. The mobile terminal will automatically display the graphic or video operation guide for this step and record the operation completion status. After the maintenance personnel submit an "application for trial operation", the system remotely unlocks and starts the wind turbine, automatically executing the two-layer verification system; If the verification fails, a secondary diagnostic report will be automatically generated, analyzing the cause and recommending supplementary treatment measures; If the verification is successful, a complete acceptance report is automatically generated, including all process node records, operation photos, call recordings, and a quantitative verification report. This report is then pushed to the wind farm manager for electronic approval, and all data is packaged, encrypted, and stored in the full traceability database.

9. The method for intelligent monitoring and rapid handling of faults in wind turbines of wind farms according to any one of claims 1 to 4, characterized in that, The second intelligent prediction model is a gradient boosting tree model; The step of outputting the failure probability of the target wind turbine within a future preset period specifically includes: By integrating historical operating data, fault data, maintenance records, and meteorological data of the target wind turbine and other wind turbines at the same site, a predictive feature set is constructed. The gradient boosting tree model is used to analyze the predicted feature set, calculate and output the probability values ​​of various preset faults occurring in the target wind turbine within a second preset time period; The method further includes: Through an incremental learning mechanism, newly added operational and fault data are automatically included in the training set at a preset training cycle to retrain the gradient boosting tree model, thereby dynamically optimizing the prediction accuracy of the gradient boosting tree model.

10. A fault intelligent monitoring and rapid handling device for wind turbines in wind farms, characterized in that, include: The data acquisition module is used to acquire multi-dimensional operating status data of the target wind turbine, including at least three dimensions: mechanical, electrical, and environmental. The vector generation module is used to preprocess the multi-dimensional operating status data at the edge computing node to generate a high-dimensional feature vector representing the real-time operating status of the target wind turbine. The information output module is used to input the high-dimensional feature vector into the first intelligent analysis model deployed in the cloud, so as to output real-time fault warning information and component fault location information for the target wind turbine. The first intelligent analysis model is trained based on historical fault data and normal operation data. The information matching module is used to automatically match and generate handling information from a preset handling solution library based on the real-time fault warning information and component fault location information; The probability output module is used to input the high-dimensional feature vector and historical operating data into the second intelligent prediction model deployed in the cloud, so as to output the failure probability of the target wind turbine in the future preset period. The treatment verification module is used to automatically collect the operating data of the target wind turbine after restarting after performing treatment operations based on the treatment information, and compare it with the preset verification benchmark to generate a quantitative treatment effect verification result.

Images