A wind turbine fault prediction system based on cloud computing

Abstract

The present application relates to the technical field of fault prediction, in particular to a wind turbine fault prediction system based on cloud computing, which comprises six modules of data management module, feature engineering module, fusion modeling module, cloud-edge collaboration module, online learning module and environment adaptation module. In use, the six modules cooperate to build a closed-loop linkage mechanism of forward execution and reverse feedback, adapt to offshore, high-cold, mountainous multi-extreme scenarios and wind turbine full life cycle operation requirements, and each module has its own function and cooperates and optimizes. The data management module provides high-quality standardized data, the feature engineering module extracts strong correlation features of faults, the fusion modeling module trains high-precision prediction models, the cloud-edge collaboration module realizes low-latency deployment and reasoning, the online learning module ensures dynamic iteration of the model, and the environment adaptation module maintains prediction accuracy in extreme environments, solving the problems of poor adaptability, low precision and slow iteration of traditional fault prediction systems.

Description

Technical Field

[0001] This invention relates to the field of fault prediction technology, and specifically to a cloud computing-based wind turbine fault prediction system. Background Technology

[0002] With the rapid development of the wind power industry, the deployment scale of wind turbine units is constantly expanding, and more and more units are being used in extreme field scenarios such as offshore, high-altitude cold, and mountainous areas. As complex large rotating machines, wind turbine units operate under high load and harsh environments for a long time. Their core components, such as gearboxes, generators, and bearings, are prone to wear, fatigue, and failure. If failures cannot be predicted in time and preventive maintenance is not carried out, it will lead to unit shutdown, power generation loss, and even safety accidents, significantly increasing maintenance costs.

[0003] Currently, wind turbine fault prediction technology has become the core of smart wind power operation and maintenance. However, it lacks systematic collaborative functions and has not formed a closed-loop management and control mechanism for the entire process. Overall, it suffers from common defects such as poor adaptability, low prediction accuracy, insufficient operating efficiency, weak anti-interference ability, and lagging iteration and update. It cannot adapt to multiple extreme scenarios and the operation needs of wind turbines throughout their entire life cycle, making it difficult to effectively support preventive operation and maintenance of wind turbines. It also cannot meet the high-precision, high-reliability, and real-time requirements of fault prediction for the large-scale and intelligent development of the wind power industry. Summary of the Invention

[0004] This invention addresses the problems of poor adaptability, low accuracy, and slow iteration in traditional fault prediction systems through the synergy of six modules, providing system support for preventive operation and maintenance of wind turbines and improving the operational stability and maintenance efficiency of wind turbines.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a cloud computing-based wind turbine fault prediction system, which includes six modules: data governance module, feature engineering module, fusion modeling module, cloud-edge collaboration module, online learning module, and environment adaptation module. Among them, the cloud computing platform provides distributed computing power support, global data storage and model scheduling management for the six major modules; The data governance module is used to process multi-source heterogeneous time-series data of wind turbines and output standardized data. The feature engineering module extracts a feature set related to wind turbine faults based on the standardized data output by the data governance module. The fusion modeling module, based on the feature set extracted by the feature engineering module, adopts a multi-model fusion approach driven by both data and physical mechanisms to train a high-precision wind turbine fault prediction model. The cloud-edge collaboration module is used to deploy the prediction model trained by the fusion modeling module, enabling collaborative deployment between the cloud and the edge. The online learning module uses actual feedback data to conduct online incremental iterative optimization of the model and update the model parameters. The environment adaptation module is used for environmental interference compensation and drift correction to ensure prediction accuracy in extreme environments.

[0006] Furthermore, the multi-source heterogeneous time-series data includes data from the unit's SCADA monitoring system, data from the condition monitoring system, meteorological and environmental monitoring data, and historical operation and maintenance records. The data governance module processes multi-source heterogeneous time-series data as follows: First, it performs a unified format conversion on the multi-source heterogeneous time-series data, converting it to CSV format. Then, it aligns the time-series data at a fixed time granularity of once per minute. For missing data, it uses linear interpolation combined with fitting of nearby time-series samples to fill in the missing data. It removes outlier data by using preset parameter normal threshold ranges. Finally, it uses the min-max normalization algorithm to scale the data to the [0,1] interval and outputs standardized time-series data.

[0007] Furthermore, the process of establishing a real-time data interaction relationship between the feature engineering module and the data governance module to extract fault-related feature sets is as follows: the standardized data is divided into segments according to a fixed time window. For key operating parameters such as vibration, temperature, and speed, a time-frequency domain joint analysis method combining wavelet transform and Fourier transform is used for preliminary feature extraction. Then, low-variance redundant features are eliminated through variance analysis, and features strongly correlated with faults are screened through mutual information method. Finally, the features are integrated to form a fault feature set that characterizes the operating status of the core components of the wind turbine gearbox, generator, and bearing.

[0008] Furthermore, the data and physical mechanism dual-driven multi-model fusion training process of the fusion modeling module is as follows: the feature set is divided into a training set, a validation set, and a test set according to a preset ratio; the LSTM deep learning model is used to perform time-series feature fitting training on the training set; at the same time, a physical mechanism model is constructed based on the structural mechanics and operational failure mechanism of the core components of the wind turbine; the output results of the deep learning model and the physical mechanism model are fused through a dynamic weighting algorithm based on the model validation accuracy; the fusion weights and single model parameters are iteratively adjusted using the validation set; and the model accuracy is verified through the test set to obtain the fault prediction model.

[0009] Furthermore, the cloud-edge collaborative deployment inference logic of the cloud-edge collaborative module is as follows: the cloud is responsible for the global multi-unit data aggregation management, the training scheduling and version management of the prediction model, and after the trained prediction model is compressed by model quantization or pruning, it is sent to the lightweight inference unit deployed at the edge. The edge inference unit is deployed near the wind turbine site, collects the unit operation data in real time and inputs it into the lightweight model to complete low-latency fault inference; when the inference result exceeds the preset fault warning threshold, a local audible and visual warning or platform warning is triggered, and the warning data and inference result are uploaded to the cloud simultaneously.

[0010] Furthermore, the actual feedback data from the online learning module includes post-operation and maintenance records of wind turbine units, fault diagnosis and confirmation results, and model prediction deviation data; The online incremental iterative optimization process of the model is as follows: the feedback data is denoised, deduplicated, cleaned, and classified with fault labels to select effective incremental iterative data; the online incremental learning algorithm is used to integrate the effective incremental data into the existing prediction model, freeze the core parameters of the model's underlying features, and update the top-level output parameters that are adapted to the incremental data, so as to avoid the model being retrained with all data and make the model adapt to the dynamic changes in the unit's operating status.

[0011] Furthermore, the environmental interference compensation and drift correction process of the environmental adaptation module is as follows: extreme environmental interference data such as sea salt spray, high-altitude and low-temperature conditions, strong mountain winds, and mountain dust are collected; core interference features are extracted through a feature selection-type interference extraction algorithm; the least squares compensation algorithm is used to offset the impact of environmental interference on the model input data and prediction results; the distribution of unit operation data and prediction deviation are monitored in real time; when the data drift or prediction deviation exceeds a preset threshold, a sliding window drift correction algorithm with a preset window size is used to dynamically correct the model input and inference logic to maintain stable prediction accuracy.

[0012] Furthermore, the six modules form a closed-loop linkage mechanism of positive execution and negative feedback. The data governance module outputs standardized data to the feature engineering module, the feature engineering module outputs feature sets to the fusion modeling module, the fusion modeling module outputs a prediction model to the cloud-edge collaboration module to complete inference and early warning, the inference results of the cloud-edge collaboration module and the actual feedback data from the field are input into the online learning module to complete model iteration, the model deviation data output by the online learning module drives the environment adaptation module to optimize the interference compensation strategy, and the environment correction parameters output by the environment adaptation module are input back into the data governance module to optimize the data anomaly removal and normalization rules.

[0013] Furthermore, the closed-loop linkage mechanism realizes the collaborative optimization of the entire system link. Through positive data flow and reverse parameter feedback, it completes the collaborative process of data processing, feature extraction, model training, inference warning, incremental iteration, and environment adaptation, suppresses prediction errors caused by the propagation of deviations in a single module, and realizes the continuous iterative upgrade of the fault prediction model.

[0014] Furthermore, the prediction system is adaptable to offshore, high-altitude, and mountainous wind farm scenarios, and covers the entire life cycle of wind turbine operation. Through least squares compensation and sliding window drift correction of the environment adaptation module, it matches the interference suppression requirements of different extreme environments. Through incremental iteration of the online learning module, it adapts to the full life cycle state changes of the unit during the break-in period, stable operation period, and aging period. The system achieves real-time and accurate fault prediction based on cloud computing, providing data support for the preventive operation and maintenance of wind turbines.

[0015] Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects: This invention constructs a closed-loop linkage mechanism with positive execution and negative feedback through the collaboration of six major modules. It is adapted to multiple extreme scenarios such as offshore, high-altitude cold, and mountainous areas, as well as the full life cycle operation requirements of wind turbine units. Each module performs its own function and works together to optimize. The data governance module provides high-quality standardized data, the feature engineering module extracts strong correlation features of faults, the fusion modeling module trains a high-precision prediction model, the cloud-edge collaboration module realizes low-latency deployment and inference, the online learning module ensures dynamic iteration of the model, and the environment adaptation module maintains the prediction accuracy in extreme environments. This solves the problems of poor adaptability, low accuracy, and slow iteration of traditional fault prediction systems. Attached Figure Description

[0016] Figure 1 This is a system diagram of the cloud computing-based wind turbine fault prediction system of the present invention. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] The present invention will now be described in further detail with reference to the accompanying drawings: Example: like Figure 1 As shown, this invention provides a cloud computing-based wind turbine fault prediction system. This prediction system comprises six modules: a data governance module, a feature engineering module, a fusion modeling module, a cloud-edge collaboration module, an online learning module, and an environment adaptation module. The cloud computing platform provides distributed computing power support, global data storage, and model scheduling management for these six modules. The cloud computing platform can adopt a Kubernetes distributed cluster architecture, with the number of nodes adjustable according to the scale of the wind farm's turbines. Considering that a single wind farm turbine generates approximately 500MB of time-series data daily, a single slave node can stably handle the parallel data processing and model training tasks of 50 turbines, balancing computing power redundancy and cost control. This ensures the system efficiently processes massive amounts of time-series data without wasting computing power. For every 100 turbines, two additional slave nodes can be added for dynamic expansion. These slave nodes are responsible for parallel data processing and model training, ensuring the system can efficiently process the massive amounts of time-series data from wind turbines.

[0020] Furthermore, the six modules form a closed-loop linkage mechanism of positive execution and negative feedback. The data governance module outputs standardized data to the feature engineering module, the feature engineering module outputs feature sets to the fusion modeling module, the fusion modeling module outputs a prediction model to the cloud-edge collaboration module to complete inference and early warning, the inference results of the cloud-edge collaboration module and the actual feedback data from the field are input into the online learning module to complete model iteration, the model deviation data output by the online learning module drives the environment adaptation module to optimize the interference compensation strategy, and the environmental correction parameters output by the environment adaptation module are input back into the data governance module to optimize the data anomaly removal and normalization rules. The response latency of the closed-loop linkage is ≤500ms, ensuring that model iteration and parameter correction can adapt to changes in the unit's operating status in real time.

[0021] Furthermore, the closed-loop linkage mechanism realizes full-link collaborative optimization of the system. Through positive data flow and reverse parameter feedback, it completes the full-process collaboration of data processing, feature extraction, model training, inference warning, incremental iteration, and environment adaptation, suppresses prediction errors caused by the propagation of deviations in a single module, and realizes continuous iterative upgrades of the fault prediction model.

[0022] Furthermore, the prediction system is adaptable to offshore, high-altitude, and mountainous wind farm scenarios, and covers the entire life cycle of wind turbine operation. Through least squares compensation and sliding window drift correction of the environment adaptation module, it matches the interference suppression requirements of different extreme environments. Through incremental iteration of the online learning module, it adapts to the full life cycle state changes of the unit during the break-in period, stable operation period, and aging period. The system achieves real-time and accurate fault prediction based on cloud computing, providing data support for the preventive operation and maintenance of wind turbines.

[0023] The data governance module is used to process multi-source heterogeneous time-series data of wind turbines and output standardized data.

[0024] Furthermore, the multi-source heterogeneous time-series data includes data from the unit's SCADA monitoring system, data from the condition monitoring system, meteorological and environmental monitoring data, and historical operation and maintenance records. The data governance module processes multi-source heterogeneous time-series data as follows: First, it performs a unified format conversion on the multi-source heterogeneous time-series data. Then, it aligns the time-series data at a fixed time granularity of once per minute. For missing data, it uses linear interpolation combined with fitting of nearby time-series samples to fill in the missing data. It removes outlier data by using preset parameter normal threshold ranges. Finally, it uses the min-max normalization algorithm to scale the data to the [0,1] interval and outputs standardized time-series data.

[0025] The feature engineering module extracts a set of features related to wind turbine faults based on the standardized data output by the data governance module.

[0026] Furthermore, the process of establishing a real-time data interaction relationship between the feature engineering module and the data governance module to extract fault-related feature sets is as follows: the standardized data is divided into segments according to fixed time windows; for key operating parameters such as vibration, temperature, and speed, a time-frequency domain joint analysis method combining wavelet transform and Fourier transform is used for preliminary feature extraction; then, low-variance redundant features are eliminated through variance analysis, and features strongly correlated with faults are screened through mutual information method; finally, the features are integrated to form a fault feature set characterizing the operating status of the core components of the wind turbine gearbox, generator, and bearing.

[0027] The fusion modeling module, based on the feature set extracted by the feature engineering module, adopts a multi-model fusion approach driven by both data and physical mechanisms to train a high-precision wind turbine fault prediction model.

[0028] Furthermore, the data and physical mechanism dual-driven multi-model fusion training process of the fusion modeling module is as follows: the feature set is divided into a training set, a validation set, and a test set according to a preset ratio. The LSTM deep learning model is used to perform time-series feature fitting training on the training set. At the same time, a physical mechanism model is constructed based on the structural mechanics and operational failure mechanism of the core components of the wind turbine. The output results of the deep learning model and the physical mechanism model are fused through a dynamic weighting algorithm based on the model validation accuracy. The fusion weights and single model parameters are iteratively adjusted using the validation set. The model accuracy is verified through the test set to obtain the fault prediction model.

[0029] The cloud-edge collaboration module is used to deploy the prediction model trained by the fusion modeling module, enabling collaborative deployment between the cloud and the edge.

[0030] Furthermore, the cloud-edge collaborative deployment inference logic of the cloud-edge collaborative module is as follows: the cloud is responsible for the global multi-unit data aggregation management, the training scheduling and version management of the prediction model, and after the trained prediction model is compressed by model quantization or pruning, it is distributed to the lightweight inference unit deployed at the edge. The edge inference unit is deployed near the wind turbine site, collects the unit operation data in real time and inputs it into the lightweight model to complete low-latency fault inference; when the inference result exceeds the preset fault warning threshold, a local audible and visual warning or platform warning is triggered, and the warning data and inference result are synchronously uploaded to the cloud.

[0031] The online learning module uses actual feedback data to conduct online incremental iterative optimization of the model and update the model parameters.

[0032] Furthermore, the actual feedback data of the online learning module includes post-operation and maintenance records of wind turbine units, fault investigation and confirmation results, and model prediction deviation data. The online incremental iterative optimization process of the model is as follows: the feedback data is denoised, deduplicated, cleaned, and classified with fault labels to select effective incremental iterative data; an online incremental learning algorithm is adopted to integrate the effective incremental data into the existing prediction model, freeze the core parameters of the model's underlying features, and update the top-level output parameters that are adapted to the incremental data, avoiding full retraining of the model and enabling the model to adapt to the dynamic changes in the unit's operating status.

[0033] The environment adaptation module is used for environmental interference compensation and drift correction to ensure prediction accuracy in extreme environments.

[0034] Furthermore, the environmental interference compensation and drift correction process of the environmental adaptation module is as follows: Extreme environmental interference data such as sea salt spray, high-altitude and low-temperature conditions, strong mountain winds, and mountain dust storms are collected. A random forest feature importance algorithm is adopted. This algorithm has strong fault tolerance for redundant interference data in extreme environments, high feature extraction efficiency, and requires no complex parameter debugging. It adapts to the extraction needs of heterogeneous interference data from wind farms and can quantify the influence of interference factors through feature importance, outperforming other feature selection algorithms. The feature importance threshold is set to ≥0.1 to extract core interference features. After testing with interference data from multiple scenarios including sea, high-altitude and mountainous areas, a threshold below 0.1 introduces a large number of invalid interference features, leading to redundancy in the compensation algorithm. A threshold above 0.1 misses key interference features. Setting the threshold to 0.1 balances feature effectiveness and algorithm efficiency. For example, salt spray concentration, low temperature value, and wind speed gradient are used. A least squares compensation algorithm is employed to offset the impact of environmental interference on the model input data and prediction results. The distribution of unit operation data and prediction deviation values ​​are monitored in real time. The sliding window size is set to 24 hours, focusing on the core components of the wind turbine unit. For example, the typical time span for the precursory features of gearbox and bearing failures is 12-24 hours. Setting the window to 24 hours can fully capture the environmental interference changes and data drift trends corresponding to the precursory features of failures. At the same time, it avoids the frequent corrections caused by a window that is too short and the accuracy lag caused by a window that is too long. The data drift threshold is ±10% of the distribution mean deviation and the prediction deviation threshold is 5%. When the amount of data drift or prediction deviation exceeds the preset threshold, the sliding window drift correction algorithm is used to dynamically correct the model input and inference logic to maintain stable prediction accuracy.

[0035] The prediction system of this invention was tested against a conventional fault prediction system over a period of 6 months. One hundred wind turbines from three typical wind farms (offshore, high-altitude, and mountainous) were selected, and one million multi-source heterogeneous time-series data points were collected, with fault data accounting for 15%. The overall performance of the two systems was compared, and the results are shown in the table below:

[0036] As shown in the table above, six indicators were selected: data processing latency, fault prediction accuracy, data transmission latency, extreme environment prediction accuracy, model iteration efficiency, and closed-loop linkage response. These indicators cover the entire performance process of the system, including data processing, fault prediction, deployment and transmission, environment adaptation, and model iteration. The test scenarios closely match actual wind farms in offshore, high-altitude, and mountainous areas. All indicators of this invention significantly outperform conventional systems. Specifically, the performance of data processing latency and data transmission latency is improved by ≥50%, the fault prediction accuracy is improved by 15%-17% (compared to 93%-95% for conventional systems, 78%-83%), the prediction accuracy is improved by 8%-12% in extreme environments, the closed-loop linkage response latency is controlled within ≤500ms (compared to ≥800ms for conventional systems), and the model iteration efficiency is improved by ≥20%.

[0037] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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 will 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 cloud computing-based wind turbine fault prediction system, characterized in that, The prediction system comprises six modules: data governance, feature engineering, fusion modeling, cloud-edge collaboration, online learning, and environment adaptation. Among them, the cloud computing platform provides distributed computing power support, global data storage and model scheduling management for the six major modules; The data governance module is used to process multi-source heterogeneous time-series data of wind turbines and output standardized data. The feature engineering module extracts a set of features related to wind turbine faults based on the standardized data output by the data governance module. The fusion modeling module, based on the feature set extracted by the feature engineering module, adopts a multi-model fusion approach driven by both data and physical mechanisms to train a high-precision wind turbine fault prediction model. The cloud-edge collaboration module is used to deploy the prediction model trained by the fusion modeling module, enabling collaborative deployment between the cloud and the edge. The online learning module uses actual feedback data to conduct online incremental iterative optimization of the model and update the model parameters. The environment adaptation module is used for environmental interference compensation and drift correction to ensure prediction accuracy in extreme environments.

2. The cloud computing-based wind turbine fault prediction system according to claim 1, characterized in that, The multi-source heterogeneous time-series data includes data from the unit's SCADA monitoring system, data from the condition monitoring system, meteorological and environmental monitoring data, and historical operation and maintenance records. The data governance module processes multi-source heterogeneous time-series data as follows: First, it performs a unified format conversion on the multi-source heterogeneous time-series data. Then, it aligns the time-series data at a fixed time granularity of once per minute. For missing data, it uses linear interpolation combined with fitting of nearby time-series samples to fill in the missing data. It removes outlier data by using preset parameter normal threshold ranges. Finally, it uses the min-max normalization algorithm to scale the data to the [0,1] interval and outputs standardized time-series data.

3. The cloud computing-based wind turbine fault prediction system according to claim 2, characterized in that, The process of establishing a real-time data interaction relationship between the feature engineering module and the data governance module to extract fault-related feature sets is as follows: the standardized data is divided into segments according to a fixed time window. For key operating parameters such as vibration, temperature, and speed, a time-frequency domain joint analysis method combining wavelet transform and Fourier transform is used for preliminary feature extraction. Then, low-variance redundant features are eliminated through variance analysis, and features strongly correlated with faults are screened through mutual information method. Finally, the features are integrated to form a fault feature set that characterizes the operating status of the core components of the wind turbine gearbox, generator, and bearing.

4. The cloud computing-based wind turbine fault prediction system according to claim 3, characterized in that, The data and physical mechanism dual-driven multi-model fusion training process of the fusion modeling module is as follows: the feature set is divided into training set, validation set and test set according to a preset ratio. The LSTM deep learning model is used to perform time series feature fitting training on the training set. At the same time, a physical mechanism model is constructed based on the structural mechanics and operation failure mechanism of the core components of the wind turbine. The output results of the deep learning model and the physical mechanism model are fused through a dynamic weighting algorithm based on the model validation accuracy. The fusion weights and single model parameters are iteratively adjusted using the validation set. The model accuracy is verified through the test set to obtain the fault prediction model.

5. The cloud computing-based wind turbine fault prediction system according to claim 4, characterized in that, The cloud-edge collaborative deployment inference logic of the cloud-edge collaborative module is as follows: The cloud is responsible for the global multi-unit data aggregation management, the training scheduling and version management of the prediction model. After the trained prediction model is quantized or pruned and compressed, it is distributed to the lightweight inference unit deployed at the edge. The edge inference unit is deployed near the wind turbine site, collects the unit operation data in real time and inputs it into the lightweight model to complete low-latency fault inference. When the inference result exceeds the preset fault warning threshold, a local audible and visual warning or platform warning is triggered, and the warning data and inference result are uploaded to the cloud simultaneously.

6. The cloud computing-based wind turbine fault prediction system according to claim 5, characterized in that, The actual feedback data from the online learning module includes post-operation and maintenance records of wind turbine units, fault diagnosis and confirmation results, and model prediction deviation data. The online incremental iterative optimization process of the model is as follows: the feedback data is denoised, deduplicated, cleaned, and classified with fault labels to select effective incremental iterative data; the online incremental learning algorithm is used to integrate the effective incremental data into the existing prediction model, freeze the core parameters of the model's underlying features, and update the top-level output parameters that are adapted to the incremental data, so as to avoid the model being retrained with all data and make the model adapt to the dynamic changes in the unit's operating status.

7. The cloud computing-based wind turbine fault prediction system according to claim 6, characterized in that, The environmental interference compensation and drift correction process of the environmental adaptation module is as follows: extreme environmental interference data such as sea salt spray, high-altitude and low-temperature, mountain strong wind and mountain dust are collected. The core interference features are extracted by feature selection interference extraction algorithm. The least squares compensation algorithm is used to offset the impact of environmental interference on the model input data and prediction results. The distribution of unit operation data and prediction deviation are monitored in real time. When the data drift or prediction deviation exceeds the preset threshold, the sliding window drift correction algorithm with a preset window size is used to dynamically correct the model input and inference logic to maintain stable prediction accuracy.

8. The cloud computing-based wind turbine fault prediction system according to claim 7, characterized in that, The six modules form a closed-loop linkage mechanism of positive execution and negative feedback. The data governance module outputs standardized data to the feature engineering module, the feature engineering module outputs feature sets to the fusion modeling module, the fusion modeling module outputs a prediction model to the cloud-edge collaboration module to complete inference and early warning, the inference results of the cloud-edge collaboration module and the actual feedback data from the field are input into the online learning module to complete model iteration, the model deviation data output by the online learning module drives the environment adaptation module to optimize the interference compensation strategy, and the environment correction parameters output by the environment adaptation module are input back into the data governance module to optimize the data anomaly removal and normalization rules.

9. The cloud computing-based wind turbine fault prediction system according to claim 8, characterized in that, The closed-loop linkage mechanism realizes the collaborative optimization of the entire system link. Through positive data flow and reverse parameter feedback, it completes the collaborative process of data processing, feature extraction, model training, inference warning, incremental iteration, and environment adaptation, suppresses prediction errors caused by the propagation of deviations in a single module, and realizes the continuous iterative upgrade of the fault prediction model.

10. The cloud computing-based wind turbine fault prediction system according to claim 9, characterized in that, The prediction system is adaptable to offshore, high-altitude, and mountainous wind farm scenarios, and covers the entire life cycle of wind turbine operation. Through least squares compensation and sliding window drift correction of the environment adaptation module, it matches the interference suppression requirements of different extreme environments. Through incremental iteration of the online learning module, it adapts to the full life cycle status changes of the unit during the break-in period, stable operation period, and aging period. The system achieves real-time and accurate fault prediction based on cloud computing, providing data support for the preventive operation and maintenance of wind turbines.

Images