Wind power key equipment state predictive maintenance method based on multi-source data fusion

Abstract

The application discloses a wind power key equipment state predictive maintenance method based on multi-source data fusion, relates to the wind power equipment operation and maintenance technical field, and comprises the following steps: a three-level fusion mechanism of original data alignment, cross-domain feature extraction and decision information synthesis is constructed, and a fusion feature set is generated; a hybrid prediction model is established based on the fusion feature set, and remaining service life, fault type and performance degradation critical value cross-validation results of equipment are output; a fault early warning state is determined according to a remaining life threshold value, a fault risk probability threshold value and a performance degradation critical threshold value; and a repair scheme is matched based on the fault early warning state. The application integrates multi-source heterogeneous data of wind power key equipment to construct a three-level fusion mechanism, combines a hybrid prediction model with an attention mechanism to output multi-dimensional equipment state cross-validation results, matches a differentiated maintenance scheme, continuously optimizes through a closed-loop feedback, and realizes precise adaptation of equipment predictive maintenance.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment operation and maintenance technology, specifically to a method for predictive maintenance of key wind power equipment based on multi-source data fusion. Background Technology

[0002] Wind power equipment is exposed to complex outdoor environments for extended periods. Key components such as wind turbine blades and gearboxes are susceptible to factors such as wind speed fluctuations and temperature changes, which can lead to malfunctions and shutdowns. Currently, maintenance is mainly based on periodic inspections, which can result in over-maintenance or untimely maintenance. This not only increases operation and maintenance costs but may also cause significant economic losses due to sudden malfunctions.

[0003] In existing technologies, wind power equipment condition monitoring relies heavily on single-type data, which makes it difficult to comprehensively reflect the actual operating status of the equipment. The data fusion mechanism is imperfect, and the spatiotemporal differences and redundant information of heterogeneous data are not effectively processed, resulting in insufficient feature representation capabilities. The prediction model is not targeted enough and cannot accurately capture the temporal dependencies and key influencing factors of equipment failures, thereby affecting the accuracy of fault warnings and the adaptability of maintenance plans. To address the above technical deficiencies, a solution is proposed. Summary of the Invention

[0004] The purpose of this invention is to solve the problems mentioned above by proposing a predictive maintenance method for key wind power equipment based on multi-source data fusion.

[0005] The objective of this invention can be achieved through the following technical solutions: The method for predictive maintenance of key wind power equipment based on multi-source data fusion includes: collecting operating status data, environmental perception data and full life cycle operation and maintenance data of key wind power equipment; constructing a three-level fusion mechanism of raw data alignment, cross-domain feature extraction and decision information synthesis; integrating heterogeneous data; and generating a fusion feature set. Based on the aforementioned fusion feature set, a hybrid prediction model is established to capture key dimensions and time-series dependencies that are strongly correlated with equipment failure in the fusion feature set, and outputs cross-validation results of equipment remaining service life, failure type, and performance degradation threshold. Based on the cross-validation results of the remaining service life of the equipment, the fault type, and the performance degradation threshold, the fault warning status is determined according to the remaining service life threshold, the fault risk probability threshold, and the performance degradation threshold. Based on the fault warning status matching maintenance plan, the frequency of routine inspections and key monitoring parts are determined based on the cross-validation results of performance degradation threshold, and maintenance measures are matched with the predicted equipment status.

[0006] Preferably, the collection of operational status data, environmental perception data, and full lifecycle operation and maintenance data of key wind power equipment is used to construct a three-level fusion mechanism for raw data alignment, cross-domain feature extraction, and decision information synthesis. This mechanism integrates heterogeneous data to generate a fused feature set, including: The key wind power equipment includes wind turbine blades, gearbox, generator, main shaft and converter. The operating status data includes vibration, temperature, speed, oil quality, voltage and current data, which are collected in real time by Internet of Things sensors deployed in the bearings, housing and oil circuit of the equipment. The environmental sensing data includes wind speed, wind direction, air humidity, ambient temperature, and salt spray concentration data, which are collected synchronously through meteorological stations deployed in the wind farm and environmental monitoring sensors built into the equipment. The full lifecycle operation and maintenance data includes the time of failure, failure description, maintenance work order details, spare parts replacement information, equipment factory parameters and previous maintenance records, which are extracted from the wind farm operation and maintenance management system.

[0007] Preferably, a three-level fusion mechanism is constructed, consisting of original data alignment, cross-domain feature extraction, and decision information synthesis, integrating heterogeneous data to obtain a fused feature set, including: Based on the acquisition frequency, measurement accuracy, and transmission stability of each data source, confidence weights are assigned. Based on these confidence weights, the original data of different dimensions are aligned in time and matched in spatial location. A data missing rate threshold is set, and invalid data segments with missing rates exceeding the threshold are removed to obtain a preliminary aligned dataset. Local spatiotemporal features and temporal dependency features are extracted from the preliminary aligned dataset. The two types of features are concatenated to generate a high-dimensional feature vector. The high-dimensional feature vector is then subjected to dimensionality reduction processing to remove redundant information and obtain a simplified feature vector. Fault-related features are extracted from the full lifecycle operation and maintenance data, including the trend of changes in operating parameters before the fault and the correspondence between historical faults and operating status. The simplified feature vector is fused with the fault-related features, the feature credibility is calculated, and a fused feature set with both completeness and relevance is generated.

[0008] Preferably, a hybrid prediction model is established based on the fused feature set, including: The hybrid prediction model strengthens the weights of feature dimensions that are strongly correlated with equipment failures, highlights the impact of key information on the prediction results, adjusts the learning rate during model training, and sets an early stopping mechanism to avoid overfitting. Using historical equipment fault data and corresponding fusion feature sets as training samples, the model is divided into training, validation, and test sets, and the model parameters are iteratively optimized.

[0009] Preferably, the cross-validation results of the remaining service life, failure type, and performance degradation threshold of the output device include: The trained hybrid prediction model outputs the predicted remaining useful life of the equipment, the failure type classification results, and the performance degradation trend curve. Cross-validation is used to verify the above results, evaluate the accuracy of remaining useful life prediction and the failure type classification effect, and determine the confidence interval of the performance degradation critical value. The results of the integrated verification are used to form cross-verification results for the remaining service life of the equipment, failure type, and performance degradation threshold.

[0010] Preferably, the fault warning status is determined based on the cross-validation results, including: The remaining life threshold is dynamically updated based on the equipment's designed service life, the average remaining life of historical faults, and the equipment's importance level. The fault risk probability threshold is divided into interval ranges and determined based on the wind farm's historical fault statistics and operation and maintenance costs. The performance degradation threshold is determined based on the equipment's rated operating parameters and the performance degradation limit value provided by the manufacturer. When the predicted remaining service life is lower than the remaining service life threshold, the failure risk probability exceeds the high threshold, and the performance degradation value exceeds the critical threshold, it is determined to be an emergency warning state. When the failure risk probability is within the threshold range, and the remaining service life and performance degradation value do not trigger the corresponding threshold, it is determined to be a general warning state. When the failure risk probability is lower than the low threshold, and the remaining service life and performance degradation value do not trigger the corresponding threshold, it is determined to be a normal state.

[0011] Preferably, the maintenance plan matched based on the fault warning status includes: If an emergency warning state is determined, an emergency maintenance plan will be matched, and first-level maintenance resources will be prioritized. Based on the priority of fault handling, key maintenance procedures and the list of necessary spare parts, maintenance work will be initiated within the specified time limit. If the situation is determined to be a general warning state, a planned maintenance scheme will be matched, and a short-term maintenance plan will be formulated based on the real-time power generation load of the wind farm and weather conditions to obtain the maintenance content and expected results. If the condition is determined to be normal, maintain the regular maintenance mode and optimize the regular inspection strategy based on the cross-validation results of the performance degradation threshold.

[0012] Preferably, the frequency of routine inspections and key monitoring areas are determined based on the cross-validation results of the performance degradation threshold, including: extracting the performance degradation rate of each component of the equipment from the performance degradation trend curve, marking the components with faster degradation rates as key monitoring areas, adjusting the frequency of routine inspections according to the proportion of key monitoring areas and the fluctuation range of the confidence interval of the performance degradation threshold, and identifying targeted inspection items such as vibration detection, temperature monitoring, and oil analysis for key monitoring areas.

[0013] Preferably, the raw data alignment stage also includes data preprocessing operations to normalize heterogeneous data of different dimensions, unify the data range, identify outliers in the data, use appropriate correction methods according to the degree of deviation of outliers, and retain the abrupt change data characteristics corresponding to the precursory signs of equipment failure.

[0014] Preferably, the method further includes a maintenance effect feedback optimization step: collecting the operating status data of the equipment after maintenance and generating a post-maintenance fusion feature set; inputting the post-maintenance fusion feature set into the hybrid prediction model to obtain the post-maintenance equipment status prediction result; comparing the equipment status prediction results before and after maintenance, calculating the failure risk reduction rate and performance degradation rate mitigation rate evaluation index; adjusting the confidence weight of the three-level fusion mechanism and the parameters of the hybrid prediction model based on the failure risk reduction rate and performance degradation rate mitigation rate evaluation index, and periodically performing incremental training of the model based on new data.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a three-level data fusion mechanism, integrating multi-source heterogeneous data from operational status, environmental perception, and full lifecycle operation and maintenance. Through data alignment, feature extraction, and information synthesis, it solves the problem of one-sided representation by single data. The generated fused feature set has both completeness and relevance, laying the foundation for accurate prediction. During the preprocessing process, the characteristics of fault precursor mutations are retained, further improving the quality of features.

[0016] 2. A hybrid prediction model is established based on the fusion feature set, which strengthens the weight of fault-related features and outputs multi-dimensional prediction results through cross-validation. This accurately captures the key dimensions and temporal dependencies of equipment faults, and improves the prediction accuracy of remaining life, fault type and performance degradation threshold. The maintenance effect feedback optimization mechanism realizes the dynamic adjustment of the model and fusion mechanism, and continuously improves the prediction accuracy.

[0017] 3. Based on multi-dimensional thresholds, determine the fault warning status, match differentiated maintenance plans, and optimize inspection strategies in combination with performance degradation. This achieves accurate matching between maintenance measures and predicted equipment status, avoids over-maintenance and untimely maintenance, reduces operation and maintenance costs, and ensures the safe and stable operation of wind power equipment. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the steps of the present invention. Detailed Implementation

[0020] 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 are within the scope of protection of the present invention.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] Please see Figure 1 As shown, the predictive maintenance method for key wind power equipment based on multi-source data fusion includes: Step 1: Collect operational status data, environmental perception data, and full lifecycle operation and maintenance data of key wind power equipment; construct a three-level fusion mechanism for raw data alignment, cross-domain feature extraction, and decision information synthesis; integrate heterogeneous data; and generate a fusion feature set. Step 2: Based on the fused feature set, establish a hybrid prediction model, capture the key dimensions and time series dependencies that are strongly correlated with equipment failure in the fused feature set, and output the cross-validation results of the remaining service life of the equipment, failure type and performance degradation threshold. Step 3: Based on the cross-validation results of the remaining service life of the equipment, the fault type, and the performance degradation threshold, determine the fault warning status according to the remaining service life threshold, the fault risk probability threshold, and the performance degradation threshold. Step 4: Match maintenance plans based on fault warning status, determine the frequency of routine inspections and key monitoring parts based on the cross-validation results of performance degradation threshold, and match maintenance measures with the predicted equipment status; Step 5: Collect the operating status data of the equipment after maintenance, and optimize the three-level fusion mechanism and hybrid prediction model through maintenance effect evaluation feedback.

[0023] The above solutions utilize multi-source data collection to cover all dimensions of equipment operation information, overcoming the limitations of single data sources; a three-level fusion mechanism enables deep integration of heterogeneous data, enhancing feature representation capabilities; a hybrid prediction model accurately captures the evolution patterns of equipment status and outputs multi-dimensional prediction results; a threshold-based early warning mechanism and differentiated maintenance schemes achieve precise operation and maintenance, avoiding over-maintenance or untimely maintenance; and a closed-loop feedback optimization mechanism ensures that the system continuously adapts to changes in equipment operation, improving the efficiency of operation and maintenance of key wind power equipment.

[0024] In some embodiments, step 1 collects operational status data, environmental perception data, and full lifecycle operation and maintenance data of key wind power equipment, constructs a three-level fusion mechanism of raw data alignment, cross-domain feature extraction, and decision information synthesis, integrates heterogeneous data, and generates a fused feature set, including: The core monitoring objects of key wind power equipment are identified, including wind turbine blades, gearboxes, generators, main shafts and converters, and three types of core data are collected. A three-level fusion mechanism is constructed, which includes raw data alignment, cross-domain feature extraction and decision information synthesis. Through data preprocessing, feature mining and information fusion, a fusion feature set with completeness, relevance and effectiveness is generated.

[0025] Specifically, the data collection for the operational status of key wind power equipment requires the deployment of an IoT sensor matrix targeting critical fault-prone areas of each piece of equipment. Vibration data is collected through accelerometers installed on vibration-sensitive parts such as bearings of rotating components and equipment housings, covering both steady-state and transient vibration signals during equipment operation. Temperature data collection covers key heat-generating components and lubrication circuits, capturing signals of equipment temperature rise and abnormal overheating. Speed ​​data is acquired through equipment-integrated encoding devices or non-contact speed sensors, reflecting the equipment's operating conditions. Oil quality data is collected for lubrication systems such as gearboxes and generators, collecting signals related to the physical and chemical properties of the oil. Electrical parameters such as voltage and current are synchronously collected through electrical sensors at the input and output terminals of converters and generators to monitor the operating status of the electrical system. All sensor data is uploaded to the data center in real time via an industrial-grade data transmission network to ensure the continuity and stability of data transmission.

[0026] Environmental perception data collection needs to be based on the geographical environment and climate characteristics of the wind farm, and a comprehensive environmental perception network needs to be constructed. Wind speed and wind direction data are collected through meteorological monitoring equipment deployed in the wind farm to capture the impact of wind resource fluctuations on equipment operation; ambient temperature and air humidity data are obtained through temperature and humidity sensors to cover environmental conditions at all times during equipment operation; for special environmental areas such as coastal areas and areas with high salt spray, additional specialized environmental sensors are deployed to collect corrosive environmental parameters such as salt spray concentration. All environmental perception data and operational status data are time-stamped and correlated to ensure data consistency in the time dimension.

[0027] Full lifecycle operation and maintenance data extraction requires systematically extracting relevant data from the wind farm operation and maintenance management system. This includes basic information such as rated parameters and design specifications at the equipment delivery stage; fault records such as fault occurrence time, fault phenomenon description, and fault location results during the operation stage; operation and maintenance records such as maintenance work order details, maintenance procedures, spare parts replacement models and cycles, and maintenance personnel configuration during the maintenance stage; and maintenance data such as inspection items, inspection methods, and inspection results from each overhaul. The extracted unstructured data is then structured, and the data format and field definitions are standardized to form a standardized full lifecycle operation and maintenance dataset.

[0028] The raw data alignment operation requires the establishment of a data source reliability assessment system, and the allocation of reliability weights based on the acquisition frequency, measurement accuracy, transmission stability and anti-interference capability of each data source. Among them, operational status data, due to its high collection frequency and direct reflection of equipment operating status, is assigned the highest confidence weight; environmental perception data, with a lower degree of external interference, is assigned a medium confidence weight; and full lifecycle maintenance data, although not real-time data, is closely related to equipment failures and is assigned a corresponding confidence weight. Using the timestamps of high-frequency operational status data as a benchmark, an interpolation synchronization method adapted to different data types is used to align the time axis of low-frequency environmental perception data and discrete maintenance data, linking all data sources on the same time scale. A data missing rate judgment standard is set, and invalid data segments with missing rates exceeding the threshold in continuous data segments are directly removed; for data segments with missing rates below the threshold, a completion algorithm based on data distribution characteristics is used for filling. A data preprocessing process is executed synchronously: standardized processing methods are used for heterogeneous data of different dimensions to unify the data value range and eliminate the impact of dimensional differences; outlier identification algorithms are used to filter out outliers in the data, and the nature of the anomaly is determined by combining equipment operation logs and fault history records. If the anomaly is a sudden change caused by a precursor to equipment failure, the original characteristics are retained; if it is noise interference during data collection or transmission, an appropriate correction method is used.

[0029] Cross-domain feature extraction requires extracting core features from the initially aligned dataset in different dimensions. On one hand, spatial feature extraction algorithms are used to mine local spatiotemporal features. By analyzing the spatial structure of time-series data such as vibration and temperature, local changes in equipment operating status are captured. On the other hand, temporal feature extraction algorithms are used to mine temporal dependency features. This captures the trend changes and correlations of data in the time dimension, reflecting the dynamic evolution of equipment status. The extracted local spatiotemporal features and temporal dependency features are concatenated to generate a high-dimensional feature vector. Since there is redundant information and noise interference in the high-dimensional features, feature dimensionality reduction algorithms are used to process the high-dimensional feature vector. By retaining principal components whose cumulative variance contribution rate meets the preset requirements and removing redundant information, a simplified feature vector with reduced dimensions and dense information is obtained.

[0030] The decision information synthesis operation requires in-depth mining of fault correlation features from the entire lifecycle operation and maintenance data. Through statistical analysis and association rule mining, key information such as the trend of changes in operating parameters before the fault occurred, the correspondence between historical fault types and operating status parameters, and the correlation between maintenance measures and fault repair effects are extracted. Fault correlation feature vectors are constructed, and a weighted fusion algorithm is used to fuse the simplified feature vectors with the fault correlation feature vectors. The fusion weights are dynamically adjusted according to the importance of the features to equipment fault prediction. The importance score of each feature is calculated by the feature importance evaluation algorithm, and the score is normalized and used as the fusion weight. The credibility of each feature after fusion is calculated. The credibility level of the features is determined by the correlation analysis between the features and historical fault data. Invalid features with credibility below the threshold are removed, and finally, a fused feature set is generated.

[0031] In some embodiments, step 2 establishes a hybrid prediction model based on the fused feature set, captures key dimensions and time-series dependencies strongly correlated with equipment failure in the fused feature set, and outputs cross-validation results of remaining equipment lifespan, failure type, and performance degradation threshold, including: The hybrid prediction model uses a fusion architecture that combines spatial feature mining, temporal dependency capture, and key feature enhancement. It uses historical equipment fault data and the corresponding fusion feature set as training samples. After dataset partitioning, parameter optimization, and training validation, it outputs multi-dimensional prediction results. The multi-fold cross-validation method is used to verify the reliability of the prediction results, forming a complete cross-validation result set.

[0032] Specifically, the model structure design needs to meet the requirements of multi-task prediction. The spatial feature mining module is used to further mine the local spatial correlation information in the fusion feature set and capture the spatial distribution characteristics of the equipment operating status. The temporal dependency capture module is used to mine the long-term dependency relationship of features in the time dimension and reflect the temporal evolution law of the equipment status. The key feature enhancement module uses an attention mechanism to calculate the correlation between each feature and the equipment fault, assign higher weights to feature dimensions that are strongly correlated with the fault, highlight the impact of key information on the prediction results, and suppress the interference of irrelevant features. The model output layer is designed as a multi-task output structure, corresponding to the three core prediction tasks of equipment remaining service life prediction, fault type classification, and performance degradation threshold calculation.

[0033] Model training requires historical equipment fault data and corresponding fused feature sets as training samples, divided into training, validation, and test sets according to a preset ratio. The training set is used for model parameter learning, the validation set for hyperparameter tuning and overfitting monitoring, and the test set for final model performance evaluation. The model training process and termination conditions are defined, and an adaptive optimization algorithm is used to adjust the learning rate. The learning step size is dynamically adjusted based on changes in the validation set loss to improve model convergence speed and training accuracy. An early stopping mechanism is implemented: when the validation set loss does not decrease for several consecutive training cycles, model training is stopped to avoid overfitting. Multiple rounds of iterative training optimize model parameters, ensuring that the model achieves preset performance metrics on both the training and validation sets.

[0034] Cross-validation requires the use of multi-fold cross-validation to verify the reliability of the model's prediction results. The training and validation sets are merged and then evenly divided into multiple independent subsets. Each subset is selected sequentially as the validation set, while the remaining subsets are used as the training set for model training and prediction. This process is repeated for all subsets, and the average of the multiple validation results is taken as the final validation result, reducing the impact of data partitioning randomness on model evaluation. Differentiated evaluation metrics are set for different prediction tasks: for equipment remaining service life prediction results, error-based metrics between predicted and actual values ​​are used to evaluate prediction accuracy; for fault type classification results, classification evaluation metrics such as accuracy, recall, and F1 score are used to evaluate classification performance; for performance degradation thresholds, confidence intervals are determined by statistically analyzing the distribution characteristics of the prediction results to ensure the reliability and stability of the thresholds.

[0035] Cross-validation result integration requires the system to integrate the results from multiple fold cross-validations, forming a complete cross-validation result set that includes predicted remaining service life of equipment and accuracy assessment indicators, fault type classification results and classification performance indicators, performance degradation thresholds and confidence intervals. Consistency analysis is performed on the validation results. If the differences between validation results from different folds exceed a preset range, the feature extraction or model training process is reviewed back, relevant parameters are optimized, and re-validation is conducted to ensure the stability and reliability of the prediction results.

[0036] In some embodiments, step 3, based on the cross-validation results of the remaining useful life of the device, the fault type, and the performance degradation threshold, determines the fault warning status according to the remaining useful life threshold, the fault risk probability threshold, and the performance degradation threshold, including: Establish a threshold adjustment mechanism, setting a remaining life threshold, a failure risk probability threshold, and a performance degradation critical threshold respectively; through multi-dimensional threshold comparison, classify the equipment status into three levels: emergency warning status, general warning status, and normal status.

[0037] Specifically, the setting of the remaining life threshold needs to take into account factors such as the equipment's designed service life, the average remaining life of historical faults, the equipment's importance level, and the wind farm's operation and maintenance cost budget. The equipment's importance level is determined based on the equipment's core position in the wind power system, the scope of the fault's impact, and the maintenance cost. Core equipment is given a more stringent threshold standard. By combining historical data statistical analysis, the average remaining life of historical faults is updated regularly, and the remaining life threshold is adjusted to ensure the adaptability and rationality of the threshold.

[0038] The failure risk probability threshold setting needs to be based on the historical failure statistics and operation and maintenance cost analysis results of the wind farm. The failure risk probability threshold is divided into multiple interval levels. By statistically analyzing the failure occurrence probability and operation and maintenance cost loss corresponding to different risk probability intervals, three core threshold intervals of high, medium and low are determined. The high threshold corresponds to the risk level with a high failure occurrence probability and large operation and maintenance cost loss, while the low threshold corresponds to the risk level with a low failure occurrence probability and small impact on system operation.

[0039] Setting performance degradation thresholds requires combining the equipment's rated operating parameters, the performance degradation limits provided by the manufacturer, historical performance degradation data, and industry standard requirements. It also requires referencing the performance degradation patterns of the equipment at different operating stages to set phased critical thresholds for key performance parameters. This reflects the performance degradation process of the equipment from normal operation to failure, ensuring that relevant signals can be captured in a timely manner at the initial stage of performance degradation.

[0040] The warning status determination operation needs to establish a multi-dimensional threshold comparison mechanism, which comprehensively compares the predicted remaining service life of the equipment, the probability of failure risk, and the performance degradation value obtained from cross-validation with preset thresholds: when the predicted remaining service life of the equipment is lower than the remaining service life threshold, the probability of failure risk exceeds the high threshold, and the performance degradation value exceeds the critical threshold, it is determined to be an emergency warning status; when the probability of failure risk is in the middle threshold range, and the predicted remaining service life is higher than the remaining service life threshold, and the performance degradation value has not exceeded the critical threshold, it is determined to be a general warning status; when the probability of failure risk is lower than the low threshold, and neither the remaining service life nor the performance degradation value has triggered the corresponding threshold, it is determined to be a normal status.

[0041] In some embodiments, step 4 matches maintenance plans based on fault warning status, determines the frequency of routine inspections and key monitoring areas based on the cross-validation results of performance degradation thresholds, and matches maintenance measures with the predicted equipment status, including: Different maintenance plans are matched for different warning states. Emergency maintenance response is initiated in the emergency warning state, planned maintenance plan is formulated in the general warning state, and routine maintenance mode is maintained in the normal state. Based on the performance degradation trend, key monitoring parts and inspection frequency are determined, and targeted inspection items are specified.

[0042] Specifically, the emergency warning state maintenance plan requires the activation of the emergency maintenance response mechanism, prioritizing the dispatch of first-level maintenance resources, including professional maintenance teams, emergency spare parts, and dedicated maintenance equipment, to ensure arrival at the site in the shortest possible time. Based on the fault type classification results, a precise maintenance plan should be formulated, clearly defining the fault handling priority, key maintenance procedures, spare parts replacement list, and quality control standards. Maintenance work should be initiated within the specified time limit, and full-process quality monitoring should be implemented during the maintenance process to ensure that the maintenance procedures meet the technical requirements. After the maintenance is completed, a strict testing and verification process should be conducted to confirm that the equipment has returned to normal before it can be put back into operation.

[0043] General maintenance plans under early warning conditions should be developed in conjunction with the real-time power generation load of the wind farm and weather conditions. The plans should avoid peak electricity consumption periods and periods of severe weather, and the maintenance time window should be reasonably arranged. A maintenance plan that can be executed in the short term should be developed, and the maintenance content, maintenance process, expected goals and safety assurance measures should be clearly defined. The maintenance content should focus on potential fault risk areas, including component testing, performance calibration, replacement of worn parts and other preventive maintenance operations, so as to reduce the probability of fault risk to a safe range through maintenance.

[0044] Normal maintenance plans should maintain the conventional maintenance mode, optimizing routine inspection strategies based on cross-validation results of performance degradation thresholds. Inspection cycles and content should be adjusted according to equipment performance degradation trends and historical maintenance data to ensure the rational allocation of maintenance resources.

[0045] The marking of key monitoring areas requires extracting the performance degradation rate of each component from the performance degradation trend curve, establishing a degradation rate assessment standard, and marking components with degradation rates exceeding a preset threshold as key monitoring areas. Priority should be given to covering high-failure-prone components such as gearbox bearings, wind turbine blades, and generator stator windings, while dynamically adjusting the coverage based on equipment operating years, maintenance history, and environmental impact factors.

[0046] Adjusting the inspection frequency requires establishing a dynamic adjustment mechanism. The inspection frequency should be determined based on the proportion of key monitoring areas and the fluctuation range of the confidence interval for performance degradation thresholds. When the proportion of key monitoring areas exceeds the preset proportion, or when the fluctuation range of the confidence interval for performance degradation thresholds is large, the regular inspection cycle should be shortened and the inspection frequency increased. When the proportion of key monitoring areas is lower than the preset proportion and the performance degradation trend is stable, the inspection cycle should be maintained or appropriately extended.

[0047] Targeted inspection projects should be clearly defined, with specific inspection items developed for different types of key monitoring areas. Appropriate testing methods should be selected based on component fault characteristics and monitoring needs. For rotating machinery components, key inspection items include vibration spectrum analysis, temperature monitoring, and oil analysis. For structural components such as wind turbine blades, key inspection items include surface damage detection, strain testing, and lightning protection system testing. For electrical equipment, key inspection items include voltage and current harmonic analysis, insulation performance testing, and wiring tightness checks.

[0048] In some embodiments, step 5 involves collecting operational status data of the equipment after maintenance, and optimizing the three-level fusion mechanism and hybrid prediction model through maintenance effectiveness evaluation feedback, including: Continuously collect equipment operation status data after maintenance, generate a post-maintenance fusion feature set and input it into the hybrid prediction model to obtain the post-maintenance equipment status prediction results; calculate maintenance effect evaluation indicators, and optimize the three-level fusion mechanism and hybrid prediction model in a closed loop based on the evaluation results.

[0049] Specifically, post-maintenance data collection and analysis requires initiating a maintenance effectiveness tracking and monitoring process after the maintenance work is completed. This involves continuously collecting equipment operating status data, with the collection period covering the stable operation phase after equipment maintenance. The data collected after maintenance is processed according to the aforementioned three-level fusion mechanism to generate a post-maintenance fusion feature set. This set is then input into the trained hybrid prediction model to obtain prediction results such as the remaining service life of the equipment after maintenance, the probability of failure risk, and the performance degradation rate.

[0050] The calculation of maintenance effectiveness evaluation indicators requires the establishment of a maintenance effectiveness evaluation system. This involves comparing the predicted equipment condition before and after maintenance to calculate core evaluation indicators. These include the failure risk reduction rate, which measures the effectiveness of maintenance measures in reducing equipment failure risk; the performance degradation rate slowdown rate, which assesses the role of maintenance measures in delaying equipment performance degradation; and, in conjunction with indicators such as maintenance cost and downtime, a comprehensive evaluation of the economy and effectiveness of the maintenance plan.

[0051] Model and fusion mechanism optimization requires establishing a closed-loop optimization mechanism based on maintenance effectiveness evaluation results. If the maintenance effectiveness fails to meet the preset target, the confidence weights of the three-level fusion mechanism are adjusted, the feature extraction algorithm and fusion strategy are optimized, and the quality of the fused feature set is improved. Simultaneously, the structural parameters and training strategy of the hybrid prediction model are adjusted to enhance the model's ability to capture changes in equipment status. The model is periodically incrementally trained based on newly added operational data, maintenance records, and fault cases to update model parameters and ensure that the model's prediction accuracy continues to optimize with data accumulation.

[0052] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predictive maintenance of key wind power equipment based on multi-source data fusion, characterized in that, include: Collect operational status data, environmental perception data, and full lifecycle operation and maintenance data of key wind power equipment, construct a three-level fusion mechanism of raw data alignment, cross-domain feature extraction, and decision information synthesis, integrate heterogeneous data, and generate a fusion feature set; Based on the aforementioned fusion feature set, a hybrid prediction model is established to capture key dimensions and time-series dependencies that are strongly correlated with equipment failure in the fusion feature set, and outputs cross-validation results of equipment remaining service life, failure type, and performance degradation threshold. Based on the cross-validation results of the remaining service life of the equipment, the fault type, and the performance degradation threshold, the fault warning status is determined according to the remaining service life threshold, the fault risk probability threshold, and the performance degradation threshold. Based on the fault warning status matching maintenance plan, the frequency of routine inspections and key monitoring parts are determined based on the cross-validation results of performance degradation threshold, and maintenance measures are matched with the predicted equipment status.

2. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 1, characterized in that, The system collects operational status data, environmental perception data, and full lifecycle maintenance data from key wind power equipment. It then constructs a three-level fusion mechanism for raw data alignment, cross-domain feature extraction, and decision information synthesis. This mechanism integrates heterogeneous data to generate a fused feature set, including: The key wind power equipment includes wind turbine blades, gearbox, generator, main shaft and converter. The operating status data includes vibration, temperature, speed, oil quality, voltage and current data, which are collected in real time by Internet of Things sensors deployed in the bearings, housing and oil circuit of the equipment. The environmental sensing data includes wind speed, wind direction, air humidity, ambient temperature, and salt spray concentration data, which are collected synchronously through meteorological stations deployed in the wind farm and environmental monitoring sensors built into the equipment. The full lifecycle operation and maintenance data includes the time of failure, failure description, maintenance work order details, spare parts replacement information, equipment factory parameters and previous maintenance records, which are extracted from the wind farm operation and maintenance management system.

3. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 1, characterized in that, A three-level fusion mechanism is constructed, consisting of raw data alignment, cross-domain feature extraction, and decision information synthesis, integrating heterogeneous data to obtain a fused feature set, including: Based on the acquisition frequency, measurement accuracy, and transmission stability of each data source, confidence weights are assigned. Based on these confidence weights, the original data of different dimensions are aligned in time and matched in spatial location. A data missing rate threshold is set, and invalid data segments with missing rates exceeding the threshold are removed to obtain a preliminary aligned dataset. Local spatiotemporal features and temporal dependency features are extracted from the preliminary aligned dataset. The two types of features are concatenated to generate a high-dimensional feature vector. The high-dimensional feature vector is then subjected to dimensionality reduction processing to remove redundant information and obtain a simplified feature vector. Fault-related features are extracted from the full lifecycle operation and maintenance data, including the trend of changes in operating parameters before the fault and the correspondence between historical faults and operating status. The simplified feature vector is fused with the fault-related features, the feature credibility is calculated, and a fused feature set with both completeness and relevance is generated.

4. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 1, characterized in that, A hybrid prediction model is established based on the fused feature set, including: The hybrid prediction model strengthens the weights of feature dimensions that are strongly correlated with equipment failures, highlights the impact of key information on the prediction results, adjusts the learning rate during model training, and sets an early stopping mechanism to avoid overfitting. Using historical equipment fault data and corresponding fusion feature sets as training samples, the model is divided into training, validation, and test sets, and the model parameters are iteratively optimized.

5. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 1, characterized in that, The cross-validation results of the remaining useful life, failure type, and performance degradation threshold of the output device include: The trained hybrid prediction model outputs the predicted remaining useful life of the equipment, the failure type classification results, and the performance degradation trend curve. Cross-validation is used to verify the above results, evaluate the accuracy of remaining useful life prediction and the failure type classification effect, and determine the confidence interval of the performance degradation critical value. The results of the integrated verification are used to form cross-verification results for the remaining service life of the equipment, failure type, and performance degradation threshold.

6. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 1, characterized in that, The fault warning status is determined based on the cross-validation results, including: The remaining life threshold is dynamically updated based on the equipment's designed service life, the average remaining life of historical faults, and the equipment's importance level. The fault risk probability threshold is divided into interval ranges and determined based on the wind farm's historical fault statistics and operation and maintenance costs. The performance degradation threshold is determined based on the equipment's rated operating parameters and the performance degradation limit value provided by the manufacturer. When the predicted remaining service life is lower than the remaining service life threshold, the failure risk probability exceeds the high threshold, and the performance degradation value exceeds the critical threshold, it is determined to be an emergency warning state. When the failure risk probability is within the threshold range, and the remaining service life and performance degradation value do not trigger the corresponding threshold, it is determined to be a general warning state. When the failure risk probability is lower than the low threshold, and the remaining service life and performance degradation value do not trigger the corresponding threshold, it is determined to be a normal state.

7. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 6, characterized in that, Based on the fault warning status, a corresponding maintenance plan is matched, including: If an emergency warning state is determined, an emergency maintenance plan will be matched, and first-level maintenance resources will be prioritized. Based on the priority of fault handling, key maintenance procedures and the list of necessary spare parts, maintenance work will be initiated within the specified time limit. If the situation is determined to be a general warning state, a planned maintenance scheme will be matched, and a short-term maintenance plan will be formulated based on the real-time power generation load of the wind farm and weather conditions to obtain the maintenance content and expected results. If the condition is determined to be normal, maintain the regular maintenance mode and optimize the regular inspection strategy based on the cross-validation results of the performance degradation threshold.

8. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 7, characterized in that, Based on the cross-validation results of the performance degradation threshold, the frequency of routine inspections and key monitoring areas are determined, including: extracting the performance degradation rate of each component of the equipment from the performance degradation trend curve, marking the components with faster degradation rates as key monitoring areas, adjusting the frequency of routine inspections according to the proportion of key monitoring areas and the fluctuation range of the confidence interval of the performance degradation threshold, and identifying targeted inspection items such as vibration detection, temperature monitoring, and oil analysis for key monitoring areas.

9. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 3, characterized in that, The raw data alignment stage also includes data preprocessing operations, which normalize heterogeneous data of different dimensions, unify the data range, identify outliers in the data, use appropriate correction methods according to the degree of deviation of outliers, and retain the abrupt data characteristics corresponding to the precursory signs of equipment failure.

10. The method for predictive maintenance of key wind power equipment based on multi-source data fusion according to claim 1, characterized in that, The method further includes a maintenance effect feedback optimization step: collecting the operating status data of the equipment after maintenance and generating a fusion feature set after maintenance; inputting the fusion feature set after maintenance into a hybrid prediction model to obtain the equipment status prediction result after maintenance; comparing the equipment status prediction results before and after maintenance, and calculating the failure risk reduction rate and performance degradation rate mitigation rate evaluation indicators. The confidence weights of the three-level fusion mechanism and the parameters of the hybrid prediction model are adjusted based on the feedback from the evaluation indicators of failure risk reduction rate and performance degradation rate slowdown rate, and the model is incrementally trained regularly based on new data.

Images