Wind turbine generator fault early warning method and system based on multi-model combination

Abstract

The invention relates to the technical field of power generation, and discloses a wind turbine generator fault early warning method and system based on multi-model combination, and the early warning method comprises the steps: 1, data collection; step 2, establishing an early warning model; 3, real-time monitoring and early warning are carried out; step 4, analyzing a fault source; step 5, estimating fault existence time; and step 6, fault subsequent condition prediction. According to a fault source analysis result, historical fault data and current operation data are combined through the system, the fault existence time is estimated, the prediction model is utilized, the system can not only predict the fault expansion speed, but also predict the influence degree of the fault on the unit performance, and therefore a more effective coping strategy is formulated, and the unit performance is improved. The beneficial effects of early warning the fault of the wind turbine generator and analyzing the fault source are achieved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power generation, in particular to a wind turbine fault early warning method and system based on multiple model combination. BACKGROUND

[0002] The wind turbine fault early warning method refers to a method for detecting whether a potential fault exists based on current and historical operation data of a wind turbine. According to the data source, the method can be divided into a fault early warning method based on vibration data, a fault early warning method based on SCADA data and a fault early warning method based on other signal sources (such as ultrasonic devices, infrared imagers, etc.). Among them, because the purchase and installation cost of vibration sensors, ultrasonic devices, infrared imagers and other equipment is high, the fault early warning method based on such data is limited in actual application. Relatively speaking, the SCADA system has been widely used in wind farms and can provide relatively complete historical operation data of the unit, so the fault early warning method based on SCADA data is more suitable for engineering application and has become the mainstream method for current wind turbine fault early warning. According to the search, a wind turbine fault early warning method based on multiple model combination is found in CN 111798650 A. The method first collects a large amount of SCADA historical data of wind speed, pitch angle, torque setting, angle between central axis and wind direction and power, adopts multiple neural network methods, and respectively establishes a full-condition power prediction sub-model of wind turbine start-stop. Secondly, based on the data sliding window, the power prediction sub-model is used to calculate the power prediction value time sequence and the operation state index. Thirdly, the kernel density estimation method is used to calculate the probability density function of the operation state index, so as to determine the basic probability distribution function of the fault early warning proposition. Finally, the evidence theory is used to fuse the early warning results of multiple power prediction sub-models, and the fusion results are compared with the set early warning threshold to obtain the fault early warning conclusion. The application can realize the full-condition fault early warning of the wind turbine and effectively improve the accuracy of the fault early warning result.

[0003] Although the prior art can early warn the wind turbine fault and prompt the working condition of the wind turbine, since the implementation principle is to model the collected data and then predict by the evidence theory, the prediction result can only indicate that there is a fault risk at the place, but cannot know where the fault originates from and cannot explicitly know the time of the fault and the time of further expansion. SUMMARY

[0004] The technical problem solved by the application is: In view of the deficiencies of the prior art, the present application provides a wind turbine fault early warning method and system based on multi-model combination, which has the advantages of not only being able to early warn wind turbine faults but also being able to analyze fault sources, being able to calculate the existence time of the fault and subsequent conditions after the fault is discovered, and the like, and solves the above technical problems.

[0005] Technical scheme: To achieve the above object, the present application provides the following technical scheme: a wind turbine fault early warning method based on multi-model combination, wherein the early warning method steps are as follows: Step one, data collection: collect various operation data of the wind turbine, such as wind speed, power output, temperature and vibration; Step two, establishment of an early warning model: use a random forest to establish a wind turbine fault early warning model, train the model to identify the difference between normal operation state and abnormal state, and set an early warning threshold value, which is the operation data ; Step three, real-time monitoring and early warning: input real-time data into the early warning model, continuously monitor the operation state of the wind turbine, and once the model output exceeds the early warning threshold value, immediately trigger the early warning mechanism; Step four, fault source analysis: when the early warning is triggered, start the fault source analysis program, use a Bayesian network to combine historical data and real-time data to analyze the specific reason for the early warning; Step five, estimation of the existence time of the fault: according to the result of the fault source analysis, combine historical fault data and current operation data to estimate the existence time of the fault; Step six, prediction of the subsequent conditions of the fault: use a prediction model to combine current fault data and historical fault evolution data to predict the subsequent development of the fault, including the speed of the expansion of the fault and the degree of influence on the performance of the unit.

[0006] A wind turbine fault early warning system based on multi-model combination, wherein the early warning system is used for the early warning method, and the early warning system is composed of a data collection unit, a real-time monitoring and early warning unit, an analysis unit and a prediction unit; The data collection unit is used to collect wind speed, power output, temperature and vibration operation data of the wind turbine: The real-time monitoring and early warning unit is used to continuously monitor the operation state of the wind turbine, and once the output exceeds the early warning threshold value, immediately trigger the early warning mechanism; The analysis unit is used to analyze the specific reason for the early warning and estimate the existence time of the fault; The prediction unit is used to predict the subsequent development of the fault according to the result of the fault source analysis, including the speed of the expansion of the fault and the degree of influence on the performance of the unit.

[0007] Preferably, the expression of the data collection unit is: Sensor data collection: Wind speed sensor: ,in It's the wind speed value. It is time. This is the actual wind speed; Power output sensor: ,in This is the power output value. It is time. It is the power converted from theoretical wind speed; Temperature sensor: Where T is the temperature value, It is time. This is the actual temperature; Vibration sensor: Where B is the vibration value, It is time. It is actual vibration; Multi-source data fusion: in It is the merged data. , , and These are the normalized wind speed, power, temperature, and vibration data, respectively. It is a fusion function.

[0008] Preferably, the multi-source data fusion uses probabilistic fusion calculation, expressed as: in: It is the fused data, which comprehensively considers information on wind speed, power, temperature, and vibration; It is the normalized first Data sources in time The value; The corresponding weights satisfy... .

[0009] Preferably, the early warning model expression of the real-time monitoring and early warning unit is: Suppose a dataset containing multiple features: wind speed, power, temperature, and vibration. ,in It is the first The feature vector of each sample These are the corresponding labels, indicating normal operating status: label 0 or abnormal status: label 1; The prediction function of a random forest is expressed as: Using the dataset to train a random forest model, the model learns how to predict the label from the feature vector ; The warning threshold is set to be the ; The mean and standard deviation of the probability value output by the model in normal operation are denoted as and respectively, so the warning threshold is set as: .

[0010] Preferably, the real-time monitoring and warning unit triggers the warning mechanism expression as: Warning model: Using the established random forest model for prediction, the probability value output by the model is , indicating the probability of the wind turbine being in an abnormal state at time ; The warning threshold is set as: The warning threshold is set to be the mean of the probability in normal operation plus or minus a threshold range: ; Trigger the warning mechanism: In real-time monitoring, for each time point , calculate the predicted abnormal probability If , trigger the warning mechanism, indicating that the wind turbine is in an abnormal state and needs further inspection and maintenance. Where is the prediction function of the random forest model, obtained by the integration of multiple decision trees. The triggering condition expression of the warning mechanism is: .

[0011] Preferably, the analysis expression of the analysis unit is: Define variables and events: Let a set of variables represent various factors leading to wind turbine failure: failure of specific components or environmental factors.

[0012] Each variable has multiple states: , where represents the variable The One state; Prior probability and conditional probability: Prior probability This indicates that, in the absence of other information, the variable... In state The probability of; Conditional probability Indicates at a given parent node In the state, variables In state The probability of; Fault source reasoning, using the variable elimination method, calculates the posterior probability of each fault factor: in This is evidence provided by real-time data; Identify the source of the fault: in, Given factors In state Evidence observed at the time The probability, It is the prior probability, and It is the total probability of observing evidence.

[0013] Preferably, the expression for estimating the fault existence time in the analysis unit is: This indicates the duration of the fault, which needs to be estimated. Historical fault data is available, including the fault start time. and end time and current real-time data, such as wind speed ,power ,temperature ,vibration Based on the analysis of the fault source, the time when the fault started was determined. and current time : in: It is the current time point. It is the time point at which the fault began.

[0014] Preferably, the expression for the rate at which the prediction unit predicts fault propagation is: in, Indicates the current stage of the fault's development; represents the fault development degree at future time .

[0015] Preferably, the prediction unit predicts the degree of influence on the unit performance, which is expressed as: unit performance is influenced by the fault development degree , a functional relationship is established: wherein the function describes the specific influence of the fault development degree on the unit performance; The specific expression is: Herein is the predicted fault development degree at future time , and is the predicted value of the unit performance at the corresponding time.

[0016] Compared with the prior art, the present application provides a wind turbine fault early warning method and system based on multi-model combination, which has the following beneficial effects: 1. The system collects various operating data of the wind turbine: wind speed, power output, temperature and vibration, and establishes an early warning mechanism using prediction models such as random forest. Once the operating data is monitored to be abnormal and exceeds the warning threshold, the system triggers an early warning, indicating that there is a fault or abnormal situation. When the early warning is triggered, the system enters the fault source analysis stage, and through the Bayesian network combined with real-time data and historical data, the system infers the specific cause of the fault. Not only can it quickly locate the problem, but also can help operators take appropriate maintenance measures in the early stage to prevent the fault from further deteriorating. According to the results of the fault source analysis, combined with historical fault data and current operating data, the system estimates the existence time of the fault, and using the prediction model, the system can not only predict the expansion speed of the fault, but also predict the degree of influence of the fault on the unit performance, so as to develop more effective coping strategies, achieving the beneficial effect of not only being able to early warn the wind turbine fault but also being able to analyze the fault source.

[0017] 2. The system uses prediction models to predict the subsequent development of the fault by combining current fault data and historical fault evolution data. This includes the expansion speed of the fault and the specific degree of influence of the fault on the wind turbine performance. The prediction formula for the expansion speed of the fault is: wherein represents the fault development degree at the current time, represents the fault development degree at future time , and the prediction of the influence of the fault on the unit performance is through the function The method is carried out, wherein The specific influence of the fault development degree on the unit performance is described, and the beneficial effects of being able to calculate the existence time of the fault and the subsequent situation after the fault is discovered by the early warning are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 It is a flow chart of the method of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] Please refer to Figure 1 A wind turbine fault early warning method based on multi-model combination, the early warning method steps are: Step one, data collection: collect various operation data of the wind turbine: wind speed, power output, temperature and vibration; Step two, establish early warning model: use random forest to establish wind turbine fault early warning model, train the model to identify the difference between normal operation state and abnormal state, and set the early warning threshold, the threshold is the operation data ; Step three, real-time monitoring and early warning: input real-time data into the early warning model, continuously monitor the operation state of the wind turbine, once the model output exceeds the early warning threshold, immediately trigger the early warning mechanism; Step four, fault source analysis: when the early warning is triggered, start the fault source analysis program, use Bayesian network to combine historical data and real-time data to analyze the specific reasons for the early warning; Step five, fault existence time estimation: according to the results of fault source analysis, combine historical fault data and current operation data to estimate the existence time of the fault; Step six, fault subsequent situation prediction: use the prediction model to combine the current fault data and the historical fault evolution data to predict the subsequent development of the fault, including the speed of fault expansion and the degree of influence on the unit performance.

[0021] A wind turbine fault early warning system based on multi-model combination, the early warning system is used for the early warning method, the early warning system is composed of a data acquisition unit, a real-time monitoring and early warning unit, an analysis unit and a prediction unit; The data acquisition unit is used to collect the wind speed, power output, temperature and vibration operation data of the wind turbine: The real-time monitoring and early warning unit is used to continuously monitor the operating state of the wind turbine, and outputs an early warning threshold value, which immediately triggers the early warning mechanism; The analysis unit is used to analyze the specific reasons for the early warning and estimate the fault existence time; The prediction unit is used to predict the subsequent development of the fault, including the speed of fault expansion and the degree of impact on the performance of the unit, based on the results of the fault source analysis.

[0022] The data acquisition unit can integrate data from different sensors through multi-source data fusion, providing more comprehensive and integrated data information. Through normalization processing and fusion functions, various data are unified into a standardized range, facilitating comparison and analysis. By fusing multiple data sources, the limitations or errors of a single sensor are compensated for, improving the accuracy and reliability of the data. Comprehensive consideration of multiple aspects of data more comprehensively reflects the actual situation. Multi-source data fusion can improve the credibility and stability of the data, reduce the risk brought by a single data source, and reduce data fluctuations and noise by integrating multiple data, improving the credibility of the data. Fusion of multi-source data mining can provide more valuable data support for subsequent analysis and decision-making. Comprehensive data helps users better understand system operation and trends.

[0023] The real-time monitoring and early warning unit uses prediction models such as random forests to monitor and warn of abnormal states, enabling rapid and accurate identification of abnormal conditions in wind turbines. Multi-model combination improves the accuracy and stability of the prediction, reducing false positives and false negatives. The system monitors the operating state of the wind turbine in real time, detects abnormal conditions in a timely manner, and triggers the early warning mechanism through the set early warning threshold value, quickly responding to abnormal conditions, which helps to take timely measures for maintenance and repair, reducing losses. The early warning threshold value is set flexibly according to actual conditions to adapt to different operating environments and requirements. By setting an appropriate threshold range, the sensitivity and accuracy of the early warning are balanced. The triggering of the early warning mechanism is based on the comparison of the abnormal probability of the model output and the early warning threshold value, realizing an automated wind turbine fault early warning system that reduces the need for human intervention and improves the efficiency of monitoring and early warning. The system considers multiple features (wind speed, power, temperature, and vibration) and uses prediction models such as random forests for comprehensive analysis and monitoring, improving the comprehensive monitoring and early warning capabilities of the wind turbine operating state.

[0024] The analysis unit carries out inference analysis on different fault sources by using prior probability and conditional probability, combines the evidence provided by real-time data, comprehensively considers the influence of various factors on the fault of the wind turbine, improves the accuracy and comprehensiveness of the analysis of the fault source, carries out inference analysis through variable elimination method, flexibly adjusts and updates the inference process of the fault source according to the actual situation, adapts to the fault diagnosis requirements under different conditions, improves the applicability and practicability of the system, estimates the existing time length of the fault relatively accurately according to the historical fault data and real-time monitoring data combined with the result of the fault source analysis, helps to quickly respond and arrange maintenance work, reduces the influence and loss caused by the fault, more accurately judges the duration of the fault through the estimation of the existing time of the fault, helps to predict the fault condition and maintenance requirement of the wind turbine, improves the effect and response speed of the early warning system, and further ensures the safe and stable operation of the wind turbine.

[0025] The prediction unit can relatively accurately predict the development degree of the fault and the value of the unit performance at the future time by establishing the functional relationship between the fault expansion speed and the unit performance, which helps to discover potential faults in advance and take corresponding prevention and maintenance measures, and the prediction of the influence degree of the fault expansion speed and the unit performance helps the system to realize timely early warning of future faults, improves the real-time monitoring and management ability of the running state of the wind turbine, and based on the prediction of the development degree of the fault and the unit performance, the system can provide prediction of the unit running condition at the future time, which helps to optimize the operation and maintenance decision and improve the health management level of the system.

[0026] Although the embodiments of the present application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for early warning of wind turbine faults based on multi-model combination, characterized in that, The steps of the early warning method are as follows: Step 1: Data Acquisition: Collect various operating data of the wind turbine: wind speed, power output, temperature, and vibration; Step 2: Establish an early warning model: Utilize random forests to build a wind turbine fault early warning model. Train the model to identify the differences between normal and abnormal states, and set an early warning threshold, which is determined by the operating data. ; Step 3: Real-time monitoring and early warning: Input real-time data into the early warning model to continuously monitor the operating status of the wind turbine. Once the model output exceeds the early warning threshold, the early warning mechanism will be triggered immediately. Step 4: Fault Source Analysis: When the warning is triggered, the fault source analysis program is started. Using a Bayesian network, combined with historical and real-time data, the specific cause of the warning is analyzed. Step 5: Fault duration estimation: Based on the results of the fault source analysis, combined with historical fault data and current operating data, estimate the duration of the fault. Step Six: Predicting the Subsequent Development of the Fault: Using a predictive model, combined with current fault data and historical fault evolution data, predict the subsequent development of the fault, including the speed of fault expansion and the degree of impact on unit performance.

2. The wind turbine fault early warning system based on multi-model combination according to claim 1, characterized in that, The early warning system is used for early warning methods, and the early warning system consists of a data acquisition unit, a real-time monitoring and early warning unit, an analysis unit, and a prediction unit. The data acquisition unit is used to collect wind speed, power output, temperature, and vibration operation data of the wind turbine. The real-time monitoring and early warning unit is used to continuously monitor the operating status of the wind turbine unit. If the output exceeds the early warning threshold, the early warning mechanism will be triggered immediately. The analysis unit is used to analyze the specific reasons that cause the warning and estimate the duration of the fault. The prediction unit is used to predict the subsequent development of a fault based on the results of fault source analysis, including the speed of fault expansion and the degree of impact on unit performance.

3. The wind turbine fault early warning system based on multi-model combination according to claim 2, characterized in that: The expression for the data acquisition unit is: Sensor data acquisition: Wind speed sensor: ,in It's the wind speed value. It is time. This is the actual wind speed; Power output sensor: ,in This is the power output value. It is time. It is the power converted from theoretical wind speed; Temperature sensor: Where T is the temperature value, It is time. This is the actual temperature; Vibration sensor: Where B is the vibration value, It is time. It is actual vibration; Multi-source data fusion: in It is the merged data. , , and These are the normalized wind speed, power, temperature, and vibration data, respectively. It is a fusion function.

4. The wind turbine fault early warning system based on multi-model combination according to claim 3, characterized in that: The multi-source data fusion uses probabilistic fusion calculation, expressed as: in: It is the fused data, which comprehensively considers information on wind speed, power, temperature, and vibration; It is the normalized first Data sources in time The value; The corresponding weights satisfy... .

5. A wind turbine fault early warning system based on multi-model combination according to claim 4, characterized in that: The early warning model expression for the real-time monitoring and early warning unit is: Suppose a dataset containing multiple features: wind speed, power, temperature, and vibration. ,in It is the first The feature vector of each sample These are the corresponding labels, indicating normal operating status: label 0 or abnormal status: label 1; The prediction function of a random forest is expressed as: Using datasets To train a random forest model, the model learns how to use feature vectors To predict tags ; The warning threshold is set based on the running data. ; set up and These are the mean and standard deviation of the probability values ​​output by the model under normal operating conditions. Therefore, the warning threshold is set as follows: 。 6. A wind turbine fault early warning system based on multi-model combination according to claim 5, characterized in that: The expression for the real-time monitoring and early warning unit to trigger the early warning mechanism is: Early warning model: The established random forest model is used for prediction, and the probability value output by the model is... , indicating time The probability of a wind turbine being in an abnormal state; Warning threshold setting: Warning threshold The average probability of being set to normal operating state Add or subtract a threshold range: ; Triggering the early warning mechanism: During real-time monitoring, for each time point Calculate the predicted anomaly probability if If this occurs, an early warning mechanism will be triggered, indicating that the wind turbine has an abnormal condition and requires further inspection and maintenance. in It is the prediction function of the random forest model, obtained by ensemble of multiple decision trees; The trigger condition expression for the early warning mechanism is: 。 7. A wind turbine fault early warning system based on multi-model combination according to claim 2, characterized in that: The fault source analysis expression in the analysis unit is: Define variables and events: Let a set of variables This represents various factors that can lead to wind turbine failures: failure of specific components or environmental factors. Each variable There are multiple states: ,in Representing variables The One state; Prior probability and conditional probability: Prior probability This indicates that, in the absence of other information, the variable... In state The probability of; Conditional probability Indicates at a given parent node In the state, variables In state The probability of; Fault source reasoning, using the variable elimination method, calculates the posterior probability of each fault factor: in This is evidence provided by real-time data; Identify the source of the fault: in, Given factors In state Evidence observed at the time The probability, It is the prior probability, and It is the total probability of observing evidence.

8. A wind turbine fault early warning system based on multi-model combination according to claim 2, characterized in that: The expression for estimating the fault existence time in the analysis unit is: This indicates the duration of the fault, which needs to be estimated. Historical fault data is available, including the fault start time. and end time and current real-time data, such as wind speed ,power ,temperature ,vibration Based on the analysis of the fault source, the time when the fault started was determined. and current time : in: It is the current time point. It is the time point at which the fault began.

9. A wind turbine fault early warning system based on multi-model combination according to claim 2, characterized in that: The expression for the rate at which the prediction unit predicts fault propagation is: in, Indicates the current stage of the fault's development; Indicates future time The degree of development of the fault.

10. A wind turbine fault early warning system based on multi-model combination according to claim 9, characterized in that: The expression for the degree of impact of the prediction unit on the unit performance is as follows: Unit performance Affected by the degree of fault development The influence of this relationship establishes a functional relationship: Where the function It describes the specific ways in which the degree of fault development affects unit performance; The specific expression is: here It is a predicted future moment. The degree of fault development, and This is the predicted value of the unit's performance at the corresponding time.

Images