Wind turbine fault prediction system based on machine learning

Abstract

The present invention relates to the technical field of wind turbine fault prediction, and specifically to a wind turbine fault prediction system based on machine learning, comprising: a deviation data detection module for determining the wind turbine's wind accuracy and outputting the deviation data detection result; a deviation change trend analysis module for identifying the deviation change development trend and predicting future deviation increases; a device real-time data acquisition and synchronization module for acquiring real-time data of devices related to the yaw system; and a device and deviation relationship analysis module for generating a judgment result of device faults and determining the specific device or component that caused the small deviation. The present invention uses multi-level device data analysis to deeply analyze the specific device that caused the deviation, accurately locate the source of the fault, ensure rapid adjustment of the yaw system's working state, prevent further aggravation of the deviation, and ensure that the wind turbine always maintains the best match with the wind direction, thereby improving power generation efficiency and reducing energy loss caused by deviation.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine fault prediction, and in particular to a wind turbine fault prediction system based on machine learning. Background Art

[0002] The efficiency of a wind turbine is closely related to its accuracy in wind direction during operation. The wind turbine continuously adjusts the direction of the nacelle through the yaw system to ensure that the wind rotor always faces the wind direction, maximizing the capture of wind energy and improving power generation efficiency. If the yaw system deviates from the wind direction, the wind rotor will not be able to face the wind direction, resulting in a decrease in power generation efficiency and may cause additional mechanical stress to the system, accelerate equipment wear, and increase maintenance costs.

[0003] Currently, traditional yaw systems rely on fixed rules and preset parameters to control the yaw angle. These methods can only handle large deviations and are unable to accurately capture and address small deviations. Over the long-term operation of a wind turbine, the accumulation of small deviations can lead to serious energy efficiency losses and mechanical problems. Existing monitoring systems often lack comprehensive analysis of device parameters and deviation data in the yaw system, making it difficult to identify and correct potential equipment failures in real time. This not only limits system maintenance efficiency but can also lead to major failures.

[0004] In addition, existing yaw system fault detection mainly relies on preset thresholds and rules, and lacks in-depth analysis of complex, multi-dimensional equipment data. Especially when dealing with the synergy between equipment, it is often difficult to make accurate judgments, which makes fault prediction delayed and results in problems not being discovered and resolved in the early stages. Summary of the Invention

[0005] The present invention provides a wind turbine fault prediction system based on machine learning.

[0006] The machine learning-based wind turbine fault prediction system includes:

[0007] Deviation data detection module: collects yaw angle and wind direction sensor data of the yaw system in real time, calculates the deviation between the yaw angle and wind direction, detects the matching degree between the yaw angle and wind direction, and determines the wind accuracy of the wind turbine. The data detection module detects small deviations in real time based on a time series regression algorithm and outputs the deviation data detection results;

[0008] Deviation change trend analysis module: Analyzes the trend of small deviations and captures the cumulative effect of small deviations in different time periods. The deviation change trend analysis module uses a time series model to identify the trend of deviation changes and predict future deviation increases.

[0009] Equipment real-time data acquisition and synchronization module: used to collect real-time data of equipment related to the yaw system, including the current, voltage, and torque of the yaw motor, and the pressure and temperature of the hydraulic system. It ensures that all equipment data is collected synchronously with the yaw angle and wind direction sensor data, and performs data standardization processing;

[0010] Equipment and deviation relationship analysis module: Based on the deviation change development trend, the relationship between the real-time data of each device and the small deviation is analyzed. The equipment and deviation relationship analysis module uses multivariate regression analysis and correlation analysis algorithms to identify which equipment parameter fluctuations show a significant correlation with the yaw angle deviation, generate equipment failure judgment results, and determine the specific equipment or component that causes the small deviation.

[0011] Optionally, the deviation data detection module specifically includes:

[0012] Yaw angle acquisition unit: connected to the yaw sensor to obtain the yaw angle data of the wind turbine nacelle in real time. The yaw angle data is collected at a set frequency; the collection frequency is set to between 1Hz and 5Hz, that is, 1 to 5 data are collected per second;

[0013] Wind direction data acquisition unit: This unit uses a wind direction sensor to acquire real-time wind direction data in the area where the wind turbine is located. The wind direction data is collected synchronously with the yaw angle data to ensure that both are collected on the same time basis. The data smoothing method is used to eliminate the interference of instantaneous fluctuations in the wind direction data.

[0014] Deviation calculation unit: The deviation calculation unit receives yaw angle data and wind direction data, and determines the wind turbine's wind accuracy by calculating the angle between the yaw angle and the wind direction. If the angle exceeds a preset threshold (3°), it is considered that there is a deviation and the deviation data detection result is output;

[0015] Small deviation detection unit: Based on the time series regression algorithm, the small deviation detection unit analyzes multiple sets of deviation data in a short period of time, identifies persistent small deviations, and outputs the small deviation results to the deviation change trend analysis module in real time to predict the deviation cumulative effect and trend.

[0016] Optionally, the deviation change trend analysis module specifically includes:

[0017] Historical deviation data storage unit: used to store historical deviation data generated by the deviation data detection module. The data includes the difference between the yaw angle and the wind direction angle and the corresponding timestamp. The historical data is stored in a set time window to ensure that it contains enough samples for trend analysis;

[0018] Time series data processing unit: Receives historical deviation data, arranges it in chronological order, builds a time series model of deviation changes, and analyzes the trend of deviation changes in different periods. The time series model adopts the ARIMA model;

[0019] Small deviation accumulation detection unit: Identifies small deviations that persist over multiple time periods and captures the cumulative effect of deviations through time series models. If the deviation does not decrease or shows an increasing trend over multiple consecutive time periods, it is considered to be cumulative, indicating that system fine-tuning may be required.

[0020] Future Deviation Prediction Unit: Based on the prediction capability of the time series model, it uses current and historical deviation data to predict deviation changes in future periods.

[0021] The ARIMA model analyzes and predicts time series data. It captures the trend and seasonal effects of time series through three parts: autoregression (AR), differencing (I), and moving average (MA). It is suitable for analyzing the accumulated deviations of wind turbines and predicting future trends.

[0022] Optionally, the ARIMA model reviews deviation data over a period of time and analyzes data change patterns. When the deviation of the yaw angle and wind direction is collected, the model identifies whether the current deviation is gradually increasing, decreasing, or remaining stable based on the deviation changes over multiple consecutive time periods. This trend analysis can help determine whether the yaw system is entering an abnormal state.

[0023] The variation trends of the deviation data include:

[0024] Short-term changes: The ARIMA model uses data from recent periods to determine whether there are short-term fluctuations in the deviation. If the deviation fluctuates significantly in the short term, it may indicate an unstable system response.

[0025] Long-term trends: By integrating data over a long period of time, the ARIMA model identifies the long-term trend of deviations and analyzes whether there is a continuous accumulation or slow change in deviations.

[0026] The ARIMA model analyzes the deviation data and captures the phenomenon of deviation accumulation. The cumulative effect is that the deviation does not decrease over multiple time periods and shows a trend of gradual increase. The ARIMA model determines whether the deviation is accumulating by identifying trend changes in the time period.

[0027] Optionally, the ARIMA model analyzes current and historical deviation data to infer future deviation trends. The prediction process includes:

[0028] Impact of current deviation data: The current deviation value affects short-term performance. If the current deviation is large, the system is expected to maintain this deviation in the short term.

[0029] Historical deviation trend: By identifying the deviation pattern (such as gradual increase, fluctuation, stability, etc.) over a period of time in the past, it is inferred that future deviations will continue the deviation pattern.

[0030] Optionally, the device real-time data acquisition and synchronization module specifically includes:

[0031] Motor data acquisition unit: collects the operating parameters of the yaw motor in real time, including motor current, voltage and torque signals. The motor data acquisition unit collects these data at a set sampling frequency by connecting to the motor sensor in the yaw system;

[0032] Hydraulic system data acquisition unit: collects real-time data from the yaw system hydraulic components, including pressure and temperature signals of the hydraulic system. Hydraulic data is obtained through hydraulic sensors and changes in the hydraulic system operation are simultaneously recorded to ensure that the pressure and temperature data can match the yaw motor data;

[0033] Data synchronization unit: This unit ensures the synchronous collection of the yaw motor's current, voltage, and torque data, the hydraulic system's pressure and temperature data, and the yaw angle and wind direction sensor data. By matching timestamps and aligning timing, it unifies the time base of device data, ensuring that data from each device can be correlated and analyzed in real time.

[0034] Data standardization processing unit: used to standardize all collected equipment data to ensure that data of different units and magnitudes can be effectively compared and analyzed. The standardization processing is based on the Z-score standardization method to convert the equipment data into a dimensionless form so that the subsequent analysis module can perform correlation analysis between equipment status and deviation.

[0035] Optionally, the device and deviation relationship analysis module specifically includes the following units:

[0036] Correlation analysis unit: Receives real-time data from various devices, including current, voltage, and torque data of the yaw motor, pressure and temperature data of the hydraulic system, and deviation data of the yaw angle and wind direction. Using a correlation analysis algorithm, it preliminarily selects device parameters that show a significant linear correlation with the deviation data.

[0037] Multivariate regression analysis unit: Based on the highly correlated equipment parameters screened by the correlation analysis unit, multivariate regression analysis is performed to refine the relationship between equipment parameters and deviations. Multivariate regression analysis quantifies the weight of the impact of each equipment parameter on deviation changes and establishes a linear regression model to evaluate the combined impact of multiple equipment parameters on small deviations. Through regression analysis, one or more equipment parameters are determined to have the greatest contribution to the deviation change, thereby accurately identifying potential faulty equipment or components.

[0038] Equipment failure judgment unit: Based on the results of the multivariate regression analysis unit, the final judgment of equipment failure is generated. If the fluctuation of a certain equipment parameter has a large impact on the deviation and is significantly correlated with the cumulative effect of small deviations, the equipment is marked as a potential fault source and fault warning information is generated.

[0039] Optionally, the correlation analysis unit uses a Pearson correlation coefficient to measure the linear relationship between each device parameter and the yaw angle deviation. The Pearson correlation coefficient compares the operating parameter of each device with the deviation between the yaw angle and the wind direction, specifically including:

[0040] Calculate differences: The correlation analysis unit calculates the overall trend of equipment operating parameters and deviation data;

[0041] Measuring correlation: Calculate the degree of correlation between the equipment operating parameters and the deviation. The Pearson correlation coefficient outputs whether the two variables change at the same time. If the parameter change of a device is positively correlated with the deviation (that is, when the device parameter increases, the deviation also increases, or when the device parameter decreases, the deviation also decreases), it is considered that the device may have a strong influence on the deviation. If the correlation coefficient is negative, it indicates the opposite relationship (that is, when one increases, the other decreases). If the correlation coefficient is close to zero, it means that there is almost no linear relationship between the two.

[0042] Preliminary equipment screening: By calculating the correlation between multiple equipment parameters and deviations, the equipment operating parameters that have a significant impact on the deviation can be quickly identified.

[0043] Optionally, the linear regression model evaluates the impact of multiple device parameters on the deviation change simultaneously, and the expression is:

[0044] Δθ t =β0+β1x 1,t +β2x 2,t +…+β i x i,t +∈ t ,in;

[0045] Δθ t Indicates the deviation value at time t, which is the difference between the yaw angle and the wind direction angle, x 1,t ,x 2,t ,…,x i,t Indicates the parameter values ​​of each device collected at time t, which respectively represent the real-time data of the device related to the yaw system, including the current x of the yaw motor 1,t , voltage x 2,t , torque x 3,t and the hydraulic system pressure x 4,t and temperature x 5,t, β0 is the constant term of the linear regression model, which represents the baseline value of the deviation when there is no device parameter fluctuation, β1,β2,…,β i Represents the regression coefficient, representing the parameters x of each device i,t The weight of the impact on the change in bias, each β i Indicates device parameter x i,t Contribution to the deviation change, the larger the weight, the greater the impact of the device on the deviation, ∈ t is a random error term; the linear regression model is used to quantify the impact of each device parameter on the deviation change.

[0046] Optionally, the device fault judgment unit generates a fault judgment based on the following conditions using the result calculated by the regression model:

[0047] Contribution judgment: If a device parameter x i,t The regression coefficient β on the deviation i If it is significantly greater than other equipment parameters, the equipment is judged to be the main source of failure, and it is determined that the equipment parameters cause deviation accumulation and cause failure;

[0048] Comprehensive Fault Score: Use the normalized regression coefficient to generate the comprehensive fault score S of the device i , expressed as follows: Among them, S i represents the failure score of the i-th device, β i is the regression coefficient of the i-th device parameter, β j is the regression coefficient of the jth device parameter, k is the total number of devices, the sum of the fault scores of all devices is 1, and the device with a high score is the source of the fault.

[0049] Beneficial effects of the present invention:

[0050] The present invention effectively avoids the problem of wind deviation in the yaw system of the wind turbine by introducing deviation detection and predictive analysis. The deviation data detection module continuously monitors the matching degree of the yaw angle and the wind direction, can accurately capture small deviations, and combine with the time series model to predict the cumulative trend of the deviation in real time. Then, through multi-level equipment data analysis, the specific equipment that causes the deviation is deeply analyzed, the fault source is accurately located, and the working state of the yaw system is ensured to be quickly adjusted to prevent the deviation from further aggravation, ensuring that the wind turbine always maintains the best matching degree with the wind direction, thereby improving the power generation efficiency and reducing the energy loss caused by the deviation.

[0051] The present invention, by introducing a time-series regression algorithm and a time series analysis model, can accurately capture the changing trend of the matching degree between the yaw angle and wind direction of a wind turbine. By continuously monitoring the matching of the yaw angle and wind direction, it can not only detect small deviations in real time, but also identify the cumulative effect of the deviations, and predict future deviation changes through a time series model. This trend prediction can effectively avoid neglecting potential problems, and can provide an early warning system that small deviations may develop into major faults, providing a reliable data basis for timely adjustment and maintenance.

[0052] The present invention uses progressive equipment fault identification. First, a preliminary screening is performed through a correlation analysis unit to quickly identify equipment parameters that are significantly correlated with yaw angle deviation. Then, a multivariate regression analysis unit further quantifies the influence weight of each equipment parameter on the deviation change, constructs a regression model of equipment parameters and deviation, and accurately evaluates the combined impact of multiple devices on the deviation. Through this progressive analysis method, the system can efficiently and accurately locate the specific equipment that causes the deviation, significantly improve the accuracy and efficiency of fault identification, and reduce misjudgments and missed judgments.

[0053] The present invention introduces a comprehensive fault scoring mechanism into the equipment fault judgment unit, and calculates the fault score of each device through the normalized regression coefficient. The scoring mechanism enables the system to intuitively compare the failure possibilities of multiple devices, thereby accurately marking potential fault sources. Combining historical data and current analysis results, the system generates an early warning to prompt operation and maintenance personnel to take timely maintenance or adjustment measures to prevent larger problems in the system. This mechanism ensures the intelligence and automation of fault prediction through real-time analysis and scoring mechanisms, greatly improving the maintenance efficiency and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 A schematic diagram of the functional modules of the system according to an embodiment of the present invention;

[0056] Figure 2 Schematic diagram of a device and deviation relationship analysis module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0058] It should be noted that references in the specification to "one embodiment," "an embodiment," "an exemplary embodiment," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes such specific features, structures, or characteristics. In addition, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0059] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0060] like Figure 1-Figure 2 As shown, the wind turbine fault prediction system based on machine learning includes:

[0061] Deviation data detection module: This module collects the yaw angle and wind direction sensor data of the yaw system in real time. By calculating the deviation between the yaw angle and wind direction, it detects the matching degree between the yaw angle and wind direction, and determines the wind turbine's wind accuracy. The data detection module detects small deviations in real time based on a time series regression algorithm and outputs the deviation data detection results.

[0062] Deviation change trend analysis module: Analyzes the trend of small deviations and captures the cumulative effect of small deviations in different time periods. The deviation change trend analysis module uses a time series model to identify the trend of deviation changes and predict future deviation increases.

[0063] Equipment real-time data acquisition and synchronization module: used to collect real-time data of equipment related to the yaw system, including the current, voltage, and torque of the yaw motor, and the pressure and temperature of the hydraulic system. It ensures that all equipment data is collected synchronously with the yaw angle and wind direction sensor data, and performs data standardization processing;

[0064] Equipment and deviation relationship analysis module: Based on the deviation change development trend, the relationship between the real-time data of each device and the small deviation is analyzed. The equipment and deviation relationship analysis module uses multivariate regression analysis and correlation analysis algorithms to identify which equipment parameter fluctuations show a significant correlation with the yaw angle deviation, generate equipment failure judgment results, and determine the specific equipment or component that causes the small deviation.

[0065] The deviation data detection module specifically includes:

[0066] Yaw angle acquisition unit: connects to the yaw sensor to obtain the yaw angle data of the wind turbine nacelle in real time. The yaw angle data is collected at a set frequency; the collection frequency is set between 1Hz and 5Hz, that is, 1 to 5 data are collected per second;

[0067] 1Hz: When the wind speed changes slowly, data can be collected once per second, which is suitable for monitoring under normal wind conditions;

[0068] 5Hz: When wind conditions fluctuate significantly or more accurate real-time feedback is needed, data can be collected 5 times per second, which is suitable for high-precision detection of small deviations.

[0069] Wind direction data acquisition unit: The wind direction data in the area where the wind turbine is located is acquired in real time through a wind direction sensor. The wind direction data is collected synchronously with the yaw angle data to ensure that both are collected on the same time basis. The data smoothing method is used to eliminate the instantaneous fluctuation interference of the wind direction data.

[0070] Deviation calculation unit: The deviation calculation unit receives yaw angle data and wind direction data, and determines the wind turbine's wind accuracy by calculating the angle between the yaw angle and the wind direction. If the angle exceeds a preset threshold (3°), it is considered that there is a deviation and the deviation data detection result is output;

[0071] Small deviation detection unit: Based on the time series regression algorithm, the small deviation detection unit analyzes multiple sets of deviation data in a short period of time, identifies persistent small deviations, and outputs the small deviation results to the deviation change trend analysis module in real time to predict the cumulative effect and trend of deviations.

[0072] The matching degree between the yaw angle and the wind direction is achieved by calculating the angle between the two. Let:

[0073] θ y Indicates the yaw angle of the wind turbine;

[0074] θ w Indicates the current wind direction angle;

[0075] The deviation angle Δθ is calculated as: Δθ = θ y -θ w, where Δθ represents the angle between the yaw angle and the wind direction. If the angle is too large, it means that the wind turbine is not accurately aligned with the wind direction, resulting in a decrease in power generation efficiency. It should be noted that if Δθ exceeds a preset threshold (for example, 3°), it is considered that there is a problem with the wind turbine's wind accuracy and there is a small deviation, which requires further analysis;

[0076] The detection of small deviations not only depends on the current deviation angle, but also needs to combine multiple data points in the time series, use regression algorithms for trend analysis, and regress the past n deviation detection data using the linear regression formula: Δθ t =β0+β1t+ε, where Δθ t θ0 represents the deviation value at time t, where t is the time variable. θ0 and β1 are parameters in the regression model. β1 represents the trend of the deviation over time (i.e., the rate of increase or decrease in the deviation). ε is the error term. Using the time series regression model, we can detect whether there are persistent small deviations (i.e., large or consistently positive β1 values). If β1 remains consistently positive over a period of time and the deviation angle does not decrease significantly, we can assume that a small deviation exists and requires attention.

[0077] The deviation change trend analysis module specifically includes:

[0078] Historical deviation data storage unit: used to store historical deviation data generated by the deviation data detection module. The data includes the difference between the yaw angle and the wind direction angle and the corresponding timestamp. The historical data is stored in a set time window to ensure that it contains enough samples for trend analysis;

[0079] Time series data processing unit: Receives historical deviation data, arranges it in chronological order, builds a time series model of deviation changes, and analyzes the trend of deviation changes in different periods. The time series model adopts the ARIMA model;

[0080] Small deviation accumulation detection unit: Identifies small deviations that persist over multiple time periods and captures the cumulative effect of deviations through time series models. If the deviation does not decrease or shows an increasing trend over multiple consecutive time periods, it is considered to be cumulative, indicating that system fine-tuning may be required.

[0081] Future Deviation Prediction Unit: Based on the prediction capability of the time series model, it uses current and historical deviation data to predict deviation changes in future periods.

[0082] The ARIMA model analyzes and predicts time series data. It captures the trend and seasonal effects of time series through three parts: autoregression (AR), differencing (I), and moving average (MA). It is suitable for analyzing the accumulated deviations of wind turbines and predicting future trends.

[0083] The ARIMA model is expressed as:

[0084]

[0085] y t : represents the deviation value Δθ at time t t ;

[0086] α: constant term;

[0087] Autoregressive coefficient, deviation value y at the i-th past moment t-i The impact weight on the current moment;

[0088] p: the order of the autoregressive model, which indicates the impact of the data in the previous moments on the current value;

[0089] θ j : Sliding mean coefficient, representing the past error term ∈ t-j The weight of the impact on the current error term;

[0090] q: the order of the moving average model, which represents the impact of past errors;

[0091] ∈ t : represents the error term at the current moment;

[0092] d: The number of differencing steps, used to make the time series data stationary.

[0093] In the present invention, the ARIMA model is used to analyze the deviation data Δθ between the yaw angle and the wind direction. t , the deviation data changes over time to form a time series, through the autoregressive part of the model Capture the correlation between deviation and historical deviation, judge how past deviation affects current deviation value, and at the same time, the sliding average part It captures the impact of historical noise on current deviations and reduces the interference of random fluctuations.

[0094] The ARI MA model reviews deviation data from the past and analyzes data change patterns. When it detects deviations between yaw angle and wind direction, it identifies whether the current deviation is gradually increasing, decreasing, or remaining stable based on the deviation changes over multiple consecutive time periods. This trend analysis can help determine whether the yaw system is entering an abnormal state.

[0095] The changing trends of deviation data include:

[0096] Short-term changes: The ARI MA model uses data from recent periods to determine whether there are short-term fluctuations in the deviation. If the deviation fluctuates significantly in the short term, it may indicate an unstable system response.

[0097] Long-term trend: By integrating data over a long period of time, the ARI MA model identifies the long-term trend of deviations and analyzes whether there is a continuous accumulation or slow change in deviations. If the deviation value is slowly increasing over multiple consecutive periods, this cumulative trend may mean that the system needs adjustment or maintenance.

[0098] The ARI MA model analyzes deviation data and captures the phenomenon of deviation accumulation. The cumulative effect is that the deviation does not decrease over multiple time periods and shows a trend of gradual increase. The ARI MA model determines whether the deviation is accumulating by identifying trend changes in the time period.

[0099] Specifically, if the system detects persistent deviations over multiple consecutive periods without a significant decrease, the model indicates that the deviations are accumulating. This cumulative effect typically indicates that the yaw system is not adjusting quickly enough, or that there are equipment issues preventing the deviations from returning to normal levels. If this accumulation persists, system performance may gradually deteriorate, leading to a more serious failure.

[0100] The ARIMA model is particularly adept at capturing this type of variability over time. If the deviation value fails to decrease significantly over multiple consecutive time periods and instead shows a gradual increase, the model will determine that the deviation is accumulating. This cumulative effect indicates that the wind turbine yaw system may be gradually deviating from normal operation. If the model detects a continuous increase in the magnitude of the deviation at regular intervals, albeit a small one, but with continued accumulation, the system will issue an alarm indicating that deviation accumulation is occurring.

[0101] The ARIMA model analyzes current and historical deviation data to infer future deviation trends. The prediction process includes:

[0102] Impact of current deviation data: The current deviation value affects short-term performance. If the current deviation is large, the system is expected to maintain this deviation in the short term.

[0103] Historical deviation trends: By identifying deviation patterns over the past period (such as gradual increase, fluctuation, stability, etc.), it is inferred that future deviations will continue this deviation pattern. For example, if historical data shows that the deviation continues to increase, the model may predict that the deviation will increase further in the next few periods.

[0104] The ARIMA model also considers the volatility of the deviation. If there are large fluctuations in the data, it may be predicted that the deviation in the future will also show a fluctuating trend.

[0105] The device real-time data acquisition and synchronization module specifically includes:

[0106] Motor data acquisition unit: collects the operating parameters of the yaw motor in real time, including motor current, voltage and torque signals. The motor data acquisition unit collects these data at a set sampling frequency by connecting to the motor sensor in the yaw system.

[0107] Hydraulic system data acquisition unit: collects real-time data from the hydraulic components of the yaw system, including pressure and temperature signals of the hydraulic system. Hydraulic data is obtained through hydraulic sensors and changes in the hydraulic system operation are recorded synchronously to ensure that the pressure and temperature data can match the yaw motor data;

[0108] Data synchronization unit: This unit ensures the synchronous collection of the yaw motor's current, voltage, and torque data, the hydraulic system's pressure and temperature data, and the yaw angle and wind direction sensor data. By matching timestamps and aligning timing, it unifies the time base of device data, ensuring that data from each device can be correlated and analyzed in real time.

[0109] Data standardization processing unit: used to standardize all collected equipment data to ensure that data of different units and magnitudes can be effectively compared and analyzed. The standardization process is based on the Z-score standardization method to convert the equipment data into a dimensionless form so that the subsequent analysis module can perform correlation analysis between equipment status and deviation.

[0110] The device-deviation relationship analysis module specifically includes the following units, which use a progressive analysis method to identify the relationship between device parameter fluctuations and small yaw angle deviations:

[0111] Correlation Analysis Unit: Receives real-time data from various devices, including current, voltage, and torque data from the yaw motor, pressure and temperature data from the hydraulic system, and deviation data from the yaw angle and wind direction. Using a correlation analysis algorithm, it preliminarily screens out device parameters that exhibit a significant linear correlation with the deviation data. The function of the correlation analysis unit is to quickly identify variables in the device that have a strong linear correlation with the deviation data, thereby reducing the computational effort of subsequent analysis and focusing on highly correlated devices.

[0112] Multivariate regression analysis unit: Based on the highly correlated equipment parameters screened by the correlation analysis unit, multivariate regression analysis is performed to refine the relationship between equipment parameters and deviations. Multivariate regression analysis quantifies the weight of the impact of each equipment parameter on deviation changes and establishes a linear regression model to evaluate the combined impact of multiple equipment parameters on small deviations. Through regression analysis, one or more equipment parameters are determined to have the greatest contribution to the deviation change, thereby accurately identifying potential faulty equipment or components.

[0113] Equipment Fault Judgment Unit: Based on the results of the multivariate regression analysis unit, it generates a final judgment on equipment faults. If the fluctuation of a certain equipment parameter has a large impact on the deviation and is significantly correlated with the cumulative effect of small deviations, the equipment is marked as a potential fault source and fault warning information is generated. The equipment fault judgment unit combines historical fault data to further verify the match between the current equipment status and typical fault modes to improve the accuracy of fault identification.

[0114] Preliminary screening of the correlation analysis unit: First, through correlation analysis, the correlation coefficient between the operating parameters and deviation data of each device is calculated. If some device parameters show a strong linear correlation with the deviation change (such as a high correlation coefficient), these devices will be marked as possible sources of influence. The correlation analysis narrows the scope of analysis for subsequent more complex multivariate regression analysis, reduces unnecessary calculations, and focuses on devices with high correlation. Detailed analysis of the multivariate regression analysis unit: After preliminarily screening out devices with high correlation, the multivariate regression analysis unit further establishes a multivariate regression model to quantify the specific impact of each device parameter on the deviation change through the regression coefficient. This step can not only identify the comprehensive impact between multiple devices, but also deal with the interdependence between device parameters, providing more accurate judgment results. After the multivariate regression analysis is completed, the system will combine the weight results in the regression analysis to determine which one or more devices have the most significant impact on the deviation change.

[0115] Preliminary screening through the correlation analysis unit reduces the computational burden of subsequent analysis and improves the system's real-time responsiveness. Multivariate regression analysis further quantifies the impact of devices on deviations, ensuring the system can accurately determine which device or devices are the root cause of the deviation. The device fault diagnosis unit combines historical data with current analysis results to provide more reliable fault predictions, avoiding misjudgments or missed detections.

[0116] The correlation analysis unit uses the Pearson correlation coefficient to measure the linear relationship between each device parameter and the yaw angle deviation. The Pearson correlation coefficient compares each device's operating parameter with the deviation between the yaw angle and wind direction. Specifically, it includes:

[0117] Calculating variances: The correlation analysis unit calculates the overall trend of device operating parameters and deviation data. If the deviation changes as the current or pressure changes over time, then these device parameters may show a certain correlation with the deviation.

[0118] Measuring correlation: Calculate the degree of correlation between the equipment operating parameters and the deviation. The Pearson correlation coefficient outputs whether the two variables change at the same time. If the parameter change of a device is positively correlated with the deviation (that is, when the device parameter increases, the deviation also increases, or when the device parameter decreases, the deviation also decreases), it is considered that the device may have a strong influence on the deviation. If the correlation coefficient is negative, it indicates the opposite relationship (that is, when one increases, the other decreases). If the correlation coefficient is close to zero, it means that there is almost no linear relationship between the two.

[0119] Initial device screening: By calculating the correlation between multiple device parameters and deviation, the system quickly identifies the device operating parameters that have a significant impact on the deviation. For example, if the correlation coefficient between current and deviation is close to 1, this indicates that there is a strong positive correlation between current changes and deviation, which may mean that the motor status has a significant impact on the yaw angle deviation. On the other hand, if the correlation coefficient between a device parameter and deviation is close to 0, the system assumes that the device parameter is unrelated to the deviation change and can temporarily exclude it as a potential fault source.

[0120] The expression of Pearson correlation coefficient is:

[0121] The parameters are defined as follows:

[0122] x i : The value of a device parameter at the i-th moment (such as the current, voltage, torque of the yaw motor, or the pressure and temperature of the hydraulic system);

[0123] y i : The deviation value Δθ between the yaw angle and wind direction at the i-th moment i ,Right now in is the yaw angle, is the wind direction angle;

[0124] Device parameter x i The mean of

[0125] Deviation value y i The mean of

[0126] r xy : Device parameter x i and the deviation value y i The Pearson correlation coefficient, correlation coefficient r xy The value range of is [-1, 1], close to 1 indicates a high positive correlation, close to -1 indicates a high negative correlation, and close to 0 indicates no significant correlation.

[0127] By calculating the device parameter x i and the deviation value y i The correlation coefficient r xy , the system can preliminarily screen out which equipment parameters have a strong linear relationship with the deviation change. If the correlation coefficient r xy If it is close to 1 or -1, it indicates that the parameter fluctuation of the equipment has a strong influence on the deviation change, and the multivariate regression analysis is carried out in the next step.

[0128] The linear regression model evaluates the impact of multiple equipment parameters on the deviation change simultaneously, and the expression is:

[0129] Δθ t =β0+β1x 1,t +β2x 2,t +…+β i x i,t +∈ t ,in;

[0130] Δθ t Indicates the deviation value at time t, which is the difference between the yaw angle and the wind direction angle, x 1,t ,x 2,t ,…,x i,t Indicates the parameter values ​​of each device collected at time t, which respectively represent the real-time data of the device related to the yaw system, including the current x of the yaw motor 1,t , voltage x 2,t , torque x 3,t and the hydraulic system pressure x 4,t and temperature x 5,t , β0 is the constant term of the linear regression model, which represents the baseline value of the deviation when there is no device parameter fluctuation, β1,β2,…,β i Represents the regression coefficient, representing the parameters x of each device i,t The weight of the impact on the change in bias, each β i Indicates device parameter x i,t Contribution to the deviation change, the larger the weight, the greater the impact of the device on the deviation, ∈ t is the random error term, which represents the portion of the deviation that cannot be explained by the equipment parameters.

[0131] Through the linear regression model, the influence of each device parameter on the deviation change is quantified. If a device parameter x 1,t The regression coefficient β1 of the deviation change of the device (such as current) is large, indicating that the parameter change of the device has a significant impact on the deviation. Multivariate regression analysis can not only identify the combined impact of multiple devices on the deviation at the same time, but also evaluate the interaction between them.

[0132] In the equipment fault judgment unit, the results of the regression model calculation are used to generate a fault judgment based on the following conditions:

[0133] Contribution judgment: If a device parameter x i,t The regression coefficient β on the deviation i Significantly greater than other device parameters (such as β i >β j , where i≠j), the device is judged to be the main fault source, and it is determined that the device parameters cause deviation accumulation and cause the fault.

[0134] Example: If β i >β threshold Then determine device x i Fault;

[0135] Comprehensive Fault Score: Use the normalized regression coefficient to generate the comprehensive fault score S of the device i , expressed as follows: Among them, S i represents the failure score of the i-th device, β i is the regression coefficient of the ith device parameter, β j is the regression coefficient of the jth device parameter, k is the total number of devices, the sum of the fault scores of all devices is 1, and the device with a high score is the source of the fault.

[0136] Based on the analysis results of the regression model, the equipment fault judgment unit can quantify the possibility of fault according to the weight of the equipment's influence on the deviation change. When the fault score Si of a certain device is significantly higher than that of other devices, the system marks the device as the source of fault and issues an early warning.

[0137] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0138] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A wind turbine fault prediction system based on machine learning, characterized by: include: Deviation data detection module: collects yaw angle and wind direction sensor data of the yaw system in real time, calculates the deviation between the yaw angle and wind direction, detects the matching degree between the yaw angle and wind direction, and determines the wind accuracy of the wind turbine. The deviation data detection module detects small deviations in real time based on a time series regression algorithm and outputs the deviation data detection results; Deviation change trend analysis module: Analyzes the trend of small deviations and captures the cumulative effect of small deviations in different time periods. The deviation change trend analysis module uses a time series model to identify the trend of deviation changes and predict future deviation increases. Equipment real-time data acquisition and synchronization module: used to collect real-time data of equipment related to the yaw system, including the current, voltage, and torque of the yaw motor, and the pressure and temperature of the hydraulic system. It ensures that all equipment data is collected synchronously with the yaw angle and wind direction sensor data, and performs data standardization processing; Equipment and Deviation Relationship Analysis Module: This module analyzes the relationship between each device's real-time data and small deviations based on deviation change trends. Using multivariate regression analysis and correlation analysis algorithms, it identifies which device parameter fluctuations are correlated with yaw angle deviations, generates equipment failure judgments, and identifies the specific device or component causing the small deviation. The equipment and deviation relationship analysis module specifically includes the following units: Correlation analysis unit: Receives real-time data from various devices, including current, voltage, and torque data of the yaw motor, pressure and temperature data of the hydraulic system, and deviation data of the yaw angle and wind direction. Using a correlation analysis algorithm, it preliminarily selects device parameters that are linearly correlated with the deviation data. Multivariate regression analysis unit: Based on the highly correlated equipment parameters screened by the correlation analysis unit, multivariate regression analysis is performed to refine the relationship between equipment parameters and deviation. Multivariate regression analysis quantifies the weight of the impact of each equipment parameter on the deviation change, and establishes a linear regression model to evaluate the combined impact of multiple equipment parameters on small deviations. Through regression analysis, it is determined that one or more equipment parameters have the greatest contribution to the deviation change; Equipment failure judgment unit: Based on the results of the multivariate regression analysis unit, the final judgment of equipment failure is generated. If the fluctuation of a certain equipment parameter has a large impact on the deviation and is significantly correlated with the cumulative effect of small deviations, the equipment is marked as a potential fault source and fault warning information is generated.

2. The wind turbine fault prediction system based on machine learning according to claim 1, characterized in that: The deviation data detection module specifically includes: Yaw angle acquisition unit: connected to the yaw sensor to obtain the yaw angle data of the wind turbine nacelle in real time. The yaw angle data is collected at a set frequency; Wind direction data acquisition unit: acquires wind direction data of the area where the wind turbine is located in real time through a wind direction sensor. The wind direction data is collected synchronously with the yaw angle data; Deviation calculation unit: The deviation calculation unit receives yaw angle data and wind direction data, determines the wind accuracy of the wind turbine by calculating the angle between the yaw angle and the wind direction, and outputs the deviation data detection result; Small deviation detection unit: Based on the time series regression algorithm, the small deviation detection unit analyzes multiple sets of deviation data in a short period of time, identifies persistent small deviations, and outputs the small deviation results to the deviation change trend analysis module in real time.

3. The wind turbine fault prediction system based on machine learning according to claim 1, characterized in that: The deviation change trend analysis module specifically includes: A historical deviation data storage unit is used to store historical deviation data generated by the deviation data detection module, wherein the historical deviation data includes the difference between the yaw angle and the wind direction angle and the corresponding timestamp; Time series data processing unit: Receives historical deviation data, arranges it in chronological order, builds a time series model of deviation changes, and analyzes the trend of deviation changes in different periods. The time series model adopts the ARIMA model; Small deviation accumulation detection unit: Identifies small deviations that persist over multiple time periods and captures the cumulative effect of deviations through time series models. If the deviation does not decrease or shows an increasing trend over multiple consecutive time periods, it is considered to be cumulative. Future Deviation Prediction Unit: Based on the prediction capability of the time series model, it uses current and historical deviation data to predict deviation changes in future periods.

4. The wind turbine fault prediction system based on machine learning according to claim 3, characterized in that: The ARIMA model reviews deviation data from the past period and analyzes data change patterns. When the deviation of yaw angle and wind direction is collected, it identifies whether the current deviation is gradually increasing, decreasing, or remaining stable based on the deviation changes over multiple consecutive periods, and determines whether the operation of the yaw system is entering an abnormal state. The variation trends of the deviation data include: Short-term changes: The ARIMA model uses data from recent periods to determine whether there are short-term fluctuations in the deviation. Long-term trends: By integrating data over a long period of time, the ARIMA model identifies the long-term trend of deviations and analyzes whether there is a continuous accumulation or slow change in deviations.

5. The wind turbine fault prediction system based on machine learning according to claim 4, characterized in that: The ARIMA model analyzes current and historical deviation data to infer future deviation trends. The prediction process includes: Impact of current deviation data: The current deviation value affects short-term performance. If the current deviation is large, the system is expected to maintain this deviation in the short term. Historical deviation trends: By identifying deviation patterns over a period of time in the past, it is inferred that future deviations will continue this deviation pattern.

6. The wind turbine fault prediction system based on machine learning according to claim 1, characterized in that: The device real-time data acquisition and synchronization module specifically includes: Motor data acquisition unit: collects the operating parameters of the yaw motor in real time, including motor current, voltage and torque signals. The motor data acquisition unit collects these data at a set sampling frequency by connecting to the motor sensor in the yaw system; Hydraulic system data acquisition unit: collects real-time data from the yaw system hydraulic components, including pressure and temperature signals of the hydraulic system. Hydraulic data is obtained through hydraulic sensors and changes in the hydraulic system operation are recorded synchronously. Data synchronization unit: This unit ensures the synchronous collection of the yaw motor's current, voltage, and torque data, the hydraulic system's pressure and temperature data, and the yaw angle and wind direction sensor data. Through timestamp matching and timing alignment, the time base of the equipment data is unified; Data standardization processing unit: used to standardize all collected device data to ensure that data of different units and magnitudes can be effectively compared and analyzed. The standardization processing is based on the Z-score standardization method.

7. The wind turbine fault prediction system based on machine learning according to claim 1, characterized in that: The correlation analysis unit uses the Pearson correlation coefficient to measure the linear relationship between each device parameter and the yaw angle deviation. The Pearson correlation coefficient compares the operating parameter of each device with the deviation between the yaw angle and the wind direction, specifically including: Calculate differences: The correlation analysis unit calculates the overall trend of equipment operating parameters and deviation data; Measuring correlation: Calculate the degree of correlation between the equipment operating parameters and the deviation. The Pearson correlation coefficient outputs whether the two variables change at the same time. If the parameter change of a device is positively correlated with the deviation, it is considered that the device may have a strong influence on the deviation. If the correlation coefficient is negative, it indicates the opposite relationship. If the correlation coefficient is close to zero, it means there is no linear relationship between the two. Preliminary equipment screening: By calculating the correlation between multiple equipment parameters and deviations, the equipment operating parameters that have a significant impact on the deviation can be quickly identified.

8. The wind turbine fault prediction system based on machine learning according to claim 7, characterized in that: The linear regression model evaluates the impact of multiple equipment parameters on the deviation change simultaneously, and the expression is: ,in; Indicates time The deviation value at the moment represents the difference between the yaw angle and the wind direction angle. Indicates time The parameter values ​​of each device collected at all times represent the real-time data of the device related to the yaw system, including the current of the yaw motor ,Voltage , torque and the pressure of the hydraulic system and temperature , is the constant term of the linear regression model, which represents the baseline value of the deviation when there is no device parameter fluctuation. Represents the regression coefficient, which represents the parameters of each device The weight of the impact on the deviation change, each Indicates device parameters The contribution to the deviation change. The larger the weight, the greater the impact of the device on the deviation. is a random error term that quantifies the effect of each device parameter on the deviation change through a linear regression model.

9. The wind turbine fault prediction system based on machine learning according to claim 8, characterized in that: In the equipment fault judgment unit, the result of the regression model calculation is used to generate a fault judgment based on the following conditions: Contribution judgment: If a device parameter Regression coefficient on bias If it is greater than other equipment parameters, the equipment is judged to be the main source of failure, and it is determined that the equipment parameters cause the accumulation of deviations and cause the failure; Comprehensive Fault Score: Use the normalized regression coefficient to generate a comprehensive fault score for the device , expressed as follows: ,in, Indicates the Failure score of each device, It is The regression coefficients of the device parameters, It is The regression coefficients of the device parameters, is the total number of devices. The sum of the fault scores of all devices is 1. The device with the highest score is the source of the fault.

Images