Gearbox fault early warning method based on wind power space-time adaptive clustering and similarity
By constructing a gearbox low-speed bearing temperature prediction model based on the method of spatio-temporal adaptive clustering and similarity in wind power, it solves the problem of inaccurate early warning of wind turbine gearbox faults in the existing technology and achieves more efficient fault early warning and maintenance management. .
Patent Information
- Application Number
- CN202411434080.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-15
- Publication Date
- 2025-01-24
AI Technical Summary
The prior art is difficult to effectively warn of the gearbox of the wind turbine assembly, resulting in unstable operation of the wind turbine assembly and increasing maintenance costs.
Using the method of adaptive cluster and similarity based on wind and space, by obtaining the weather information and operating status data of the wind farm, the data cleaning, missing value filling and cluster analysis are performed. Models to achieve fault warning.
The accuracy and timeliness of the gearbox fault warning of the wind turbine is improved, the maintenance cost is reduced, and the stable operation of the wind turbine is ensured.
Smart Images
Figure CN119353165A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of wind power, and in particular relates to a gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity. Background Art
[0002] With the national "carbon peak" and "carbon neutrality" goals, wind power generation has been developed and utilized on a large scale in recent years. According to statistics, during the life cycle of a wind turbine, the operation and maintenance costs of onshore units can account for 15%-20% of the total power generation revenue, while this proportion of offshore wind turbines is as high as 25%-30%. In order to ensure the long-term stable operation of wind turbines, the key lies in the development of a high-precision wind turbine status detection system. As the core component of a wind turbine, the gearbox, the detection and early warning of gearbox abnormalities are an important part of ensuring the healthy and stable operation of wind turbines. Summary of the invention
[0003] The technical problem to be solved by the present invention is to provide a gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity in view of the deficiencies in the prior art.
[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: A gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity, comprising: Obtain meteorological information and wind turbine operating status data from wind farm wind towers and wind turbine SCADA systems; According to the spatiotemporal correlation of wind power data, the data of multiple units at the same time and the data of a single unit before and after are used to clean abnormal data and fill missing values in the wind turbine operating status data; Conduct cluster analysis on wind turbine operating status data; The meteorological data and wind turbine operating status data are processed using the Pearson correlation coefficient method to obtain characteristic data used as model input; Constructing a gearbox temperature prediction model, and inputting the characteristic data into the prediction model to obtain predicted temperature data of the low-speed bearing of the gearbox under normal operating conditions; Obtaining a prediction error based on actual temperature data and predicted temperature data of a low-speed bearing of a gearbox, and calculating a distribution characteristic and a confidence interval of the prediction error; The distribution characteristics of the actual values of the low-speed bearing temperature of the gearbox of similar wind turbines are analyzed based on the clustering results. If the actual value of the low-speed bearing temperature of the gearbox of a wind turbine is greater than the upper limit of the confidence interval of other wind turbines of the same type by 10 degrees Celsius, an alarm message will be issued.
[0005] Furthermore, it also includes: using the prediction model to predict the temperature data of the gearbox low-speed bearing in the next day in a sliding window mode, combining the gearbox low-speed bearing temperature data of similar units, and monitoring the operating status of the gearbox low-speed bearings of multiple units in the wind farm.
[0006] Furthermore, the method for filling missing values in the wind turbine operating status data includes: The operating status data of each unit are matched and combined at the same time, ensuring that each column is a certain operating status data of the same unit, and the same row is a certain operating status data of multiple units at the same time; according to , and Fill in missing values, where is the missing value to be filled, is the local mean on the time scale, is the local mean on the spatial scale, n is the number of existing data near the missing value, m is the number of wind turbines, and p is the unit with missing data.
[0007] Furthermore, after the missing values of the wind turbine operating status data are filled, the meteorological data and the filled wind turbine operating status data are normalized.
[0008] Furthermore, the method for clustering analysis of wind turbine operating status data includes: clustering analysis of wind turbine operating status data using a self-organizing map neural network to cluster wind turbines with similar operating status into one category.
[0009] Furthermore, the method for constructing a gearbox temperature prediction model includes: Initialize the running parameters of the whale algorithm and randomly generate the position coordinates of individual whales; According to the individual whales, the least squares support vector machine was used to build a prediction model and calculate the temperature; Save the whale individual with the smallest calculation error; The position of individual whales is updated iteratively. In each iteration, the prediction model is rebuilt based on the updated individual whales, the temperature is calculated, and the individual whale with the smallest calculation error is saved. After the iteration, the prediction model constructed with the optimal whale individual is obtained.
[0010] Furthermore, the objective function of the least squares support vector machine is: , , ; in is the objective function of the least squares support vector, γ is the penalty factor, represents the loss function, W is the normal vector of the hyperplane, φ(x k ) is the kernel function, x k is the support vector, b is a constant term, M is the number of support vectors, is the support vector output, is the kernel function width; Update and utilization of individual whale positions in the whale algorithm , Calculate, where D is the distance between the current position and the global optimal position of the population, X i (t) is the position of the ith whale at the tth iteration, X * (t) is the global optimal position of the whale population at the tth iteration, A is the convergence factor, and C is the swing factor.
[0011] Furthermore, the prediction error is calculated as follows: ,in, is the predicted value of the equipment operating status parameter, is the true value of the equipment operating status parameter, is the difference between the predicted value and the true value.
[0012] Furthermore, the distribution characteristics of the prediction error and the confidence interval calculation method are as follows: , ;in is the confidence level [ ] is the lower boundary of the distribution range of the model prediction error, is the confidence level [ ] is the upper boundary of the distribution range of the model prediction error; the confidence interval is .
[0013] Furthermore, the confidence level is 99.73%.
[0014] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity. The method obtains meteorological information and wind turbine operating status data from the wind farm wind tower and wind turbine SCADA system, and fills in the missing data values using the spatiotemporal distribution characteristics between wind turbine data. On this basis, the self-organizing map neural network algorithm is used to cluster the units with similar operating status in the wind farm into one category; the Pearson correlation coefficient method is used to select features with high correlation with the predicted target variable as model input, so as to reduce the model training time and improve the prediction accuracy; the WOA-LSSVM (whale algorithm-least squares support vector machine) model is used to mine The correlation characteristics between the unit operating status data and the gearbox low-speed bearing temperature are used to build a gearbox low-speed bearing temperature early warning model based on WOA-LSSVM; the prediction error under normal operating conditions of the wind turbine is used to calculate the confidence interval with a confidence level of 99.73%, and the distribution characteristics of the actual values of the gearbox low-speed bearing temperature of similar wind turbines are compared and analyzed to provide fault early warning for the gearbox operating status of multiple units; the WOA-LSSVM model is used to predict the gearbox low-speed bearing temperature data of the next day in sliding window mode, and the gearbox low-speed bearing temperature data of similar units are combined to carry out multi-unit inspection, so as to realize real-time monitoring of the operating status of multiple units in the wind farm. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The present invention is further described in detail below in conjunction with the accompanying drawings.
[0016] Figure 1 : Flowchart of the present invention; Figure 2 : Prediction effect diagram of the prediction model of the present invention; Figure 3 : Temperature distribution characteristics of low-speed bearing of gearbox of wind turbine on December 21; Figure 4 : Temperature distribution characteristics of low-speed bearing of gearbox of wind turbine on December 22; Figure 5 : Temperature distribution characteristics of low-speed bearing of gearbox of wind turbine on December 30. DETAILED DESCRIPTION
[0017] In order to better understand the present invention, the content of the present invention is further clearly described below in conjunction with the embodiments and the accompanying drawings, but the protection content of the present invention is not limited to the following embodiments. In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details.
[0018] Example 1: See Figure 1-Figure 5,The purpose of this embodiment is to provide a gear box fault early warning method based on ,wind power spatiotemporal adaptive clustering and similarity.
[0019] Figure 1 As shown, the wind turbine gearbox fault early warning method includes: S100, obtains historical wind speed, wind direction, ambient temperature and other meteorological data from the wind farm wind tower and wind turbine SCADA system, as well as wind turbine operating status data such as power, generator speed, gearbox oil tank temperature, gearbox low-speed bearing temperature, gearbox high-speed bearing temperature, gearbox inlet oil temperature, and cabin temperature. The data acquisition time resolution is 1 minute.
[0020] S200. According to the spatiotemporal correlation of wind power data, the data of multiple units at the same time and the data of a single unit before and after are used to clean abnormal data and fill missing values in the wind turbine operating status data, and normalize the meteorological data and the wind turbine operating status data.
[0021] Methods for filling missing values include: The operating status data of each unit are matched and combined at the same time, ensuring that each column is a certain operating status data of the same unit, and the same row is a certain operating status data of multiple units at the same time; According to formulas (1), (2) and (3), the missing values of each operating status data of the wind turbine SCADA data are filled: (1) (2) (3) In formulas (1), (2) and (3), is the missing value to be filled, is the local mean on the time scale, is the local mean on the spatial scale, n is the number of existing data near the missing value, here n=10; m is the number of wind turbines, and p is the unit with missing data.
[0022] In order to reduce the learning error during model training and improve the learning speed, the data needs to be standardized before it is input into the model to unify the dimensions of different feature data. According to formula (4), the meteorological data and the wind turbine operating status data after missing value filling are normalized: (4) In formula (4) is the value after data standardization; Indicates the maximum value in the data; Indicates the minimum value.
[0023] S300, cluster analysis is performed on the wind turbine operating status data, and wind turbines with similar operating status are clustered into one category. This step is implemented using a self-organizing map neural network algorithm, and the self-organizing map network learning process is as follows: (a) Initialize connection weights: Initialize learning rate , initialize each neuron in the range [0,1] The weight matrix ; (b) The sample set Normalized to a value between 0 and 1, where M represents the number of samples and D represents the feature dimension of a single sample; (c) Randomly select a sample from M samples and input it into the competition layer to calculate the sample dot product ,in , , select the winning node with the largest dot product, and define the initial winning neighborhood based on the winning node .
[0024] (d) Input M samples into the competitive layer in sequence and calculate the sample dot product ,in , , select the winning node with the largest dot product, and redefine the winning neighborhood based on the winning node .
[0025] (e) Adjust the weights of the nodes in the winning neighborhood. The weight update formula is as follows: (5) In formula (5), Represents the current iteration number. is the weight update learning rate, and the neighborhood neurons and the winning neuron The topological distance between The function is expressed as , It represents the difference between the weight of the current sample and a certain point in the winning field.
[0026] (f) Repeat steps (d)-(e) until the number of iterations is met or the learning rate drops to a certain value.
[0027] (g) All samples are input and finally classified and stored according to the clustering results.
[0028] S400, using the Pearson correlation coefficient method (Pearson correlation coefficient) to process the meteorological data and the wind turbine operating status data to obtain feature data used as model input. This step can select highly correlated feature variables as model input, which can reduce model training time, improve prediction accuracy, and reduce training time.
[0029] The calculation formula is as follows: (6) In the formula, and represents the i-th feature of variables X and Y, and Represents the mean of variables X and Y.
[0030] S500, construct a gearbox low-speed bearing temperature prediction model, and use the prediction model to predict the gearbox low-speed bearing temperature data under normal operating conditions. The prediction model uses the characteristic data obtained in step S400 as input and uses the predicted gearbox low-speed bearing temperature data as output.
[0031] This step uses the whale algorithm and the least squares support vector machine to build a gearbox low-speed bearing temperature prediction model. Specifically, this step includes: (a) Initialize the operating parameters of the whale algorithm and randomly generate the position coordinates of individual whales.
[0032] (b) A prediction model is constructed based on individual whales using the least squares support vector machine and temperature is calculated.
[0033] The objective function of the least squares support vector machine can be calculated using equations (7), (8) and (9): (7) (8) (9) In formulas (7), (8) and (9), is the objective function of the least squares support vector, γ is the penalty factor, represents the loss function, W is the normal vector of the hyperplane, φ(x k ) is the kernel function, x k is the support vector, b is a constant term, M is the number of support vectors, is the support vector output, is the kernel function width.
[0034] (c) Save the whale individual with the smallest calculation error.
[0035] (d) Iteratively update the position of individual whales, returning to step (b) and re-executing after each update.
[0036] The update of the whale's position can be calculated using equations (10) and (11): (10) (11) Where D is the distance between the current position and the global optimal position of the population, X i (t) is the position of the ith whale at the tth iteration, X * (t) is the global optimal position of the whale population at the tth iteration, A is the convergence factor, and C is the swing factor.
[0037] (c) After the iteration, the prediction model constructed with the optimal whale individual is obtained.
[0038] S600 , combining actual temperature data to obtain a prediction error, and calculating a distribution characteristic and a confidence interval of the prediction error.
[0039] Specifically, this step includes: (a) The prediction error calculation process under normal operating conditions is shown in the following formula: (12) In formula (12), is the predicted value of the equipment operating status parameter, is the true value of the equipment operating status parameter, is the difference between the predicted value and the true value.
[0040] (b) The Gaussian distribution function and confidence interval of the model prediction error can be calculated using equations (13) and (14): (13) (14) In formulas (13) and (14), is the confidence level [ ] is the lower boundary of the distribution range of the model prediction error, is the confidence level [ ] is the upper boundary of the distribution range of the model prediction error. According to the upper and lower boundary values of the model calculation error, the confidence level [ The confidence interval of the predicted value of the equipment operating status parameter under ] is The preferred confidence level is 99.73%.
[0041] S700: Analyze the distribution characteristics of the actual values of the low-speed bearing temperature of the gearbox of similar wind turbines based on the clustering results. When the actual value of the low-speed bearing temperature of the gearbox of a certain wind turbine is greater than the upper limit of the confidence interval of other wind turbines of the same type by 10 degrees Celsius, issue an alarm message.
[0042] S800 uses the prediction model to predict the gearbox low-speed bearing temperature data of the next day in sliding window mode, and combines the gearbox low-speed bearing temperature data of similar units to conduct multi-unit inspection, so as to realize real-time monitoring of the operating status of gearbox low-speed bearings of multiple units in the wind farm.
[0043] The present invention is verified by taking a wind farm in eastern China as an example. The SCADA system operation fault report of the wind farm records that the A15 unit has an abnormal gearbox low-speed bearing temperature on December 22, 2016. We select the data on December 21, December 22 and December 30 as verification data, and calculate the confidence interval under the confidence level of 99.73% according to the distribution characteristics of the normal prediction error of the gearbox low-speed bearing temperature, and compare and analyze the distribution characteristics of the actual value of the gearbox low-speed bearing temperature of similar wind turbines. When the actual value of the gearbox low-speed bearing temperature of a certain wind turbine is greater than the upper limit of the confidence interval of other wind turbines of the same type by 10 degrees Celsius, the wind turbine sends an alarm message.
[0044] Figure 2 In order to utilize the prediction results of the low-speed bearing temperature of multiple unit gearboxes of the present invention, Figure 3 , Figure 4 and Figure 5 The comparative analysis diagrams are of A5, A6, A11, A15 and A16 wind turbines with similar operating status on December 21, December 22 and December 30. The actual operation report of the wind farm shows that the A15 unit failed at 10:19-18:20 on December 22, 2016, and the cause of the failure was the high temperature alarm of the low-speed bearing of the gearbox.
[0045] The following table shows the analysis results of the present invention on the operating status of the gearboxes of the A5, A6, A11, A15 and A16 wind turbines: date A5 A6 A11 A15 A16 December 21 Normal operation Normal operation Normal operation At 23:00, the temperature exceeds the upper limit of the confidence interval of similar units by 10 degrees Celsius, and the warning model issues a warning message every 10 minutes Normal operation December 22 Normal operation Normal operation Normal operation At 2:55, 8:00 and 10:20, the temperature exceeded the upper limit of the confidence interval of similar units by 10°C. The warning model issued a warning signal every 10 minutes. Normal operation December 30 Normal operation Normal operation Normal operation Normal operation Normal operation It can be seen from the results that the calculation results of the gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity proposed in the present invention are consistent with the actual operation results of the wind turbine set, which further proves the accuracy and superiority of the gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity.
[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in the field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.
Claims
1. A gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity, characterized in that: include: Obtain meteorological information and wind turbine operating status data from wind farm wind towers and wind turbine SCADA systems; According to the spatiotemporal correlation of wind power data, the data of multiple units at the same time and the data of a single unit before and after are used to clean abnormal data and fill missing values in the wind turbine operating status data; Conduct cluster analysis on wind turbine operating status data; The meteorological data and wind turbine operating status data are processed using the Pearson correlation coefficient method to obtain characteristic data used as model input; Constructing a gearbox temperature prediction model, and inputting the characteristic data into the prediction model to obtain predicted temperature data of the low-speed bearing of the gearbox under normal operating conditions; Obtaining a prediction error based on actual temperature data and predicted temperature data of a low-speed bearing of a gearbox, and calculating a distribution characteristic and a confidence interval of the prediction error; The distribution characteristics of the actual values of the low-speed bearing temperature of the gearbox of similar wind turbines are analyzed based on the clustering results. If the actual value of the low-speed bearing temperature of the gearbox of a wind turbine is greater than the upper limit of the confidence interval of other wind turbines of the same type by 10 degrees Celsius, an alarm message will be issued.
2. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 1 is characterized in that: It also includes: using the prediction model to predict the temperature data of the gearbox low-speed bearing in the future day in a sliding window mode, combining the gearbox low-speed bearing temperature data of similar units, and monitoring the operating status of the gearbox low-speed bearings of multiple units in the wind farm.
3. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 1 is characterized in that: The methods for filling missing values in wind turbine operating status data include: The operating status data of each unit are matched and combined at the same time, ensuring that each column is a certain operating status data of the same unit, and the same row is a certain operating status data of multiple units at the same time; according to , and Fill in missing values, where is the missing value to be filled, is the local mean on the time scale, is the local mean on the spatial scale, n is the number of existing data near the missing value, m is the number of wind turbines, and p is the unit with missing data.
4. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 3 is characterized in that: After filling the missing values of the wind turbine operating status data, the meteorological data and the filled wind turbine operating status data are also normalized.
5. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 1 is characterized in that: The method for clustering analysis of wind turbine operating status data includes: clustering analysis of wind turbine operating status data using a self-organizing map neural network, and clustering wind turbines with similar operating status into one category.
6. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 1 is characterized in that: The method of building a gearbox temperature prediction model includes: Initialize the running parameters of the whale algorithm and randomly generate the position coordinates of individual whales; According to the individual whales, the least squares support vector machine was used to build a prediction model and calculate the temperature; Save the whale individual with the smallest calculation error; The position of individual whales is updated iteratively. In each iteration, the prediction model is rebuilt based on the updated individual whales, the temperature is calculated, and the individual whale with the smallest calculation error is saved. After the iteration, the prediction model constructed with the optimal whale individual is obtained.
7. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 6 is characterized in that: The objective function of the least squares support vector machine is: , , ; in is the objective function of the least squares support vector, γ is the penalty factor, represents the loss function, W is the normal vector of the hyperplane, φ(x k ) is the kernel function, x k is the support vector, b is a constant term, M is the number of support vectors, is the support vector output, is the kernel function width; Update and utilization of individual whale positions in the whale algorithm , Calculate, where D is the distance between the current position and the global optimal position of the population, X i (t) is the position of the ith whale at the tth iteration, X * (t) is the global optimal position of the whale population at the tth iteration, A is the convergence factor, and C is the swing factor.
8. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 1 is characterized in that: The prediction error is calculated as follows: ,in, is the predicted value of the equipment operating status parameter, is the true value of the equipment operating status parameter, is the difference between the predicted value and the true value.
9. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 8 is characterized in that: The distribution characteristics of the prediction error and the confidence interval calculation method are as follows: , ;in is the confidence level [ ] is the lower boundary of the distribution range of the model prediction error, is the confidence level [ ] is the upper boundary of the distribution range of the model prediction error; the confidence interval is .
10. The gearbox fault early warning method based on wind power spatiotemporal adaptive clustering and similarity according to claim 9 is characterized in that: The confidence level is 99.73%.