Machine Learning-Based Wind Turbine Operating Status Monitoring Method

Abstract

The present invention relates to the technical field of data processing, and particularly to a method for monitoring the operating state of a wind turbine based on machine learning. When the wind turbine is operating, this method performs logarithmic transformation on each data in the data sequence corresponding to multiple data indicators within a preset time period to obtain the corresponding logarithmic values, and constructs a scatter plot of all the logarithmic values; obtains the neighborhood radius and the minimum number of samples of the DBSCAN algorithm, and obtains the clustering results of all data points in the scatter plot according to the neighborhood radius and the minimum number of samples; obtains multiple boundary points according to the clustering results, obtains the isolation degree of each boundary point according to the distribution characteristics of the data points within the neighborhood radius of each boundary point, and obtains the abnormal points in the scatter plot according to the isolation degree of each boundary point; monitors the operating state of the wind turbine according to the abnormal points in the scatter plot, thereby improving the accuracy of the DBSCAN algorithm in monitoring the operating state of the wind turbine.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a method for monitoring the operating state of a wind turbine based on machine learning. Background Art

[0002] The monitoring of the operating state of a wind turbine plays a crucial role in the wind energy industry. Monitoring the operating state of a wind turbine can timely detect potential problems and abnormalities, so as to take preventive maintenance measures, reduce equipment damage and downtime, and improve the reliability and safety of the equipment.

[0003] Currently, various sensors are used to monitor the working states of various components of the wind turbine and environmental conditions in real time, and the operating data of the wind turbine is collected and transmitted in real time through a remote monitoring system. The collected data is analyzed and processed in real time to obtain the health indicators of the wind turbine, and the health state of the wind turbine is evaluated according to the health indicators. However, in the process of data collection and analysis, the data will contain noise points and abnormal points, and the DBSCAN algorithm is a powerful density clustering algorithm that classifies data points into core points, boundary points, and noise points according to the density connectivity between data points. In the traditional method, only the noise points are used as abnormal points to evaluate the health state of the wind turbine.

[0004] However, in this process, there are some abnormal points that are highly similar to normal data points in terms of features and are covered within the neighborhood radius of the core points as boundary points. The DBSCAN algorithm cannot accurately identify such situations, resulting in a reduction in the accuracy of anomaly detection.

[0005] Therefore, how to improve the accuracy of using the DBSCAN algorithm for monitoring the operating state of a wind turbine has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, an embodiment of the present invention provides a method for monitoring the operating state of a wind turbine based on machine learning to solve the problem of how to improve the accuracy of using the DBSCAN algorithm for monitoring the operating state of a wind turbine.

[0007] An embodiment of the present invention provides a method for monitoring the operating state of a wind turbine based on machine learning, and the method includes the following steps:

[0008] When the wind turbine is operating, at least two data indicators within a preset time period are respectively obtained, data sequences corresponding to each data indicator are obtained, each data in each of the data sequences is subjected to a logarithmic transformation to obtain corresponding logarithmic values, and a scatter plot of all the logarithmic values is constructed;

[0009] When clustering all data points in the scatter plot using the DBSCAN algorithm, obtain the neighborhood radius and minimum number of samples of the DBSCAN algorithm, and obtain the clustering result of all data points in the scatter plot according to the neighborhood radius and the minimum number of samples;

[0010] Obtain at least one boundary point in the scatter plot according to the clustering result. For any boundary point, obtain the isolation degree of the boundary point according to the distribution characteristics of the data points within the neighborhood radius of the boundary point. Obtain the isolation degrees of all boundary points in the scatter plot, and obtain the outlier points in the scatter plot according to the isolation degrees of all boundary points in the scatter plot;

[0011] Monitor the operating state of the wind turbine according to the outlier points in the scatter plot.

[0012] Preferably, the obtaining the clustering result of all data points in the scatter plot according to the neighborhood radius and the minimum number of samples includes:

[0013] Divide the data points in the scatter plot into core points, boundary points, and noise points according to the neighborhood radius and the minimum number of samples. Take each boundary point and noise point in the scatter plot as a complex point. For any core point, form a cluster class by combining the complex points belonging to the core point with the core point, and obtain at least one cluster class.

[0014] Preferably, the obtaining the isolation degree of the boundary point according to the distribution characteristics of the data points within the neighborhood radius of the boundary point includes:

[0015] Obtain the neighborhood data complexity coefficient of the boundary point according to the number of cluster classes included within the neighborhood radius of the boundary point;

[0016] Obtain the weighted density eigenvalue of the boundary point according to the Euclidean distance between the boundary point and the core point of the cluster class where the boundary point is located;

[0017] Perform weighted summation on the neighborhood data complexity coefficient and the weighted density eigenvalue to obtain the isolation degree of the boundary point.

[0018] Preferably, the obtaining the neighborhood data complexity coefficient of the boundary point according to the number of cluster classes included within the neighborhood radius of the boundary point includes:

[0019] Take each complex point within the neighborhood radius of the boundary point as a target complex point, count the number of target complex points, count the cluster classes to which each target complex point belongs to obtain the number of cluster classes, and take the ratio of the number of cluster classes to the number of target complex points as the neighborhood data complexity coefficient of the boundary point.

[0020] Preferably, obtaining the weighted density eigenvalue of the boundary point according to the Euclidean distance between the boundary point and the core point of the cluster class where the boundary point is located includes:

[0021] Obtaining the neighborhood area according to the neighborhood radius, and taking the ratio between the number of target complex points and the neighborhood area as a first variable;

[0022] Taking the core point to which the boundary point belongs as a target core point, calculating the Euclidean distance between the boundary point and the target core point, and taking the difference between the neighborhood radius and the Euclidean distance as a second variable;

[0023] Taking the product between the first variable and the second variable as the weighted density of the boundary point, calculating the average value of the weighted densities of all boundary points in the scatter plot, and obtaining the weighted density eigenvalue of the boundary point according to the average value.

[0024] Preferably, obtaining the weighted density eigenvalue of the boundary point according to the average value includes:

[0025] Calculating the absolute value of the difference between the weighted density of the boundary point and the average value as a first eigenvalue, and taking the first eigenvalue as the independent variable of the hyperbolic tangent function to obtain the weighted density eigenvalue of the boundary point.

[0026] Preferably, obtaining the outlier in the scatter plot according to the isolation degree of all boundary points in the scatter plot includes:

[0027] For any isolation degree, if the isolation degree is within the preset isolation degree range, then marking the boundary point corresponding to the isolation degree as an outlier.

[0028] The beneficial effects of the embodiments of the present invention compared with the prior art are:

[0029] When the wind turbine is operating, at least two data metrics within a preset time period are acquired respectively, and data sequences corresponding to each data metric are obtained. Each data in each of the data sequences is subjected to a logarithmic transformation to obtain corresponding logarithmic values, and a scatter plot of all the logarithmic values is constructed. When using the DBSCAN algorithm to cluster all the data points in the scatter plot, the neighborhood radius and the minimum number of samples of the DBSCAN algorithm are acquired, and the clustering result of all the data points in the scatter plot is obtained according to the neighborhood radius and the minimum number of samples. At least one boundary point in the scatter plot is obtained according to the clustering result. For any boundary point, the degree of isolation of the boundary point is obtained according to the distribution characteristics of the data points within the neighborhood radius of the boundary point. The degrees of isolation of all the boundary points in the scatter plot are obtained, and the outlier points in the scatter plot are obtained according to the degrees of isolation of all the boundary points in the scatter plot. The operating state of the wind turbine is monitored according to the outlier points in the scatter plot. Among them, when the wind turbine is operating, two data metrics are acquired, and the data in the data metrics are logarithmically transformed to enhance the distribution characteristics of the data points, reduce the interval between the data points, make the densely distributed data points more prominent and recognizable, and at the same time reduce the sensitivity of the data to outliers, which is helpful for analysis and identification. A scatter plot of all the logarithmically transformed values is constructed. At this time, the axis range of the scatter plot can effectively display the distribution of the data points without including too much or too little blank space, optimize the clustering effect of the DBSCAN algorithm, obtain boundary points on the scatter plot using the DBSCAN algorithm, analyze the degrees of isolation of all the boundary points and the types and distribution characteristics of the data points within the neighborhood radius of the boundary points, calculate the degrees of isolation of the boundary points, mark the boundary points that meet the requirements as outlier points, complete the discrimination of all the boundary points, and improve the accuracy of the DBSCAN algorithm for monitoring the operating state of the wind turbine. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0031] Figure 1 FIG. 9 is a flowchart of a method for monitoring the operating state of a wind turbine based on machine learning provided in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] The following details the embodiments of the present disclosure, and the examples of the embodiments are shown in the drawings. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present disclosure and should not be construed as limiting the present disclosure.

[0033] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are only examples of devices and methods consistent with some aspects of the present disclosure.

[0034] In order to illustrate the technical solution of the present invention, specific embodiments will be used for illustration below.

[0035] See Figure 1 , which is a method flowchart of a method for monitoring the operating state of a wind turbine based on machine learning provided in the first embodiment of the present invention. As Figure 1 shown, the method may include:

[0036] Step S101, when the wind turbine is operating, respectively obtain at least two data metrics within a preset time period, obtain the data sequence corresponding to each data metric, perform a logarithmic transformation on each data in each data sequence to obtain the corresponding logarithm value, and construct a scatter plot of all logarithm values.

[0037] Because the amount of data on the operating state of the wind turbine is large, using a scatter plot can help analyze the relationship between different operating parameters of the wind turbine. At the same time, the scatter plot can also help identify and locate abnormal data points. However, if a scatter plot is directly constructed with time as the X-axis and the data size as the Y-axis, the scatter plot will become very crowded, covering up some distribution characteristics of the data and the correlation between the data. Moreover, the data points show a linear change as a whole in the scatter plot, and the density change is not obvious. As a result, when using the DBSCAN algorithm to cluster and divide the data in the scatter plot, problems such as over-segmentation or failure to reasonably divide may occur. Therefore, when the wind turbine is operating, respectively obtain at least two data metrics within a preset time period, obtain the data sequence corresponding to each data metric, perform a logarithmic transformation on each data in each data sequence to obtain the corresponding logarithm value, and construct a scatter plot of all logarithm values.

[0038] Among them, the specific process of obtaining the data sequence of each data metric is:

[0039] First, consider the correlation in physical meaning between pairwise data indicators, and obtain two data indicators that are required and have a correlation in physical meaning. For example, the relationship between rotational speed and power: the higher the rotational speed, the greater the power; the relationship between temperature and humidity: the higher the temperature, the lower the humidity, etc. Set the preset time period to 1 hour, which is not limited here, and the implementer can set it according to the implementation scenario. According to these two data indicators, obtain the data sequences corresponding to each data indicator within 1 hour. To ensure the accuracy and consistency of each data in the data sequences, after obtaining all the data sequences, perform data preprocessing on each data in all the data sequences to achieve the purpose of data cleaning. Among them, data preprocessing includes but is not limited to: data cleaning, filling in missing values, deleting samples or features where missing values are located, detecting and handling outliers, deleting duplicate samples or features; data normalization to ensure that the data conforms to specific standards or specifications, such as unit consistency, unified data format, etc.; data transformation, for example, normalizing or standardizing the data to ensure the comparability of different indicators. Among them, data preprocessing techniques belong to the prior art and will not be elaborated here.

[0040] Among them, perform a logarithmic transformation on each data in each data sequence to obtain the corresponding logarithmic value, and construct a scatter plot of all the logarithmic values. The specific process is as follows:

[0041] To enhance the change trend of the data in the data sequence, reduce the data spacing, make the densely distributed data more prominent and recognizable, and help subsequent analysis and outlier identification, perform a base-10 logarithmic transformation on each data in the data sequence corresponding to each data indicator to obtain the corresponding logarithmic value, and construct a scatter plot of all the logarithmic values. Then the calculation formula for the logarithmic value is:

[0042]

[0043] Among them, y represents the value after logarithmic transformation, that is, the logarithmic value, and x represents the data in the data sequence.

[0044] It should be noted that if two data indicators to be obtained have a correlation in physical meaning, such as the relationship between rotational speed and power, the relationship between temperature and humidity, etc., then these two data indicators are respectively used as the X-axis and Y-axis of the scatter plot. If there is only one data indicator to be obtained, or two or more data indicators to be obtained do not have a correlation in physical meaning, then time is used as the X-axis of the scatter plot, and the corresponding logarithmic values in the data sequence corresponding to the data indicator are used as the Y-axis of the scatter plot.

[0045] Step S102, when clustering all the data points in the scatter plot using the DBSCAN algorithm, obtain the neighborhood radius and the minimum number of samples of the DBSCAN algorithm, and obtain the clustering results of all the data points in the scatter plot according to the neighborhood radius and the minimum number of samples.

[0046] First, determine the neighborhood radius and minimum number of samples of the DBSCAN algorithm through empirical adjustment or verification techniques, and then run the DBSCAN algorithm on the scatter plot for clustering to obtain the clustering results of all data points in the scatter plot, which are used to screen out abnormal points.

[0047] Among them, the clustering results of all data points in the scatter plot include:

[0048] According to the neighborhood radius and the minimum number of samples, divide the data points in the scatter plot into core points, boundary points, and noise points. Take each boundary point and noise point in the scatter plot as complex points. For any core point, combine the complex points belonging to the core point with the core point to form a cluster class, and obtain at least one cluster class.

[0049] The DBSCAN algorithm divides all data points in the scatter plot into core points, boundary points, and noise points according to the neighborhood radius and the minimum number of samples. Among them, data points with the number of sample points within the neighborhood radius greater than or equal to the minimum number of samples are core points. Data points that do not belong to core points but are within the neighborhood radius of a certain core point are boundary points. Data points that are neither core points nor boundary points are noise points. Take each boundary point and each noise point in the scatter plot as complex points. Each core point represents a cluster class. For any core point, the complex points belonging to the core point are the cluster members of the cluster class where the core point is located. Similarly, all data points in the scatter plot are divided into at least one cluster class. It should be noted that the DBSCAN algorithm belongs to the prior art and will not be elaborated here.

[0050] Step S103, obtain at least one boundary point in the scatter plot according to the clustering result. For any boundary point, obtain the isolation degree of the boundary point according to the distribution characteristics of the data points within the neighborhood radius of the boundary point, obtain the isolation degrees of all boundary points in the scatter plot, and obtain the abnormal points in the scatter plot according to the isolation degrees of all boundary points in the scatter plot.

[0051] There may be some abnormal points among the boundary points. These abnormal points are highly similar to normal data points in terms of features and are covered within the neighborhood radius of the core points during the operation of the DBSCAN algorithm, making it impossible for the DBSCAN algorithm to identify them. That is, these abnormal points are not classified as noise points but as boundary points, resulting in a decrease in the accuracy of anomaly detection. Therefore, in the example of the present invention, according to step S102, the clustering results of all data points in the scatter plot can be obtained, and then each boundary point in the scatter plot can be obtained based on the clustering results. By analyzing the isolation degree of each boundary point and the types and distribution characteristics of the data points within the neighborhood radius of the boundary point, the isolation degree of each boundary point is calculated, and the boundary points that meet the requirements are marked as abnormal points, completing the discrimination of all boundary points and improving the accuracy of anomaly detection by the DBSCAN algorithm.

[0052] Among them, taking any boundary point as an example, the specific process of calculating the isolation degree of this boundary point is as follows:

[0053] (1) Obtain the neighborhood data complexity coefficient of the boundary point according to the number of cluster classes included within the neighborhood radius of the boundary point, including:

[0054] Regarding each complex point within the neighborhood radius of the boundary point as a target complex point, count the number of the target complex points, count the cluster classes to which each of the target complex points belongs to obtain the number of cluster classes, and take the ratio of the number of cluster classes to the number of target complex points as the neighborhood data complexity coefficient of the boundary point.

[0055] In an embodiment, for the i-th boundary point in the scatter plot, count the number of all data points except the core points within the neighborhood radius of the i-th boundary point, that is, the number of target complex points, count the cluster classes where each target complex point is located to obtain the number of cluster classes, and obtain the neighborhood data complexity coefficient of the i-th boundary point according to the number of target complex points and the number of cluster classes. Then the calculation formula for the neighborhood data complexity coefficient of the i-th boundary point is:

[0056]

[0057] Among them, represents the neighborhood data complexity coefficient of the i-th boundary point, represents the number of cluster classes corresponding to the target complex point corresponding to the i-th boundary point, represents the number of target complex points corresponding to the i-th boundary point.

[0058] It should be noted that, the larger, the more the number of cluster classes existing within the neighborhood range of the i-th boundary point, The closer it is to 1, that is, the closer the neighborhood data complexity coefficient of the $i$-th boundary point is to 1, it indicates that this boundary point may be closer to other cluster classes or belong to the boundary region of multiple cluster classes, and the greater the possibility of being an outlier or an ambiguous data point.

[0059] (2)Obtaining the weighted density eigenvalue of the boundary point according to the Euclidean distance between the boundary point and the core point of the cluster class where the boundary point is located, including:

[0060] Obtaining the neighborhood area according to the neighborhood radius, and taking the ratio between the number of target complex points and the neighborhood area as the first variable;

[0061] Taking the core point to which the boundary point belongs as the target core point, calculating the Euclidean distance between the boundary point and the target core point, and taking the difference between the neighborhood radius and the Euclidean distance as the second variable;

[0062] Taking the product between the first variable and the second variable as the weighted density of the boundary point, calculating the average value of the weighted densities of all boundary points in the scatter plot, and obtaining the weighted density eigenvalue of the boundary point according to the average value.

[0063] In an embodiment, calculate the area covered by the neighborhood radius, calculate the Euclidean distance between the $i$-th boundary point and the core point to which the $i$-th boundary point belongs, and obtain the weighted density of the $i$-th boundary point according to the area and the Euclidean distance. Among them, the Euclidean distance is prior art and will not be elaborated here. Then the calculation formula for the weighted density of the $i$-th boundary point is:

[0064]

[0065] Among them, represents the weighted density of the $i$-th boundary point, represents the number of target complex points corresponding to the $i$-th boundary point, represents the neighborhood radius, represents the area covered by the neighborhood radius, represents the Euclidean distance between the $i$-th boundary point and the core point to which the $i$-th boundary point belongs.

[0066] It should be noted that represents the comprehensive situation of the Euclidean distance between the $i$-th boundary point and the core point and the data density within the neighborhood radius of the $i$-th boundary point. The smaller the weighted density, the farther the $i$-th boundary point is from the core point, and the fewer the data points within the neighborhood radius of the $i$-th boundary point, and the more likely it is to be an outlier.

[0067] Similarly, the weighted density of all boundary points in the scatter plot is calculated to obtain the average value of the weighted density, and then the weighted density eigenvalue of the i-th boundary point is obtained according to the average value of the weighted density and the weighted density of the i-th boundary point.

[0068] The weighted density characteristic value of the boundary point is obtained according to the average value, including:

[0069] The absolute value of the difference between the weighted density of the boundary point and the average value is calculated as a first eigenvalue, and the first eigenvalue is used as an independent variable of a hyperbolic tangent function to obtain a weighted density eigenvalue of the boundary point.

[0070] In one embodiment, the weighted density eigenvalue of the i-th boundary point is obtained according to the average value of the weighted densities of all boundary points in the scatter plot and the weighted density of the i-th boundary point. The calculation formula of the weighted density eigenvalue is:

[0071]

[0072] in, represents the weighted density eigenvalue of the i-th boundary point, tanh() represents the hyperbolic tangent function, represents the weighted density of the i-th boundary point, represents the weighted density of the j-th boundary point, represents the number of all boundary points in the scatter plot, Indicates the absolute value symbol.

[0073] It should be noted that represents the average value of the weighted density of all boundary points in the scatter plot, The larger the value, the greater the difference between the i-th boundary point and the overall density of all boundary points in the scatter plot. The further the i-th boundary point is from the core point to which it belongs, the more likely it is an outlier. The tanh() function is used to convert The value of is fixed between [0, 1].

[0074] (3) Performing a weighted summation of the neighborhood data complexity coefficient and the weighted density eigenvalue to obtain the degree of isolation of the boundary point.

[0075] In one embodiment, the degree of isolation of the i-th boundary point is obtained based on the neighborhood data complexity coefficient and the weighted density feature value of the i-th boundary point. The calculation formula of the isolation degree is:

[0076]

[0077] in, Indicates the degree of isolation of the i-th boundary point, represents the weighted density eigenvalue of the i-th boundary point, represents the neighborhood data complexity coefficient of the i-th boundary point, represents the first weight value, represents the second weight value.

[0078] It should be noted that, set to 0.65, set to 0.35, there is no limitation here, and the implementer can set according to the implementation scenario, and then the value range of is (0, 1), the closer it is to 1, it indicates that there are noise points within the neighborhood radius of the i-th boundary point and the number of cluster classes within the neighborhood radius is relatively large. The i-th boundary point is farther away from the core point and the number of data points within the neighborhood radius is smaller. Furthermore, the i-th boundary point is more likely to be distributed in a sparse density area and is more likely to be an outlier. Similarly, the isolation degree of each boundary point is analyzed to obtain the isolation degree of all boundary points in the scatter plot. Further, according to the isolation degree of all boundary points in the scatter plot, the outliers in the scatter plot are obtained, including:

[0079] For any isolation degree, if the isolation degree is within the preset isolation degree range, then mark the boundary point corresponding to the isolation degree as an outlier.

[0080] In an implementation manner, set the preset isolation degree range to (0.75, 1), there is no limitation here, and the implementer can set according to the implementation scenario. For any boundary point, if the outlier degree of this boundary point is within (0.75, 1), then mark this boundary point as an outlier. Similarly, the isolation degree of each boundary point is analyzed to obtain all outliers in the scatter plot.

[0081] Step S104, monitor the operating state of the wind turbine according to the outliers in the scatter plot.

[0082] According to the outliers screened out from all boundary points in the scatter plot in step S103, further, consider the noise points and outliers in the scatter plot as abnormal values during the operation of the wind turbine, and consider the core points and boundary points not marked as outliers in the scatter plot as normal values. Monitor the operating state of the wind turbine according to the abnormal values and normal values, and at the same time visually display the monitored data and analysis results to provide intuitive reports and charts. Take corresponding measures according to the monitoring results to ensure the safe and efficient operation of the wind turbine.

[0083] It is worth noting that the focus of the present invention is on how to screen out outliers from the boundary points obtained by clustering using the DBSCAN algorithm. Monitoring the operating state data of the wind turbine according to the abnormal values and taking corresponding measures belong to the prior art and will not be elaborated here.

[0084] In summary, the present invention provides a method for monitoring the operating state of a wind turbine based on machine learning, which includes obtaining at least two data metrics within a preset time period during the operation of the wind turbine, corresponding to obtaining data sequences for each data metric, performing logarithmic transformation on each data in each of the data sequences to obtain corresponding logarithmic values, and constructing a scatter plot of all the logarithmic values; when using the DBSCAN algorithm to cluster all the data points in the scatter plot, obtaining the neighborhood radius and the minimum number of samples of the DBSCAN algorithm, and obtaining the clustering result of all the data points in the scatter plot according to the neighborhood radius and the minimum number of samples; obtaining at least one boundary point in the scatter plot according to the clustering result, for any boundary point, obtaining the isolation degree of the boundary point according to the distribution characteristics of the data points within the neighborhood radius of the boundary point, obtaining the isolation degrees of all the boundary points in the scatter plot, and obtaining the abnormal points in the scatter plot according to the isolation degrees of all the boundary points in the scatter plot; monitoring the operating state of the wind turbine according to the abnormal points in the scatter plot. Among them, when the wind turbine is operating, two data metrics are obtained, and the data in the data metrics are logarithmically transformed to enhance the distribution characteristics of the data points, reduce the interval between the data points, make the densely distributed data points more prominent and recognizable, and at the same time make the data less sensitive to outliers, which is helpful for analysis and identification. A scatter plot of all the logarithmically transformed values is constructed. At this time, the axis range of the scatter plot can effectively display the distribution of the data points without containing too much or too little blank space, optimizing the clustering effect of the DBSCAN algorithm. The DBSCAN algorithm is used on the scatter plot to obtain boundary points, analyze the isolation degrees of all the boundary points and the types and distribution characteristics of the data points within the neighborhood radius of the boundary points, calculate the isolation degrees of the boundary points, mark the boundary points that meet the requirements as abnormal points, complete the discrimination of all the boundary points, and improve the accuracy of the DBSCAN algorithm for monitoring the operating state of the wind turbine.

[0085] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for monitoring the operating state of a wind turbine based on machine learning, characterized in that, The method for monitoring the operating state of a wind turbine based on machine learning includes: When the wind turbine is operating, at least two data metrics within a preset time period are respectively acquired, data sequences corresponding to each data metric are obtained, each data in each of the data sequences is logarithmically transformed to obtain corresponding logarithmic values, and a scatter plot of all the logarithmic values is constructed; When clustering all the data points in the scatter plot using the DBSCAN algorithm, the neighborhood radius and the minimum number of samples of the DBSCAN algorithm are acquired, and the clustering result of all the data points in the scatter plot is obtained according to the neighborhood radius and the minimum number of samples; At least one boundary point in the scatter plot is obtained according to the clustering result. For any boundary point, the isolation degree of the boundary point is obtained according to the distribution characteristics of the data points within the neighborhood radius of the boundary point, the isolation degrees of all the boundary points in the scatter plot are obtained, and the abnormal points in the scatter plot are obtained according to the isolation degrees of all the boundary points in the scatter plot; The operating state of the wind turbine is monitored according to the abnormal points in the scatter plot; The obtaining the isolation degree of the boundary point according to the distribution characteristics of the data points within the neighborhood radius of the boundary point includes: Each complex point within the neighborhood radius of the boundary point is used as a target complex point, the number of the target complex points is counted, the cluster classes to which each of the target complex points belongs are counted to obtain the number of cluster classes, and the ratio of the number of cluster classes to the number of target complex points is used as the neighborhood data complexity coefficient of the boundary point; The neighborhood area is obtained according to the neighborhood radius, and the ratio between the number of target complex points and the neighborhood area is used as a first variable; the core point to which the boundary point belongs is used as a target core point, the Euclidean distance between the boundary point and the target core point is calculated, and the difference between the neighborhood radius and the Euclidean distance is used as a second variable; the product of the first variable and the second variable is used as the weighted density of the boundary point, the average value of the weighted densities of all the boundary points in the scatter plot is calculated, the absolute value of the difference between the weighted density of the boundary point and the average value is used as a first eigenvalue, and the first eigenvalue is used as the independent variable of the hyperbolic tangent function to obtain the weighted density eigenvalue of the boundary point; The neighborhood data complexity coefficient and the weighted density eigenvalue are weighted and summed to obtain the isolation degree of the boundary point.

2. The method for monitoring the operating state of a wind turbine based on machine learning according to claim 1, characterized in that The obtaining the clustering result of all the data points in the scatter plot according to the neighborhood radius and the minimum number of samples includes: The data points in the scatter plot are divided into core points, boundary points, and noise points according to the neighborhood radius and the minimum number of samples. Each boundary point and noise point in the scatter plot is used as a complex point. For any core point, the complex points belonging to the core point and the core point are combined into a cluster class to obtain at least one cluster class.

3. The method for monitoring the operating state of a wind turbine based on machine learning according to claim 1, characterized in that, The obtaining the abnormal points in the scatter plot according to the isolation degrees of all the boundary points in the scatter plot includes: For any degree of isolation, if the degree of isolation is within the preset range of the degree of isolation, mark the boundary point corresponding to the degree of isolation as an abnormal point.

Images