Abnormal state detection and joint interpretation method and device for wind generating set

Abstract

The invention discloses a wind generating set abnormal state detection and joint interpretation method and device. The method comprises the steps that multi-dimensional operation data are acquired from an SCADA system of a wind generating set and preprocessed; clustering by adopting a k-means clustering algorithm, and selecting the calculated center features of the center cluster to construct a new feature data set; training a wind turbine generator abnormal state detection model by using a classification decision lifting algorithm, and detecting and identifying an abnormal state; establishing a simplified model output by the detection model, calculating a global contribution degree of an input feature to an abnormal state detection result by using an SHAP algorithm, and providing global explanation; establishing a local model of the detection model instance, generating local explanation for a specific abnormal instance by adopting an LIME algorithm, and analyzing contribution of key features in the specific abnormal instance; according to the method, the accuracy and interpretability of anomaly detection of the wind generating set are improved, so that solid data support and theoretical basis are provided for maintenance and optimization of a complex system.

Description

Technical Field

[0001] The present invention relates to the field of wind power generation technology, in particular to the field related to wind generator set state monitoring and abnormality detection technology, and more specifically to a method and device for abnormal state detection and joint interpretation of a wind generator set. Background Art

[0002] As the proportion of wind power in the global energy structure continues to rise, the scale of wind farms has gradually expanded, and the number and complexity of wind turbines have also increased significantly. In large wind farms, wind turbines often operate in clusters, and their operating status is directly related to the power generation efficiency and safety of the entire wind farm. However, due to the complex operating environment of wind turbines and the variety of fault types, traditional monitoring and anomaly detection methods face many challenges in practical applications.

[0003] Traditional methods mainly rely on the construction of physical models and rule-based anomaly detection. Physical models usually require accurate equipment parameters and operating environment information, which is often difficult to obtain in practice or is limited by the complexity and nonlinear characteristics of the model, resulting in the accuracy and timeliness of anomaly detection being limited; rule-based detection methods rely on predefined thresholds and empirical rules, which makes it difficult to meet the detection needs of new or complex faults.

[0004] With the development of data-driven technology, anomaly detection methods based on machine learning can automatically learn the normal operating modes and abnormal characteristics of wind turbines by utilizing a large amount of historical operating data, and realize real-time detection of abnormal conditions; however, traditional machine learning methods often face two major problems in application: first, when dealing with high-dimensional data and complex features, they are prone to fall into a "black box" state and lack sufficient explanatory ability, making it difficult for operation and maintenance personnel to understand and trust the detection results; second, data missing and noise problems will significantly affect the performance of the detection model, resulting in a decrease in the accuracy and reliability of anomaly detection.

[0005] Therefore, how to provide a method and device for abnormal state detection and joint interpretation of wind turbines that can improve the accuracy of abnormality detection of wind turbines, overcome the lack of interpretability caused by the "black box" effect of traditional machine learning methods, and deal with missing values ​​and noise problems in high-dimensional data is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0006] In view of this, the present invention provides a method and device for detecting and jointly interpreting abnormal conditions of a wind turbine generator set to solve some of the technical problems mentioned in the background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set comprises the following steps:

[0009] S1. Obtain multi-dimensional operation data from the SCADA system of the wind turbine and perform pre-processing;

[0010] S2. The pre-processed wind turbine SCADA data is clustered using the k-means clustering algorithm, and the central features of the calculated central cluster are selected to construct a new feature data set;

[0011] S3. Based on the new feature data set constructed from the extracted central feature data, the classification decision boosting algorithm is used to train the abnormal state detection model of the wind turbine to detect and identify the abnormal state;

[0012] S4. Establish a simplified model of the output of the abnormal state detection model of the wind turbine, use the SHAP algorithm to calculate the global contribution of the input features to the abnormal state detection results, and provide a global explanation;

[0013] S5. Establish a local model of the wind turbine abnormal state detection model instance, use the LIME algorithm to fit the local linear model, generate local explanations for specific abnormal instances, and analyze the contribution of key features in specific abnormal instances.

[0014] Preferably, the specific contents of the preprocessing in step S1 include:

[0015] S11. Based on the SCADA data of wind turbines, the missing classification features in the data are filled using the majority filling method;

[0016] S12. Based on the SCADA data of the wind turbine generator set, the mean filling method is used to fill the missing numerical features in the data;

[0017] S13. The filled SCADA data of the wind turbine generator set, i.e., the pre-processed SCADA data of the wind turbine generator set, is used as a pre-training data set.

[0018] Preferably, the specific content of step S2 is:

[0019] S21. The optimal number of clusters is determined by using the elbow method. The total sum of squared deviations under different k values ​​is calculated, and the k value with which the total sum of squared deviations is significantly reduced is selected as the optimal number of clusters.

[0020] S22. Clustering the wind turbine characteristics in the pre-trained data set of the pre-processed wind turbine SCADA data according to the determined optimal number of clusters;

[0021] S23. According to the completed wind turbine feature clustering results, calculate the distance from the feature to the cluster center, and select the feature with the smallest distance as the central feature;

[0022] S24. Use the central features selected by clustering to construct a new feature set to reduce collinearity and redundancy between features.

[0023] Preferably, the specific content of step S3 is:

[0024] S31. Using the classification decision boosting algorithm to establish a wind turbine abnormal state detection model, starting from a constant model, the classification prediction value is initialized to the mean of the target value;

[0025] S32. Calculate the gradient of each sample and the diagonal elements of the Hessian matrix in each iteration based on the initialized classification prediction value;

[0026] S33. construct a histogram for each splitting point of each feature, calculate the gain of each splitting point using the histogram value, select the splitting point with the largest gain for splitting, and then construct a decision tree based on the selected optimal splitting point;

[0027] S34. Use the currently constructed tree to update the prediction value of the model. After multiple iterations, the final model is the weighted sum of all base learners.

[0028] Preferably, the specific content of step S4 is:

[0029] S41. Construct a simplified model of the abnormal state detection model of the wind turbine generator set, and approximate the output of the original model in the form of a linear combination;

[0030] S42. Calculate the difference between the simplified model and the original model to meet local accuracy, missingness, and consistency;

[0031] S43. Calculate the SHAP value of each feature in the simplified model, perform global visualization based on the SHAP value of each feature, and display the importance ranking of the features affecting the abnormal state of the wind turbine generator set and the correlation between the features in the figure.

[0032] Preferably, the specific content of step S5 is:

[0033] S51 generates a new data set by locally perturbing the original data to simulate the behavior of the model in the local space;

[0034] S52. The Euclidean distance is used to measure the similarity between the generated new data set sample and the original data sample, and is used to assign a weight to each perturbed sample. The higher the weight, the more similar the perturbed sample is to the original sample.

[0035] S53. A weighted linear regression model is used to fit the disturbance samples and the corresponding model outputs, and a local explanation model is obtained by minimizing the weighted loss function;

[0036] S54. Based on the constructed local explanation model, a single instance in the wind turbine SCADA data is analyzed to calculate the abnormal state probability of the single instance.

[0037] Preferably, the method for abnormal state detection and joint interpretation of a wind turbine generator set further includes: S6. Visualizing the abnormal state detection results and interpretation results, and triggering an alarm mechanism based on preset rules.

[0038] A wind turbine abnormal state detection and joint interpretation system, based on the wind turbine abnormal state detection and joint interpretation method, comprises a data acquisition module, a preprocessing module, a feature selection module, an abnormal state detection module, a global interpretation module, a local interpretation module and a visualization module;

[0039] The acquisition module is used to obtain multi-dimensional operation data from the SCADA system of the wind turbine;

[0040] A preprocessing module, used for preprocessing the acquired multi-dimensional operation data;

[0041] The feature selection module is used to cluster the pre-processed wind turbine SCADA data using the k-means clustering algorithm, and select the central features of the calculated central cluster to construct a new feature data set;

[0042] The abnormal state detection module is used to construct a new feature data set based on the extracted central feature data, train the abnormal state detection model of the wind turbine using the classification decision boosting algorithm, and detect and identify the abnormal state;

[0043] The global explanation module is used to establish a simplified model of the output of the abnormal state detection model of the wind turbine, calculate the global contribution of the input features to the abnormal state detection results using the SHAP algorithm, and provide a global explanation;

[0044] The local explanation module is used to establish a local model of the abnormal state detection model instance of the wind turbine. The LIME algorithm is used to fit the local linear model to generate local explanations for specific abnormal instances and analyze the contribution of key features in specific abnormal instances.

[0045] The visualization module is used to visualize the detection and interpretation results and trigger the alarm mechanism.

[0046] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set.

[0047] A processing terminal comprises a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, the abnormal state detection and joint interpretation method of a wind turbine generator set is implemented.

[0048] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and device for detecting and jointly interpreting abnormal states of wind turbines. Based on data-driven machine learning methods and interpretability models, by integrating KMeans clustering algorithm and classification decision improvement algorithm, it can effectively extract central features in wind turbine operation data, build an efficient anomaly detection model, and use SHAP and LIME algorithms to provide global and local result interpretations, thereby overcoming the problem of "black box" models being difficult to understand and lacking transparency in traditional methods.

[0049] For different types of missing data and noise, mode filling and mean filling methods are used to effectively handle them, which improves data quality and enhances the robustness and applicability of the model.

[0050] In the complex and multi-dimensional operating environment of wind turbines, the present invention can not only accurately detect abnormal conditions, but also provide detailed explanations and analyses for operation and maintenance personnel, quickly respond to abnormal conditions, reduce the risk of failure, and improve the overall operating efficiency and stability of the wind farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0052] Figure 1 A schematic diagram of a wind turbine abnormal state detection and joint interpretation method provided by the present invention;

[0053] Figure 2 A schematic diagram of the framework of the abnormal state detection and joint interpretation method of a wind turbine generator set provided by the present invention;

[0054] Figure 3 A schematic diagram of calculating central features using the k-means clustering algorithm provided by the present invention;

[0055] Figure 4A schematic diagram of feature selection for the k-means clustering algorithm provided by the present invention;

[0056] Figure 5 A schematic diagram of the accuracy of the abnormal state detection model of a wind turbine generator set based on the classification decision improvement algorithm provided by the present invention;

[0057] Figure 6 A schematic diagram of a confusion matrix of a wind turbine abnormal state detection model based on a classification decision boosting algorithm provided by the present invention;

[0058] Figure 7 A schematic diagram of the predicted recall rate of the wind turbine abnormal state detection model based on the classification decision boosting algorithm provided by the present invention;

[0059] Figure 8 A schematic diagram of the SHAP global feature interpretation provided by the present invention;

[0060] Fig. 9 A schematic diagram of the correlation between the SHAP features provided by the present invention;

[0061] Fig.10 A schematic diagram of the abnormal state prediction of a single LIME instance provided by the present invention;

[0062] Fig.11 Schematic diagram of the impact features of a single LIME instance provided by the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] The embodiment of the present invention discloses a method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set. Figure 1 and Figure 2 , including the following steps:

[0065] S1. Obtain multi-dimensional operation data from the SCADA system of the wind turbine and perform pre-processing;

[0066] S2. The pre-processed wind turbine SCADA data is clustered using the k-means clustering algorithm, and the central features of the calculated central cluster are selected to construct a new feature data set;

[0067] S3. Based on the new feature data set constructed from the extracted central feature data, the classification decision boosting algorithm is used to train the abnormal state detection model of the wind turbine to detect and identify the abnormal state;

[0068] S4. Establish a simplified model of the output of the abnormal state detection model of the wind turbine, use the SHAP algorithm to calculate the global contribution of the input features to the abnormal state detection results, and provide a global explanation;

[0069] S5. Establish a local model of the wind turbine abnormal state detection model instance, use the LIME algorithm to fit the local linear model, generate local explanations for specific abnormal instances, and analyze the contribution of key features in specific abnormal instances.

[0070] In this embodiment, the multi-dimensional operating data includes but is not limited to parameters such as wind speed, temperature, current, and voltage.

[0071] In order to further implement the above technical solution, the specific contents of the preprocessing in step S1 include:

[0072] S11. Based on the SCADA data of wind turbines, the missing classification features in the data are filled using the majority filling method;

[0073] S12. Based on the SCADA data of the wind turbine generator set, the mean filling method is used to fill the missing numerical features in the data;

[0074] S13. The filled SCADA data of the wind turbine generator set, i.e., the pre-processed SCADA data of the wind turbine generator set, is used as a pre-training data set.

[0075] In this embodiment, the mode filling method is specifically:

[0076]

[0077] Among them, Mode X is the mode of feature X, c is a specific category of categorical features, I(x i =c) is the indicator function, when x i When it belongs to category c, the value of the indicator function is 1, otherwise it is 0;

[0078] The mean imputation method is as follows:

[0079]

[0080] Where X is a set of numerical features, n is the number of all values ​​in the numerical feature set, m is the number of missing values ​​in the numerical feature set, and x i is the i-th non-missing value, μ X is the mean of feature X.

[0081] In order to further implement the above technical solution, the specific content of step S2 is:

[0082] S21. The optimal number of clusters is determined by using the elbow method. The total sum of square deviations under different k values ​​is calculated, and the k value with which the total sum of square deviations is significantly reduced is selected as the optimal number of clusters.

[0083] S22. Based on the determined optimal number of clusters, the k-means clustering algorithm is used to cluster the wind turbine characteristics in the pre-trained data set of the pre-processed wind turbine SCADA data, such as Figure 3 ;

[0084] S23. According to the completed wind turbine feature clustering results, calculate the distance from the feature to the cluster center, and select the feature with the smallest distance as the central feature, such as Figure 4 ;

[0085] S24. Use the central features selected by clustering to construct a new feature set to reduce collinearity and redundancy between features.

[0086] In this embodiment, in step S21, the total deviation sum of squares is:

[0087]

[0088] Among them, C i is the i-th cluster, μ is the center point of the i-th cluster;

[0089] In step S23, the selection equation of the central feature is:

[0090]

[0091] Among them, x * is the central feature, i is the ith cluster, μ is the center point of the ith cluster, and the intra-cluster features are {x1,x2,…,x m}.

[0092] like Figure 3 , shows the clustering results after reducing high-dimensional data to three-dimensional space through principal component analysis (PCA). Different colors represent different clustering labels. The coordinate axes represent the three principal components. The positions of data points are determined by the scores of these principal components. The colors from purple to yellow represent the differences in clustering labels, showing the distribution and clustering of data points in the space after dimensionality reduction.

[0093] like Figure 4, shows the variance changes of different features. The horizontal axis represents the feature index, and the vertical axis represents the variance of each feature. The blue line in the figure represents the variance of all original features, and the red dots mark the selected features. The selected features are distributed in areas with larger variances, indicating that these features may have higher representativeness or distinguishing ability in the overall data.

[0094] In order to further implement the above technical solution, the specific content of step S3 is:

[0095] S31. Using the classification decision boosting algorithm to establish a wind turbine abnormal state detection model, starting from a constant model, the classification prediction value is initialized to the mean of the target value;

[0096] The equation for initializing the classification prediction value to the mean of the target value is:

[0097]

[0098] in, is the predicted value, y i is a single target value, i is the number of the specific target value, and n is the number of the overall target values;

[0099] S32. Calculate the gradient of each sample and the diagonal elements of the Hessian matrix in each iteration based on the initialized classification prediction value;

[0100] The calculation equations for the sample gradient and the diagonal elements of the Hessian matrix are:

[0101]

[0102] Among them, g j is the gradient, is the loss function, y j is the actual label value of the sample, is the predicted value of the model, h j are the diagonal elements of the Hessian matrix;

[0103] S33. construct a histogram for each splitting point of each feature, calculate the gain of each splitting point using the histogram value, select the splitting point with the largest gain for splitting, and then construct a decision tree based on the selected optimal splitting point;

[0104] The histogram value construction equation is:

[0105]

[0106] Where b represents the candidate value of the split point;

[0107] The equation for calculating the gain of each split point from the histogram value is:

[0108]

[0109] Among them, b is the candidate value of the split point, Gain is the gain function, is the gradient sum of all samples in the left child node, is the gradient sum of all samples in the right child node, is the Hessian matrix sum of all samples in the left child node, is the Hessian matrix sum of all samples in the right child node, λ is the regularization parameter, and γ is the splitting penalty parameter;

[0110] S34. Use the currently constructed tree to update the prediction value of the model. After multiple iterations, the final model is the weighted sum of all base learners.

[0111] The iterative training equation is:

[0112]

[0113] The weighted sum equation of the base learner is:

[0114]

[0115] Among them, η is the learning rate, f m is the weak learner trained in the mth round.

[0116] like Figure 5 As shown in the figure, the ROC curve of the classification effect of the constructed abnormal state detection model of wind turbine generator set in different categories is displayed, with the horizontal axis being the false positive rate and the vertical axis being the true positive rate; the figure includes the ROC curves of category 0 and category 1, and the AUC is 0.99, indicating that the classification performance of the model in these two categories is very good; in addition, the AUC of the micro-average ROC curve is 1.00, and the AUC of the macro-average ROC curve is 0.99. On the whole, the classification effect of the constructed abnormal state detection model of wind turbine generator set in all categories is close to perfect, indicating that it has extremely high discrimination ability.

[0117] like Figure 6As shown in the figure, it is the classification report of the constructed abnormal state detection model of wind turbine generator set in different categories, showing the precision, recall rate, F1 value and support number; the horizontal axis represents the four evaluation indicators, and the vertical axis represents the classification results of category 0 and category 1: for category 1, the precision rate is 0.997, the recall rate is 0.959, the F1 value is 0.978, and the support number is 413; for category 0, the precision rate is 0.988, the recall rate is 0.999, the F1 value is 0.993, and the support number is 1369; overall, the classification performance of the model on both types of data is very good, especially on category 0, the F1 value is close to perfect, indicating that the classification effect of the constructed abnormal state detection model of wind turbine generator set in most categories is relatively balanced, and the support number also shows that the number of samples in category 0 is large.

[0118] like Figure 7 As shown in the figure, the precision-recall curve of the constructed wind turbine abnormal state detection model is displayed, with the horizontal axis being the recall rate and the vertical axis being the precision rate. The solid line represents the binary precision-recall curve of the model, and the dotted line represents the average precision rate of 0.99. It can be seen that the precision of the model remains close to 1.0 within a higher recall rate range, indicating that the constructed wind turbine abnormal state detection model can maintain high precision and recall rates in most cases. The rapid decline of the curve only occurs in the area close to the recall rate of 1.0, indicating that in very rare cases, the precision of the model will drop slightly, but the overall performance is still very excellent.

[0119] In order to further implement the above technical solution, the specific content of step S4 is:

[0120] S41. Construct a simplified model of the abnormal state detection model of the wind turbine generator set, and approximate the output of the original model in the form of a linear combination;

[0121] The equation for the linear combination is:

[0122]

[0123] Among them, g(z′) is the output of the global explanation model, φ0 is a constant term, which represents the model output when all features are missing, and φ i is the importance value of feature i, representing the contribution of the feature to the model output, z′ is the simplified input variable, usually binary 0 or 1, indicating whether the feature exists, and M is the total number of features;

[0124] S42. Calculate the difference between the simplified model and the original model to meet local accuracy, missingness, and consistency;

[0125] The local accuracy equation is:

[0126]

[0127] The missing equation is:

[0128]

[0129] The consistency equation is:

[0130]

[0131] Where f(z′) is the output of the original model under the simplified input z′, f′(z′) is the output of the improved model under the simplified input z', and z′\i is the feature i removed from the simplified input z';

[0132] S43. Calculate the SHAP value of each feature in the simplified model, perform global visualization based on the SHAP value of each feature, and display the importance ranking of the features that affect the abnormal state of the wind turbine generator set and the correlation between the features in the figure, such as Figure 8 and Fig. 9 ;

[0133] The calculation equation of SHAP value is:

[0134]

[0135] Among them, φ i (f, x) is the SHAP value of feature i in model f, indicating its contribution to the model output, φ i (f′, x) is the SHAP value of feature i in the improved model f′.

[0136] Figure 8 , shows the feature importance analysis based on SHAP (Shapley Additive Explanations) values. The horizontal axis is the SHAP value of the model output, which indicates the contribution of each feature to the model prediction results. The vertical axis lists different features, such as "Gearbox_Oil-2_Temperature", "Tower Acceleration Lateral", etc. Each point in the figure represents the SHAP value of a data point, and the color indicates the high and low feature values, from blue (low) to red (high). The shape corresponding to each feature represents the SHAP value distribution of the feature. The wider the part, the more samples there are in this range. Through this figure, we can see the positive and negative effects of certain features on the model output in different value ranges. The redder the point, the greater the impact of high feature values ​​on positive SHAP values, and the bluer the point, the greater the impact of low feature values ​​on negative SHAP values. This figure can observe the contribution of each feature to the model prediction and the changes of these contributions in different feature value ranges.

[0137] Fig. 9, shows the SHAP value distribution of the feature "TemperatureBatteryBox-1", which is used to explain the contribution of this feature to the model prediction results. The horizontal axis represents the value range of "TemperatureBatteryBox-1", and the vertical axis represents the corresponding SHAP value. The color bar represents the value of another feature "Gearbox_Oil-2_Temperature", which gradually changes from blue (low temperature) to red (high temperature); from the figure, we can see that the vast majority of data points are concentrated in the lower "TemperatureBattery Box-1" value range, and the SHAP values ​​of these points are close to 0, indicating that within this range, the feature contributes less to the model prediction, while in the high value area (close to 100,000), the SHAP value of the point increases or decreases slightly, indicating that in these cases "TemperatureBatteryBox-1" has more influence on the model prediction. The figure reflects the impact of this feature on the model output in different value ranges. The color bar combined with the temperature value of another feature can show the impact of the interaction between multiple features on the model prediction.

[0138] In order to further implement the above technical solution, the specific content of step S5 is:

[0139] S51 generates a new data set by locally perturbing the original data to simulate the behavior of the model in the local space;

[0140] The equation for generating a new data set is:

[0141] z=x+ò

[0142] Among them, z is the new data, ò is the disturbance value, and x is the original data;

[0143] S52. The Euclidean distance is used to measure the similarity between the generated new data set sample and the original data sample, and is used to assign a weight to each perturbed sample. The higher the weight, the more similar the perturbed sample is to the original sample.

[0144] The similarity calculation equation is:

[0145]

[0146] Among them, π x (z) is the weight of sample z, distance(x,z) is the distance between samples x and z, and σ is an adjustment parameter used to control the decay speed of similarity;

[0147] S53. A weighted linear regression model is used to fit the disturbance samples and the corresponding model outputs, and a local explanation model is obtained by minimizing the weighted loss function;

[0148] The equation for minimizing the weighted loss function is:

[0149]

[0150] Among them, ξ(g, f, π x ) is the weighted square loss function, g is the local explanation model, f is the complex prediction model constructed, π x (z) is the weight used to measure the similarity between the perturbation sample z and the explanation sample x, and Ω(g) is the complexity penalty term used to avoid overfitting;

[0151] S54. Based on the constructed local interpretation model, a single instance in the wind turbine SCADA data is analyzed and the abnormal state probability of the single instance is calculated, such as Fig.10 and 11 .

[0152] Fig.10 , showing the important features and their contributions in two different categories (0 and 1). The left side of the figure represents category 0 and the right side represents category 1. Each bar graph represents the contribution of a feature to the category. The larger the value, the longer the bar, indicating that the feature has a greater impact on the category. In category 0, the value of the "State and Fault" feature is 2, with the highest contribution (0.85), followed by the "Gearbox_T1_High_Speed_Shaft_Temperature" feature (contribution is 0.06); the classification probability is displayed below the bar graph of category 0, and the blue "0" indicates that the probability of category 0 is 1.00 and the probability of category 1 is 0; in category 1, the value of the "Gearbox_Oil_Temperature" feature is 46, with a lower contribution (0.03), and other features such as "Tower Acceleration The contribution of "Lateral" and "Gearbox_Oil-1_Temperature" is also relatively small; this shows that the influence of the features of category 1 is more dispersed. This figure is used to explain the model's prediction process for the two categories, highlighting the main features of each category and its contribution to the prediction results.

[0153] Fig.11, showing the correspondence between features and values. The left side lists the feature names and their values, and the right side indicates whether these feature values ​​are true (True). The feature names in the figure include "State and Fault", "Gearbox_T1_High_Speed_Shaft_Temperature", etc. Each feature is followed by a specific value, such as "State and Fault=2" and "Gearbox_Oil_Temperature=46", etc. The values ​​of all features are marked as "True", indicating that these features and their corresponding values ​​are valid in the current situation. The alternating blue and orange background in the figure helps to distinguish different features and enhances readability. This table format is used to display the specific feature values ​​of the model input or to explain the role of the feature in a specific prediction.

[0154] In order to further implement the above technical solution, a method for abnormal state detection and joint interpretation of a wind turbine generator set also includes: S6. Visualizing the abnormal state detection results and interpretation results, and triggering an alarm mechanism based on preset rules.

[0155] A wind turbine abnormal state detection and joint interpretation system, based on a wind turbine abnormal state detection and joint interpretation method, includes a data acquisition module, a preprocessing module, a feature selection module, an abnormal state detection module, a global interpretation module, a local interpretation module and a visualization module;

[0156] The acquisition module is used to obtain multi-dimensional operation data from the SCADA system of the wind turbine;

[0157] A preprocessing module, used for preprocessing the acquired multi-dimensional operation data;

[0158] The feature selection module is used to cluster the pre-processed wind turbine SCADA data using the k-means clustering algorithm, and select the central features of the calculated central cluster to construct a new feature data set;

[0159] The abnormal state detection module is used to construct a new feature data set based on the extracted central feature data, train the abnormal state detection model of the wind turbine using the classification decision boosting algorithm, and detect and identify the abnormal state;

[0160] The global explanation module is used to establish a simplified model of the output of the abnormal state detection model of the wind turbine, calculate the global contribution of the input features to the abnormal state detection results using the SHAP algorithm, and provide a global explanation;

[0161] The local explanation module is used to establish a local model of the abnormal state detection model instance of the wind turbine. The LIME algorithm is used to fit the local linear model to generate local explanations for specific abnormal instances and analyze the contribution of key features in specific abnormal instances.

[0162] The visualization module is used to visualize the detection and interpretation results and trigger the alarm mechanism.

[0163] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set.

[0164] A processing terminal comprises a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, a method for abnormal state detection and joint interpretation of a wind turbine generator set is implemented.

[0165] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0166] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set, characterized in that: The following steps are involved: S1. Obtain multi-dimensional operation data from the SCADA system of the wind turbine and perform pre-processing; S2. The pre-processed wind turbine SCADA data is clustered using the k-means clustering algorithm, and the central features of the calculated central cluster are selected to construct a new feature data set; S3. Based on the new feature data set constructed from the extracted central feature data, the classification decision boosting algorithm is used to train the abnormal state detection model of the wind turbine to detect and identify the abnormal state; S4. Establish a simplified model of the output of the abnormal state detection model of the wind turbine, use the SHAP algorithm to calculate the global contribution of the input features to the abnormal state detection results, and provide a global explanation; S5. Establish a local model of the wind turbine abnormal state detection model instance, use the LIME algorithm to fit the local linear model, generate local explanations for specific abnormal instances, and analyze the contribution of key features in specific abnormal instances.

2. A method for abnormal state detection and joint interpretation of a wind turbine generator set according to claim 1, characterized in that: The specific contents of the preprocessing in step S1 include: S11. Based on the SCADA data of wind turbines, the missing classification features in the data are filled using the majority filling method; S12. Based on the SCADA data of the wind turbine generator set, the mean filling method is used to fill the missing numerical features in the data; S13. The filled SCADA data of the wind turbine generator set, i.e., the pre-processed SCADA data of the wind turbine generator set, is used as a pre-training data set.

3. A method for abnormal state detection and joint interpretation of a wind turbine generator set according to claim 1, characterized in that: The specific content of step S2 is: S21. The optimal number of clusters is determined by using the elbow method. The total sum of squared deviations under different k values ​​is calculated, and the k value with which the total sum of squared deviations is significantly reduced is selected as the optimal number of clusters. S22. Clustering the wind turbine characteristics in the pre-trained data set of the pre-processed wind turbine SCADA data according to the determined optimal number of clusters; S23. According to the completed wind turbine feature clustering results, calculate the distance from the feature to the cluster center, and select the feature with the smallest distance as the central feature; S24. Use the central features selected by clustering to construct a new feature set to reduce collinearity and redundancy between features.

4. A method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set according to claim 1, characterized in that: The specific content of step S3 is: S31. Using the classification decision boosting algorithm to establish a wind turbine abnormal state detection model, starting from a constant model, the classification prediction value is initialized to the mean of the target value; S32. Calculate the gradient of each sample and the diagonal elements of the Hessian matrix in each iteration based on the initialized classification prediction value; S33. construct a histogram for each splitting point of each feature, calculate the gain of each splitting point using the histogram value, select the splitting point with the largest gain for splitting, and then construct a decision tree based on the selected optimal splitting point; S34. Use the currently constructed tree to update the prediction value of the model. After multiple iterations, the final model is the weighted sum of all base learners.

5. A method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set according to claim 1, characterized in that: The specific content of step S4 is: S41. Construct a simplified model of the abnormal state detection model of the wind turbine generator set, and approximate the output of the original model in the form of a linear combination; S42. Calculate the difference between the simplified model and the original model to meet local accuracy, missingness, and consistency; S43. Calculate the SHAP value of each feature in the simplified model, perform global visualization based on the SHAP value of each feature, and display the importance ranking of the features affecting the abnormal state of the wind turbine generator set and the correlation between the features in the figure.

6. A method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set according to claim 1, characterized in that: The specific content of step S5 is: S51 generates a new data set by locally perturbing the original data to simulate the behavior of the model in the local space; S52. The Euclidean distance is used to measure the similarity between the generated new data set sample and the original data sample, and is used to assign a weight to each perturbed sample. The higher the weight, the more similar the perturbed sample is to the original sample. S53. A weighted linear regression model is used to fit the disturbance samples and the corresponding model outputs, and a local explanation model is obtained by minimizing the weighted loss function; S54. Based on the constructed local explanation model, a single instance in the wind turbine SCADA data is analyzed to calculate the abnormal state probability of the single instance.

7. A method for detecting and jointly interpreting abnormal conditions of a wind turbine generator set according to claim 1, characterized in that: Also includes: S6. Visualize the abnormal state detection results and interpretation results, and trigger the alarm mechanism based on preset rules.

8. A wind turbine abnormal state detection and joint interpretation system, characterized in that: A method for abnormal state detection and joint interpretation of a wind turbine generator set according to any one of claims 1 to 7, comprising a data acquisition module, a preprocessing module, a feature selection module, an abnormal state detection module, a global interpretation module, a local interpretation module and a visualization module; The acquisition module is used to obtain multi-dimensional operation data from the SCADA system of the wind turbine; A preprocessing module, used for preprocessing the acquired multi-dimensional operation data; The feature selection module is used to cluster the pre-processed wind turbine SCADA data using the k-means clustering algorithm, and select the central features of the calculated central cluster to construct a new feature data set; The abnormal state detection module is used to construct a new feature data set based on the extracted central feature data, train the abnormal state detection model of the wind turbine using the classification decision boosting algorithm, and detect and identify the abnormal state; The global explanation module is used to establish a simplified model of the output of the abnormal state detection model of the wind turbine, calculate the global contribution of the input features to the abnormal state detection results using the SHAP algorithm, and provide a global explanation; The local explanation module is used to establish a local model of the abnormal state detection model instance of the wind turbine. The LIME algorithm is used to fit the local linear model to generate local explanations for specific abnormal instances and analyze the contribution of key features in specific abnormal instances. The visualization module is used to visualize the detection and interpretation results and trigger the alarm mechanism.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the abnormal state detection and joint interpretation method of a wind turbine generator set described in any one of claims 1 to 7 is implemented.

10. A processing terminal, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the computer program, the abnormal state detection and joint interpretation method of a wind turbine generator set as described in any one of claims 1 to 7 is implemented.

Images