Fault diagnosis method and system for fan blade icing and blade quality imbalance

Abstract

The invention discloses a fan blade icing and blade quality imbalance fault diagnosis method and system. The method comprises the following steps: acquiring operation data and blade icing information of a plurality of fans; according to the blade icing information, icing state labels are marked on the operation data, and fault data are obtained; fault features in the fault data are extracted, the fault features are sorted according to the importance degree, and an optimal feature set is generated; selecting an optimal feature in the optimal feature set according to a square error minimization criterion to generate an optimal decision tree; and performing classification based on the optimal decision tree to obtain a diagnosis detection result containing fault information. The method has the advantages of high diagnosis precision and the like.

Description

Technical Field

[0001] The present invention mainly relates to the field of wind power technology, and in particular to a method and system for diagnosing wind turbine blade icing and blade mass imbalance faults. Background Art

[0002] The rapid development of wind power has inevitably brought some problems, such as wind turbine icing problems, wind turbine blade mass imbalance problems, etc. When the wind turbine is running in a faulty state, the quality of the power it generates is low and it cannot be connected to the grid normally, which will cause serious waste of resources. Moreover, when the wind turbine runs in a faulty state for a period of time, the entire system will begin to have different degrees of failures, causing the wind turbine to be unable to operate under normal conditions.

[0003] Onshore wind turbines are generally installed in places with abundant wind resources such as cold plateaus and mountain ridges. These places are usually at high altitudes and low temperatures, which can easily cause blade icing, which in turn leads to reduced wind turbine efficiency, changes in blade shape, and even induces blade breakage. In addition, most wind turbines are mainly operated outdoors in relatively harsh environments, and wind and sand wear or icing may cause impeller mass imbalance. Continuing to operate with an unbalanced impeller mass will aggravate the fatigue vibration of components such as blades, and cause serious damage to the wind turbine. If the state changes of the wind turbine are monitored and discovered in a timely manner, and the fault state and fault location are identified, the wind turbine can be maintained and repaired in a timely manner, thereby greatly reducing the loss and waste of wind power generation resources. Therefore, research on fault diagnosis of wind turbine icing problems and wind turbine blade mass imbalance problems is of great significance.

[0004] Solving the problem of wind turbine blade imbalance and icing through modeling is actually tracking and analyzing the internal physical changes during the blade failure process, so as to determine the fault state through the corresponding data. This method is based on the study of the mechanism and usually requires additional sensors to be placed on the wind turbine blades. The disadvantages of inconvenience in practical application and increased operation and maintenance costs of wind farms make it impossible to be widely used in practice.

[0005] The data-driven modeling method does not require the installation of additional sensors. It collects and monitors data through the supervisory control and data acquisition system (SCADA system), from which the feature quantity is mined to achieve the diagnosis of the fault state. This method often requires a large amount of data as support, and the disadvantage is that it lacks interpretability and operational guidance. Summary of the invention

[0006] In view of the technical problems existing in the prior art, the present invention provides a method and system for diagnosing fan blade icing and blade mass imbalance faults with high diagnostic accuracy.

[0007] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0008] A method for diagnosing fan blade icing and blade mass imbalance faults, comprising the steps of:

[0009] Obtain operating data and blade icing information of multiple wind turbines;

[0010] According to the blade icing information, the operating data is labeled with an icing status label to obtain fault data;

[0011] Extract fault features from fault data, sort the fault features according to their importance, and generate the optimal feature set;

[0012] The optimal feature in the optimal feature set is selected according to the criterion of minimizing the square error to generate the optimal decision tree;

[0013] Classification is performed based on the optimal decision tree to obtain diagnostic test results containing fault information.

[0014] Preferably, the operation data of the wind turbine includes wind speed, power, cabin temperature, ambient temperature and front, rear, left and right acceleration of the cabin.

[0015] Preferably, the fault characteristics include power residual characteristics, wind speed power ratio, rotation speed residual characteristics, temperature difference characteristics and vibration acceleration vector synthesis characteristics.

[0016] Preferably, the fault features are sorted by importance by logistic regression and support vector machine to generate an optimal feature set, which specifically includes:

[0017] The fault data is divided into training set D by resampling 1j and validation set D 2j ;

[0018] Using the training set D 1j All features 1j Training the random forest fault diagnosis model Mi j (θ), in the validation set D 2j Mi j (θ) Evaluation;

[0019] By training the model M ij (θ), calculate its performance score, and 1j Sort by importance from high to low, select features from important to unimportant, and form a feature set S ij , where i = 1, 2, ..., m;

[0020] For the feature set, use the training set D 1j The feature set S ij Training model Mi j(θ), in the validation set D 2j Perform model evaluation and obtain model Mi j (θ) in the feature set S ij The score on

[0021] Set S the number of features with the same number of resampled h times ij Average the model scores and get the number of features with the highest score;

[0022] Use the number of features to execute the algorithm on the original training set D to obtain the optimal feature set S max .

[0023] Preferably, the random forest classification fault diagnosis model integrates multiple decision tree basic units through the idea of ​​ensemble learning, including:

[0024] The optimal features are selected to generate a decision tree by minimizing the square error; the decision tree is defined as:

[0025]

[0026] In the formula, x is the input, f(x) is the output, and M means that the input space is divided into M units: R1, R2, ..., R m , and in each unit R m There is a fixed output value mc on it;

[0027] The left and right branch areas of the jth feature of the decision tree model and the cut point s of this feature are defined as:

[0028] R1(j,s)={xx (j) ≤s},R2(j,s)={xx (j) >s}

[0029] The optimal segmentation feature j and segmentation point s are obtained by solving the following formula:

[0030]

[0031] In the formula, y i For x i The corresponding label values, c1 and c2 are:

[0032] C1=ave(y i x i ∈R1(j,s)),C2=ave(y i x i ∈R2(j,s))

[0033] Starting from the above root node, find the optimal segmentation feature and segmentation point to divide the input space into two regions, and then repeat this division process for each region, and finally obtain m division units R1, R2, ...R m , according to the criterion of minimum square error, the optimal output value of each unit is obtained; each unit R m The optimal value of is all samples x of this unit i The corresponding y i The mean of is defined as:

[0034] C m =ave(y i x i ∈R m ).

[0035] Preferably, the rules for generating a decision tree include:

[0036] Select k sub-training sample sets from the total training sample set and build k decision trees;

[0037] At each node of the decision tree, m indicators are randomly selected from n indicators. According to the principle of minimum node impurity, the optimal features are selected from m candidate indicators to classify and grow the nodes. The decision tree is allowed to grow fully until the impurity of each leaf node is minimized. At the same time, the decision tree is not pruned.

[0038] Repeat the above steps to traverse the pre-built k decision trees, and form a random forest from the k decision trees;

[0039] The classification results of new unknown samples to be tested are predicted based on the k decision trees that have been grown, and the decision is made based on the majority vote of the k decision trees.

[0040] Preferably, the decision tree is pruned, starting from the bottom of the complete decision tree T0 and continuously pruning to the root node to form a subtree sequence {T0, T1, ..., T n}, then use the cross-validation method to test the subtree on the validation set and select the optimal subtree;

[0041] The loss function of the subtree in the above pruning operation is defined as:

[0042] C α (T) = C(T) + αT |

[0043] In the formula, T is an arbitrary subtree, C(T) is the prediction error of the training data, |T| is the number of leaf nodes of the subtree, which is a parameter greater than 0 and is used to weigh the degree of fit of the training data and the complexity of the model. α (T) represents the overall loss of subtree T when the parameter is α;

[0044] Starting from T0, calculate the α of each node from bottom to top, cut off the subtree with the smallest α in T0 to get T1, and then use T1 as the whole tree, calculate the α of each node from bottom to top, cut off the subtree with the smallest α to get T2, until T n Finally, the validation set is used to train the subtrees {T0, T1, ..., T n}Test, the decision tree with the smallest square error is the optimal decision tree.

[0045] The present invention also discloses a fan blade icing and blade mass imbalance fault diagnosis system, comprising:

[0046] A data acquisition module, used to obtain the operating data of multiple wind turbines and blade icing information;

[0047] A data preprocessing module is used to label the operating data with an icing status label according to the blade icing information to obtain fault data;

[0048] The feature selection module is used to extract fault features from fault data, sort the fault features according to their importance, and generate the optimal feature set;

[0049] A decision tree generation module, used for selecting the best feature in the best feature set to generate an optimal decision tree according to the criterion of minimizing the square error;

[0050] The random forest module is used to perform classification based on the optimal decision tree to obtain diagnostic detection results containing fault information.

[0051] The present invention further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described above are executed.

[0052] The present invention also discloses a computer device, comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, and when the computer program is run by the processor, the steps of the method described above are executed.

[0053] Compared with the prior art, the advantages of the present invention are:

[0054] The present invention adopts a wrapped feature selection method in feature selection, and combines it with cross-validation automatic optimization. Compared with a single model, it has a more powerful feature judgment ability. In addition, the present invention adopts a random forest classification method to extract some samples and features for all decision trees, effectively avoiding overfitting, with stable performance and suitable for parallel computing. Compared with other diagnostic methods, it simplifies the complexity of model training and has higher diagnostic performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 The flowchart of the fault diagnosis method in an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of data label classification in data preprocessing in the present invention.

[0057] Figure 3 It is a schematic diagram of the statistical results of data labels in the present invention.

[0058] Figure 4 This is a graph showing how the classification accuracy of the feature selection method proposed in the present invention changes with the number of selected feature variables.

[0059] Figure 5 This is a structural diagram of the random forest algorithm in the present invention.

[0060] Figure 6 This is the confusion matrix result diagram obtained by using the algorithm model of the present invention to perform fault diagnosis on the blade icing test set.

[0061] Figure 7 This is a confusion matrix result diagram obtained by performing fault diagnosis on the blade mass imbalance test set using the algorithm model of the present invention. DETAILED DESCRIPTION

[0062] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0063] like Figure 1 As shown, an embodiment of the present invention provides a method for diagnosing wind turbine blade icing and blade mass imbalance faults based on a random forest model, which specifically includes the following steps:

[0064] Data collection: Collect real-time operating data of multiple wind turbines and blade icing information through the SCADA system;

[0065] Icing state labeling: label the icing state of each wind turbine blade according to the icing information in each operating period; preprocess the operating data (including wind turbine operating parameters: unit operating state, wind speed, power, cabin temperature, ambient temperature, front and rear and left and right acceleration information of the cabin, etc., see Table 1) to obtain the preprocessed data, and divide it into a test set and a training set;

[0066] Table 1 SCADA system status data

[0067]

[0068] Perform feature selection on the preprocessed data, obtain power residual features, wind speed power ratio, speed residual features, temperature difference features, vibration acceleration vector synthesis features, etc. according to the wind turbine operating sequence data, and reduce the feature dimensions to obtain the optimal feature subset;

[0069] A random forest fault diagnosis model was established using the optimal feature subset. The features in the optimal feature subset were used as input, and the current icing state label, future icing state label and icing state confidence were used as the output of the model to verify the fault diagnosis accuracy.

[0070] like Figure 2-Figure 3 As shown, data preprocessing mainly includes marking the fault data row as 1 ( Figure 3 89%), normal data rows are marked as 0 ( Figure 3 6%), invalid data ( Figure 3 5% of the data are not included in the evaluation. Variables that are useless for modeling are deleted to reduce the impact of invalid information on the model, solve the problem of missing and abnormal data, and make the data better fit the model. The data set is divided into training set and test set in a ratio of 2:1.

[0071] like Figure 4 As shown, the above feature selection method adopts RFECV method, and outputs the feature importance ranking through logistic regression and support vector machine, including:

[0072] The data set is divided into training set D by resampling 1j and validation set D 2j ;

[0073] Using the training set D 1j All features 1j Training model Mi j (θ), in the validation set D 2j Mi j (θ) Evaluation;

[0074] By training the model M ij (θ) calculates its performance score and j 1j Sort by importance from high to low, select features from important to unimportant, and form a feature set S ij , where i = 1, 2, ..., m;

[0075] For the feature set, use the training set D 1j The feature set S ij Training model Mi j (θ), in the validation set D 2j Perform model evaluation and obtain model Mi j (θ) in the feature set S ij The score on

[0076] Set S the number of features with the same number of resampled h times ij Average the model scores and get the number of features with the highest score;

[0077] Use the number of features to execute the algorithm on the original training set D to obtain the optimal feature set S max .

[0078] like Figure 5 As shown in the figure, the random forest classification fault diagnosis model integrates multiple decision tree basic units through the idea of ​​ensemble learning, including:

[0079] The above decision tree is generated by selecting the optimal feature through the criterion of minimizing the square error; the decision tree is defined as:

[0080]

[0081] In the formula, x is the input, f(x) is the output, and M means that the input space is divided into M units: R1, R2, ..., R m , and in each unit R m There is a fixed output value mc on it;

[0082] The jth feature of the above decision tree model and the left and right branch areas of this feature split point s are defined as:

[0083] R1(j,s)={xx (j) ≤s},R2(j,s)={xx (j) >s}

[0084] The optimal segmentation feature j and segmentation point s are obtained by solving the following formula:

[0085]

[0086] In the formula, y i For x i The corresponding label values, c1 and c2 are:

[0087] C1=ave(y i x i ∈R1(j,s)),C2=ave(y i x i ∈R2(j,s))

[0088] Starting from the above root node, find the optimal segmentation feature and segmentation point to divide the input space into two regions, and then repeat this division process for each region, and finally obtain m division units R1, R2, ...R m , according to the criterion of minimum square error, the optimal output value of each unit is obtained; each unit R m The optimal value of is all samples x of this unit i The corresponding y i The mean of is defined as:

[0089] C m =ave(yi x i ∈R m ).

[0090] The above rules for generating decision trees include:

[0091] Select k sub-training sample sets from the total training sample set and build k decision trees;

[0092] At each node of the decision tree, m indicators are randomly selected from n indicators. According to the principle of minimum node impurity, the optimal features are selected from m candidate indicators to classify and grow the nodes. The decision tree is allowed to grow fully until the impurity of each leaf node is minimized. At the same time, the decision tree is not pruned.

[0093] Repeat the above steps to traverse the pre-built k decision trees, and form a random forest from the k decision trees;

[0094] The classification result of the new unknown sample to be tested is predicted based on the k decision trees that have been grown, and is determined by the majority vote of the k decision trees.

[0095] The above decision tree pruning starts from the bottom of the complete decision tree T0 and prunes to the root node to form a subtree sequence {T0, T1, ..., T n}, then use the cross-validation method to test the subtree on the validation set and select the optimal subtree;

[0096] The loss function of the subtree in the above pruning operation is defined as:

[0097] C α (T) = C (T) + αT

[0098] In the formula, T is an arbitrary subtree, C(T) is the prediction error of the training data, |T| is the number of leaf nodes of the subtree, which is a parameter greater than 0 and is used to weigh the degree of fit of the training data and the complexity of the model. α (T) represents the overall loss of subtree T when the parameter is α;

[0099] Starting from T0, calculate the α of each node from bottom to top, cut off the subtree with the smallest α in T0 to get T1, and then use T1 as the whole tree, calculate the α of each node from bottom to top, cut off the subtree with the smallest α to get T2, until T n Finally, the validation set is used to train the subtrees {T0, T1, ..., T n}Test, the decision tree with the smallest square error is the optimal decision tree.

[0100] Specifically, the parameter settings of random forest include:

[0101] max_features: The maximum number of eigenvalues ​​to consider when partitioning the data set. It needs to be appropriately reduced for cases with a large number of eigenvalues.

[0102] max_depth: The maximum depth of the decision tree. When there is little data or features, this value can be left unset. The default value can complete the model training well. If the model has a large number of samples and features, you can try to limit the value. The commonly used value range is 10-100, which can improve the training accuracy and efficiency while avoiding overfitting of data.

[0103] min_samples_split: The minimum number of samples required to split the internal node (i.e., judgment condition). If the number of node samples is less than min_samples_split, the judgment tree will not be tried to split further. When the sample size is small, the value should be smaller to improve the training accuracy. If the sample size is very large, this value should be increased to speed up the training of the judgment tree and avoid overfitting of the data.

[0104] min_samples_leaf: The minimum number of samples for leaf nodes (i.e. classification). It limits the minimum number of samples for leaf nodes. If the number of leaf nodes is less than the number of samples, they will be pruned together with their sibling nodes. When the sample size is small, the value should be small. If the sample size is very large, this value should be increased to speed up the training of the judgment tree and improve efficiency.

[0105] min_weight_fraction_leaf: The minimum sample weight of a leaf node (i.e., classification). This value limits the minimum value of the sum of all sample weights of a leaf node. If it is less than this value, it will be pruned together with its sibling nodes. The default value is 0, which means that the weight issue is not considered and all samples have the same weight. Weights need to be considered for special data.

[0106] max_leaf_nodes: maximum number of leaf nodes, default value 0, limit the maximum number of leaf nodes, the purpose is to prevent overfitting; if the limit is added, the algorithm will build the optimal decision tree within the maximum number of leaf nodes. If there are not many features, you can ignore this value, but if there are many features, you can add a limit to achieve the best training model results.

[0107] min_impurity_decrease: minimum impurity for node partitioning, the default value is 0. To limit the growth of the decision tree, the impurity of the node (Gini coefficient, information gain, mean square error, absolute difference) must be greater than this threshold, otherwise the node will no longer generate child nodes.

[0108] presort: pre-sort data, the default value is False, indicating whether to pre-sort data before fitting to speed up tree building. For classification with very large data sets, presort = true will cause the entire classification to become slow; when the data set is small and the tree depth is limited, presort = true will speed up classification. Whether to enable it needs to be set according to the actual situation.

[0109] The present invention adopts a wrapped feature selection method in feature selection, and combines it with cross-validation automatic optimization. Compared with a single model, it has a more powerful feature judgment ability. In addition, the present invention adopts a random forest classification method to extract some samples and features for all decision trees, effectively avoiding overfitting, with stable performance and suitable for parallel computing. Compared with other diagnostic methods, it simplifies the complexity of model training and has higher diagnostic performance.

[0110] After the above model is established, the model performance needs to be evaluated and then optimized. For binary classification problems, confusion matrix related indicators are often used for evaluation:

[0111]

[0112] Among them, TP is the number of samples that are actually positive and predicted as positive, FN is the number of samples that are actually positive and predicted as negative, FP is the number of samples that are actually negative and predicted as positive, and TN is the number of samples that are actually negative and predicted as negative.

[0113] Accuracy, precision, recall and F1-score:

[0114] Accuracy refers to the ratio of the total number of samples that the classifier can predict as positive examples and negative examples as negative examples when using the trained model to predict the test data, and is defined as:

[0115]

[0116] Precision refers to the proportion of samples predicted to be positive that are actually positive, and is defined as:

[0117]

[0118] The recall rate refers to the ratio of the number of correctly predicted positive samples to the total number of positive samples, which is defined as:

[0119]

[0120] The concept of F1-score is to harmonize the precision and recall and is defined as:

[0121]

[0122] Figure 6 The confusion matrix results obtained by random forest classification for fault diagnosis on the blade icing test set are as follows: Figure 6 Display data calculation, the random forest classification model shows that the corresponding accuracy is 99.98%, the precision is 100%, the recall is 99.49%, and the F1-score is 1; Figure 7 The confusion matrix results obtained by random forest classification for fault diagnosis on the blade mass imbalance test set are as follows: Figure 7 The data calculation shows that the random forest classification model has a corresponding accuracy of 99.71%, a recall rate of 100%, a recall rate of 99.67%, and an F1-score of 1. It can be seen that the random forest classification diagnosis model has a high diagnosis accuracy for blade icing and blade imbalance, achieving the goal of ensuring the model diagnosis performance on the basis of simplifying the complexity of model training.

[0123] The embodiment of the present invention further provides a fan blade icing and blade mass imbalance fault diagnosis system based on a random forest model, comprising:

[0124] A preprocessing module is used to preprocess the fault data, obtain the preprocessed data, and divide it into a test set and a training set;

[0125] A data input module is used to input the data preprocessed by the preprocessing module into the pre-trained random forest model;

[0126] The feature selection module is used to output the feature importance ranking based on the pre-processed data input by the data input module, using the RFECV method, logistic regression and support vector machine to generate the optimal feature subset;

[0127] The decision tree generation and pruning module is used to select the optimal features to generate a decision tree based on the criterion of minimizing the square error, and continuously prune from the bottom to the root node to form a subtree sequence, and cross-validate to obtain the optimal subtree;

[0128] The random forest module is used to set the random forest parameters, perform classification based on the optimal decision tree, and obtain diagnostic detection results containing fault information.

[0129] The present invention further discloses a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed. The present invention also discloses a computer device, including a memory and a processor connected to each other, on which a computer program is stored, and when the computer program is executed by the processor, the steps of the above method are executed. The system, medium and device of the present invention correspond to the above method and also have the advantages described in the above method.

[0130] The present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiment when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable storage media include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. The memory is used to store computer programs and / or modules, and the processor implements various functions by running or executing computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory may include high-speed random access memory and may also include non-volatile memory, such as a hard disk, an internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0131] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for diagnosing fan blade icing and blade mass imbalance faults, characterized in that: Includes steps: Obtain operating data and blade icing information of multiple wind turbines; According to the blade icing information, the operating data is labeled with an icing status label to obtain fault data; Extract fault features from fault data, sort the fault features according to their importance, and generate the optimal feature set; The optimal feature in the optimal feature set is selected according to the criterion of minimizing the square error to generate the optimal decision tree; Classification is performed based on the optimal decision tree to obtain diagnostic test results containing fault information.

2. The method for diagnosing wind turbine blade icing and blade mass imbalance faults according to claim 1, characterized in that: The operating data of the wind turbine include wind speed, power, cabin temperature, ambient temperature and acceleration of the cabin front, back, left and right.

3. The method for diagnosing wind turbine blade icing and blade mass imbalance faults according to claim 2, characterized in that: The fault characteristics include power residual characteristics, wind speed power ratio, rotation speed residual characteristics, temperature difference characteristics and vibration acceleration vector synthesis characteristics.

4. The method for diagnosing wind turbine blade icing and blade mass imbalance faults according to claim 1, 2 or 3, characterized in that: Logistic regression and support vector machine are used to sort fault features according to their importance to generate the optimal feature set, including: The fault data is divided into training set D by resampling 1j and validation set D 2j ; Using the training set D 1j All features 1j Training the random forest fault diagnosis model Mi j (θ), in the validation set D 2j Mi j (θ) Evaluation; By training the model M ij (θ), calculate its performance score, and j 1j Sort by importance from high to low, select features from important to unimportant, and form a feature set S ij , where i = 1, 2, ..., m; For the feature set, use the training set D 1j The feature set S ij Training model Mi j (θ), in the validation set D 2j Perform model evaluation and obtain model Mi j (θ) in the feature set S ij The score on Set S the number of features with the same number of resampled h times ij Average the model scores and get the number of features with the highest score; Use the number of features to execute the algorithm on the original training set D to obtain the optimal feature set S max .

5. The method for diagnosing wind turbine blade icing and blade mass imbalance faults according to claim 4, characterized in that: The random forest classification fault diagnosis model integrates multiple decision tree basic units through the idea of ​​ensemble learning, including: The optimal features are selected to generate a decision tree by minimizing the square error; the decision tree is defined as: In the formula, x is the input, f(x) is the output, and M means that the input space is divided into M units: R1, R2, ..., R m , and in each unit R m There is a fixed output value mc on it; The left and right branch areas of the jth feature of the decision tree model and the cut point s of this feature are defined as: R1(j,s)={xx (j) ≤s},R2(j,s)={xx (j) >s} The optimal segmentation feature j and segmentation point s are obtained by solving the following formula: In the formula, y i For x i The corresponding label values, c1 and c2 are: C1=ave(y i x i ∈R1(j,s)),C2=ave(y i x i ∈R2(j,s)) Starting from the above root node, find the optimal segmentation feature and segmentation point to divide the input space into two regions, and then repeat this division process for each region, and finally obtain m division units R1, R2, ...R m , according to the criterion of minimum square error, the optimal output value of each unit is obtained; each unit R m The optimal value of is all samples x of this unit i The corresponding y i The mean of is defined as: C m =ave(and i x i ∈R m )。 6. The method for diagnosing wind turbine blade icing and blade mass imbalance faults according to claim 5, characterized in that: The rules for generating a decision tree include: Select k sub-training sample sets from the total training sample set and build k decision trees; At each node of the decision tree, m indicators are randomly selected from n indicators. According to the principle of minimum node impurity, the optimal features are selected from m candidate indicators to classify and grow the nodes. The decision tree is allowed to grow fully until the impurity of each leaf node is minimized. At the same time, the decision tree is not pruned. Repeat the above steps to traverse the pre-built k decision trees, and form a random forest from the k decision trees; The classification results of new unknown samples to be tested are predicted based on the k decision trees that have been grown, and the decision is made based on the majority vote of the k decision trees.

7. The method for diagnosing wind turbine blade icing and blade mass imbalance faults according to claim 6, characterized in that: To prune the decision tree, we first start from the bottom of the complete decision tree T0 and continue to prune to the root node to form a subtree sequence {T0, T1, ..., T n }, then use the cross-validation method to test the subtree on the validation set and select the optimal subtree; The loss function of the subtree in the above pruning operation is defined as: C α (T)=C(T)+αT| In the formula, T is an arbitrary subtree, C(T) is the prediction error of the training data, |T| is the number of leaf nodes of the subtree, which is a parameter greater than 0 and is used to weigh the degree of fit of the training data and the complexity of the model. α (T) represents the overall loss of subtree T when the parameter is α; Starting from T0, calculate the α of each node from bottom to top, cut off the subtree with the smallest α in T0 to get T1, and then use T1 as the whole tree, calculate the α of each node from bottom to top, cut off the subtree with the smallest α to get T2, until T n Finally, the validation set is used to train the subtrees {T0, T1, ..., T n }Test, the decision tree with the smallest square error is the optimal decision tree.

8. A fan blade icing and blade mass imbalance fault diagnosis system, characterized in that: include: A data acquisition module, used to obtain the operating data of multiple wind turbines and blade icing information; A data preprocessing module is used to label the operating data with an icing status label according to the blade icing information to obtain fault data; The feature selection module is used to extract fault features from fault data, sort the fault features according to their importance, and generate the optimal feature set; A decision tree generation module, used for selecting the best feature in the best feature set to generate an optimal decision tree according to the criterion of minimizing the square error; The random forest module is used to perform classification based on the optimal decision tree to obtain diagnostic detection results containing fault information.

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 computer program performs the steps of the method according to any one of claims 1 to 7.

10. A computer device comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 7.

Images