Fault text-based classification method, device, equipment and storage medium

Abstract

The application discloses a kind of based on the classification method of fault text, device and storage medium, belong to wind power generation field.The method includes: the fault text obtained from wind farm is as fault text training set, and fault text training set is stored in wind power corpus database;From the fault text of fault text training set, the first fault characteristic word associated with wind generating set is extracted;According to the multiple candidate classification numbers of pre-set, first fault characteristic word is classified respectively, and the classification result corresponding to each candidate classification number is obtained;According to classification result, the candidate classification number of maximum difference degree of different categories first fault characteristic word is determined as target classification number;Based on target classification number, the second fault characteristic word associated with wind generating set in the fault text to be classified is classified, and the second fault characteristic word after classification is obtained.According to the embodiment of the application, the accuracy of the classification of the content in the fault text can be improved.

Description

Technical Field

[0001] This application belongs to the field of wind power generation, and in particular relates to a classification method, apparatus, equipment and storage medium based on fault text. Background Technology

[0002] With the rapid development of wind power technology, the number of wind turbines is constantly increasing, and correspondingly, fault data of wind turbine generator sets is also accumulating. This fault data can provide strong support for fault diagnosis and analysis of wind turbine generator sets.

[0003] Fault data can be presented as fault text. To facilitate fault diagnosis and analysis, the content of the fault text needs to be pre-classified. Currently, the content of the fault text can be manually classified by staff responsible for information classification. However, the content of fault text is quite complex, and the classified content may contain mixed content from different categories, resulting in low accuracy in the classification of the fault text content. Summary of the Invention

[0004] This application provides a classification method, apparatus, device, and storage medium based on fault text, which can improve the accuracy of classifying the content in fault text.

[0005] In a first aspect, embodiments of this application provide a classification method based on fault text, comprising: using fault text obtained from a wind farm as a fault text training set, the fault text training set being stored in a wind power corpus database; extracting first fault feature words associated with wind turbine generators from the fault text in the fault text training set; classifying the first fault feature words according to a preset number of candidate classifications, obtaining classification results corresponding to each number of candidate classifications; determining the number of candidate classifications with the greatest difference between the first fault feature words of different categories as the target number of classifications; and classifying second fault feature words associated with wind turbine generators in the fault text to be classified based on the target number of classifications, obtaining the classified second fault feature words.

[0006] In some possible embodiments, after obtaining the classified second fault feature words, the method further includes: controlling the classified second fault feature words to be displayed on a display device for fault analysis of the wind turbine generator set.

[0007] In some possible embodiments, the classification result includes the probability of the first fault feature word being assigned to each category; based on the classification result, the number of candidate categories with the greatest difference in the first fault feature words of different categories is determined as the target number of categories, including: under any number of candidate categories, calculating the absolute value of the probability difference, the absolute value of the probability difference includes the absolute value of the difference in the probability of the first fault feature word being assigned to different categories; under a number of candidate categories, the absolute value of the probability difference is determined as the difference in the first fault feature words of different categories; the number of candidate categories corresponding to the largest absolute value of the probability difference is determined as the target number of categories.

[0008] In some possible embodiments, the classification result includes the probability of the first fault feature word being assigned to each category; classifying the first fault feature word according to a preset number of candidate categories to obtain the classification result corresponding to each number of candidate categories, including: for each of the multiple candidate categories, using a linear discriminant analysis algorithm to classify the first fault feature word into categories with a number equal to the number of candidate categories, and obtaining the probability of the first fault feature word being assigned to each category.

[0009] In some possible embodiments, classifying the second fault feature words associated with the wind turbine generator in the fault text to be classified based on the target number of classifications includes: setting the target number of classifications as the number of classifications of the classifier, training the classifier using fault training data labeled with categories to obtain the target classifier; and using the target classifier to classify the second fault feature words associated with the wind turbine generator in the fault text to be classified.

[0010] In some possible embodiments, the number of candidate classifications includes at least two of the following: 2, 3, 4, and 5.

[0011] In some possible embodiments, before classifying the second fault feature words associated with wind turbine generators in the fault text to be classified based on the target number of classifications, the method further includes: obtaining a fault text test set from a wind power corpus database, the fault text test set including fault texts; extracting third fault feature words associated with wind turbine generators from the fault texts in the obtained fault text test set; classifying the third fault feature words based on the target number of classifications; and obtaining the classification accuracy of the classified third fault feature words.

[0012] Based on the target number of classifications, the second fault feature words in the fault text to be classified are classified, including: when the classification accuracy of the third fault feature words is higher than or equal to the preset test standard accuracy, the second fault feature words in the fault text to be classified are classified based on the target number of classifications.

[0013] In some possible embodiments, extracting a first fault feature word associated with a wind turbine generator from fault texts in a fault text training set includes: segmenting the fault texts in the acquired fault text training set to obtain feature words of the fault texts; removing invalid characters from the feature words; and extracting the first fault feature word from the feature words from which invalid characters have been removed.

[0014] In some possible embodiments, the method further includes controlling the wind turbine generator based on the results of a fault analysis of the wind turbine generator.

[0015] Secondly, embodiments of this application provide a classification device based on fault text, comprising: an acquisition module, used to obtain fault text from a wind farm as a fault text training set, the fault text training set being stored in a wind power corpus database; an extraction module, used to extract first fault feature words associated with wind turbine generators from the fault text in the fault text training set; a traversal processing module, used to classify the first fault feature words according to a preset number of candidate classifications, obtaining classification results corresponding to each number of candidate classifications; a classification number determination module, used to determine the candidate classification number with the greatest difference between different categories of first fault feature words as the target classification number based on the classification results; and a classification module, used to classify second fault feature words associated with wind turbine generators in the fault text to be classified based on the target classification number, obtaining classified second fault feature words.

[0016] Thirdly, embodiments of this application provide a fault text-based classification device, including: a processor and a memory storing computer program instructions; the processor executes the computer program instructions to implement the fault text-based classification method of the first aspect.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the fault text-based classification method of the first aspect.

[0018] This application provides a method, apparatus, device, and storage medium for classifying fault text. It utilizes fault text obtained from wind farms to form a fault text training set. According to a preset number of candidate classification numbers, it classifies first fault feature words extracted from the fault text training set, obtaining classification results corresponding to each candidate classification number. Based on the classification results, it selects the candidate classification number that maximizes the difference between the first fault feature words of different categories after classification as the target classification number. The target classification number is then used to classify fault feature words associated with wind turbine generators in the fault text to be classified. The greater the difference between the first fault feature words of different categories, the higher the accuracy of classification using that candidate classification number. By selecting the candidate classification number with the greatest difference between the first fault feature words of different categories to classify fault feature words associated with wind turbine generators in the fault text to be classified, the accuracy of classifying the content in the fault text is improved. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of an embodiment of the fault text-based classification method provided in this application;

[0021] Figure 2 A flowchart of another embodiment of the fault text-based classification method provided in this application;

[0022] Figure 3 A flowchart of yet another embodiment of the fault text-based classification method provided in this application;

[0023] Figure 4 A schematic diagram of the structure of an embodiment of the fault text-based classification device provided in this application;

[0024] Figure 5 A schematic diagram of another embodiment of the fault text-based classification device provided in this application;

[0025] Figure 6 This is a schematic diagram of an embodiment of the fault text-based classification device provided in this application. Detailed Implementation

[0026] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are intended only to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0027] With the rapid development of wind power technology, the number of wind turbines is constantly increasing, and correspondingly, the fault data of wind turbine generators is also accumulating. This fault data provides strong support for fault diagnosis and analysis. Fault data can be presented in the form of fault text. To facilitate fault diagnosis and analysis, the content of the fault text needs to be pre-classified. Currently, the content of the fault text can be manually classified by dedicated personnel. However, the content of fault text in the wind power field has its own inherent complexity, and the classified content may contain a mixture of different categories, leading to lower accuracy in the classification of the fault text content. For example, the fault text may include a fault information table, with items such as fault name, fault time, wind farm name, fault description, and handling measures. Since the fault information table is filled out by staff, there may be some overlap between the fault description and handling measures. For example, content that should be included in the handling measures may be included in the fault description, or the same part may exist in both the handling measures and the fault description. Staff specifically responsible for information classification will categorize information according to the items in the fault information form, which will result in mixed information after classification, reduced accuracy, and adverse effects on subsequent fault analysis and other processes using the classified information.

[0028] This application provides a classification method, apparatus, device, and storage medium based on fault text. It can extract fault feature words associated with wind turbine generators from a fault text training set formed by fault texts obtained from wind farms, classify the first fault feature words according to different candidate classification data, and use the candidate classification data that maximizes the difference between the first fault feature words of different categories as the final number of classifications used to classify the fault text, thereby improving the accuracy of classification of the content in the fault text.

[0029] The first aspect of this application provides a classification method based on fault text, which can be applied to a fault text-based classification apparatus or device, i.e., it can be executed by the fault text-based classification apparatus or device. The fault text-based classification apparatus or device may include a controller or processor, and may also include other structures, without limiting the specific structure of the fault text-based classification apparatus or device herein. Figure 1 A flowchart illustrating an embodiment of the fault text-based classification method provided in this application. Figure 1 As shown, the classification method based on fault text may include steps S101 to S105.

[0030] In step S101, the fault text obtained from the wind farm is used as the fault text training set.

[0031] A wind farm comprises multiple wind turbine generators, and fault text can be obtained based on these generators. Fault text may include historical fault work orders, technical guidance documents, fault analysis reports, etc., and is not limited thereto. Fault text may contain a large amount of data, at least some of which can provide information support for fault analysis. For example, fault text may include, but is not limited to, fault name information, time information, wind farm name information, fault description information, and fault handling measures information.

[0032] A fault text training set can be formed using fault texts. The fault text training set may include at least one fault text. The fault text training set is used to determine the number of classifications required to classify fault feature words in the fault texts. Specifically, the fault text training set can be stored in a wind power corpus database, which stores various types of wind power corpora to provide information support for other tasks.

[0033] In step S102, the first fault feature word associated with the wind turbine generator is extracted from the fault text in the fault text training set.

[0034] The fault text training set contains multiple sentences, which can be segmented into feature words. From these feature words, the first fault feature word associated with the wind turbine generator set is extracted. The first fault feature word may contain wind turbine generator set elements and / or wind turbine generator set fault elements. Specifically, the first fault feature word can characterize fault-related features and / or features related to wind turbine generator set faults in the fault text. For example, the first fault feature word associated with the wind turbine generator set may include feature words containing the names of wind turbine generator set components, feature words containing descriptions of wind turbine generator set faults, feature words containing names of wind turbine generator set faults, feature words containing operation descriptions of the wind turbine generator set, etc., and is not limited here.

[0035] In some examples, the fault texts in the acquired fault text training set can be segmented to obtain the feature words of the fault texts. Invalid characters in the feature words are then removed. The first fault feature word is extracted from the feature words from which invalid characters have been removed. Invalid characters may include, but are not limited to, special characters, stop words, and words unrelated to wind turbine generators, or one or more of these. The specific range of invalid characters can be set according to the scenario, requirements, experience, etc., and is not limited here.

[0036] For example, the fault text includes the statement "1. The hydraulic pump pressure relay contact is not sensitive; 2. A new pressure relay needs to be replaced". After segmenting this statement in the fault text, the characteristic words of the fault text include "1", "、", "hydraulic pump", "pressure relay", "contact", "not sensitive", ";", "needs", and "replace". Invalid characters "1", "、", ";", "needs", and "replace" can be removed. The first fault characteristic words extracted can include "hydraulic pump", "pressure relay", "contact", "not sensitive", "replace", and "new".

[0037] In step S103, the first fault feature word is classified according to the preset number of candidate categories, and the classification result corresponding to each number of candidate categories is obtained.

[0038] N candidate classification numbers can be preset. Using each candidate classification number, the first fault feature word is classified, resulting in N classification results, where N is an integer greater than 1. For example, if the preset candidate classification numbers include 2, 3, and 4, three classification results can be obtained: the first classification result corresponding to candidate classification number 2 includes the first fault feature word divided into two categories; the second classification result corresponding to candidate classification number 3 includes the first fault feature word divided into three categories; and the third classification result corresponding to candidate classification number 4 includes the first fault feature word divided into four categories.

[0039] The number of candidate categories can include more than two candidate categories, and the specific number can be set according to the scenario, needs, experience, etc., and is not limited here. For example, the number of candidate categories can include, but is not limited to, one or more of 2, 3, 4, and 5, that is, the number of candidate categories can include two or more of 2, 3, 4, and 5.

[0040] The classification algorithm used in this application is not limited. For example, a dimensionality reduction classification algorithm, such as Linear Discriminant Analysis (LDA) or Bayesian model algorithm, can be used.

[0041] In step S104, based on the classification results, the number of candidate categories with the greatest difference in the first fault feature words of different categories is determined as the target number of categories.

[0042] The classification result characterizes the classification outcome, and may include the first fault feature word after being classified into the candidate category number, or the probability of the first fault feature word being classified into each category. Based on the classified first fault feature word, the difference between the first fault feature words in different categories can be determined; the greater the difference between the first fault feature words in different categories, the more accurate the classification of the first fault feature word. The target category number is used to classify fault feature words associated with wind turbine generators in the fault text to be classified, thereby improving the accuracy of fault feature word classification in the fault text to be classified.

[0043] For example, the preset number of candidate categories can be 2 or 3. For candidate category numbers of 2 and 3, the first fault feature words extracted from the wind power fault work order are classified respectively. When the number of candidate categories is 2, the first fault feature words classified into category 1 include "unit," "fault," "pressure," "damage," "fan," "contactor," "converter," "circuit breaker," "voltage," "wind farm," "temperature," and "water cooling," etc.; the first fault feature words classified into category 2 include "fault," "inspection," "detection," "replacement," "update," "pitch," "elimination," "tightening," "measurement," "observation," "restore to normal," and "view," etc. Based on the first fault feature words classified according to candidate category number 2, it can be seen that category 1 mainly contains fault information, that is, category 1 represents fault-related information; category 2 mainly contains operational information, that is, category 2 represents operational information. The differences between the first fault feature words of the different categories are relatively large, and the distinctions are relatively clear. When there are 3 candidate categories, the first fault characteristic words classified into category 1 include "unit", "fault", "pressure", "damage", "fan", "inspection", "update", "contactor", "water cooling", etc.; the first fault characteristic words classified into category 2 include "inspection", "replacement", "fault", "voltage", "pitch", "measurement", "converter", "circuit breaker", "temperature", "observation", "elimination", "restoration to normal", etc.; the first fault characteristic words classified into category 3 include "fault", "wind farm", "tightening", "pitch", "elimination", "viewing", etc. Based on the first fault feature words obtained after classifying the candidate data into three categories, we can see that Category 1 mainly contains fault name information, that is, Category 1 represents fault name information; Category 2 mainly contains fault description information, that is, Category 2 represents fault description information; and Category 3 mainly contains treatment measures information, that is, Category 3 represents treatment measures information. However, many first fault feature words are mixed together in these three categories. For example, first fault feature words that are biased towards treatment measures, such as "replacement" and "measurement", are classified into Category 2, which is fault description information. First fault feature words such as "elimination" and "pitch change" are classified into both Category 2 and Category 3. The difference between the first fault feature words of each category after classification is small and the distinction is not clear.

[0044] In some examples, the differences between first fault feature words of different categories can be determined by the probability that the first fault feature word is classified into different categories. The greater the difference between the probabilities of the first fault feature word being classified into different categories, the greater the difference between the first fault feature words of different categories.

[0045] In step S105, based on the target classification number, the second fault feature words associated with the wind turbine generator set in the fault text to be classified are classified to obtain the classified second fault feature words.

[0046] The second fault feature words are the fault feature words in the fault text to be classified that are associated with the wind turbine generator set. The method for obtaining the second fault feature words can be referred to in the above embodiment for retrieving the first fault feature words associated with the wind turbine generator set from the fault text in the fault text training set, and will not be repeated here.

[0047] Based on the target number of categories, the second fault feature words associated with wind turbine generators in the fault text to be classified are divided into categories according to the target number of categories. For example, if the target number of categories is 3, then the second fault feature words associated with wind turbine generators in the fault text are divided into three categories, resulting in the second fault feature words divided into three categories.

[0048] The second fault feature words after classification have been classified according to the most accurate number of classifications, which can be used more effectively for fault analysis and detection, thereby further improving the accuracy of fault analysis and detection processes.

[0049] In some examples, after step S105, the classified second fault feature words can also be controlled to be displayed on a display device for fault analysis of the wind turbine generator set. The display device can be independent of the fault text-based classification device or equipment, or it can be integrated into the fault text-based classification device or equipment; this is not limited thereto. Displaying the classified second fault feature words on the display device facilitates further operations by the user based on the displayed second fault feature words. Fault analysis may include fault warning, fault detection, fault handling, etc., and is not limited thereto.

[0050] Optionally, the wind turbine generator set can be controlled based on the results of the fault analysis described above. For example, the control of the wind turbine generator set may include, but is not limited to: shutting down the wind turbine generator set, shutting down a sector of the wind turbine generator set, adjusting the yaw system of the wind turbine generator set, adjusting the pitch system of the wind turbine generator set, etc.

[0051] In this embodiment, fault texts obtained from wind farms to form a fault text training set are used to classify the first fault feature words extracted from the fault texts in the fault text training set according to a preset number of candidate classifications, obtaining classification results corresponding to each candidate classification number. Based on the classification results, the candidate classification number that maximizes the difference between the first fault feature words of different categories after classification is selected as the target classification number. The target classification number is then used to classify the fault feature words associated with wind turbine generators in the fault text to be classified. The greater the difference between the first fault feature words of different categories, the higher the accuracy of classification using that candidate classification number. By selecting the candidate classification number with the greatest difference between the first fault feature words of different categories to classify the fault feature words associated with wind turbine generators in the fault text to be classified, the accuracy of classifying the content in the fault text is improved. Using the method of this embodiment to classify the content of fault texts, the classification accuracy can reach over 92%.

[0052] In some embodiments, the classification result may include the probability of the first fault feature word being assigned to each category. Figure 2 A flowchart of another embodiment of the fault text-based classification method provided in this application. Figure 2 and Figure 1 The difference is that, Figure 1 Step S103 can be further refined as follows: Figure 2 Step S1031, Figure 1 Step S104 can be further refined as follows: Figure 2 Steps S1041 to S1043 in the process, Figure 1 Step S105 can be further refined as follows: Figure 2 Steps S1051 and S1052 in the process.

[0053] In step S1031, for each of the multiple candidate categories, for example, the first fault feature word is divided into categories equal to the number of candidate categories using a linear discriminant analysis algorithm, and the probability of the first fault feature word being classified into each category is obtained.

[0054] If N candidate categories are preset, the first fault feature word is classified once for each candidate category. The number of categories into which the first fault feature word is classified under a certain number of candidate categories is the same as that number of candidate categories.

[0055] The probability of classifying the first fault feature word into a category can be obtained based on the probability of the first fault feature word in the fault text and the probability of the topic represented by this category in the fault text. In some examples, the probability of classifying the first fault feature word into a category can be obtained according to the following formula (1):

[0056] P(A|W)=P(A|Z)×P(Z|W) (1)

[0057] Where A is the first fault feature word, W is the fault text, Z is the subject represented by a certain category, P(A|W) is the probability of the first fault feature word in the fault text, P(A|Z) is the probability of the first fault feature word being classified into a certain category, and P(Z|W) is the probability of the subject represented by the classified category in the fault text.

[0058] In step S1041, the absolute value of the probability difference is calculated under any given number of candidate categories.

[0059] The absolute value of the probability difference includes the absolute value of the difference between the probabilities of the first fault feature word being classified into different categories.

[0060] In some examples, if the number of candidate categories is 2, the absolute value of the probability difference can be the absolute value of the difference between the probability of the first fault feature word being classified into category 1 and the probability of the first fault feature word being classified into category 2.

[0061] In other examples, if the number of candidate categories is an integer greater than 2, the absolute value of the probability difference can be the minimum of the absolute values ​​of the differences between the probabilities of the first fault feature being assigned to each of the two different categories, or the maximum of the absolute values ​​of the differences between the probabilities of the first fault feature being assigned to each of the two different categories, or the average of the absolute values ​​of the differences between the probabilities of the first fault feature being assigned to each of the two different categories, which is not limited here. For example, if the number of candidate categories is 3, the absolute value of the difference between the probability of the first fault feature word being classified into category 1 and the probability of the first fault feature word being classified into category 2 can be determined as the first absolute value; the absolute value of the difference between the probability of the first fault feature word being classified into category 2 and the probability of the first fault feature word being classified into category 3 can be determined as the second absolute value; the absolute value of the difference between the probability of the first fault feature word being classified into category 1 and the probability of the first fault feature word being classified into category 3 can be determined as the third absolute value; the minimum value among the first, second, and third absolute values ​​can be determined as the absolute value of the probability difference; or, the maximum value among the first, second, and third absolute values ​​can be determined as the absolute value of the probability difference; or, the average value among the first, second, and third absolute values ​​can be determined as the absolute value of the probability difference.

[0062] In step S1042, the absolute value of the conditional probability difference of a candidate classification number is determined as the difference degree of the first fault feature word of different categories.

[0063] In step S1043, the number of candidate categories corresponding to the largest absolute value of probability difference is determined as the number of target categories.

[0064] The larger the absolute value of the probability difference, the lower the probability that the first fault feature word is classified into more than two categories simultaneously, meaning the higher the classification accuracy. Determining the number of candidate categories corresponding to the largest absolute value of the probability difference as the target number of categories ensures more accurate classification of fault feature words in the fault text to be classified in subsequent processes.

[0065] For example, the preset number of candidate categories includes 2, 3, 4, and 5. Table 1 shows the probability of the first fault feature word being classified into different categories under each number of candidate categories. Table 1 is as follows:

[0066] Table 1

[0067]

[0068]

[0069] As shown in Table 1, the absolute value of the probability difference is the largest when the number of candidate categories is 2. Therefore, 2 is taken as the number of target categories.

[0070] In step S1051, the target number of categories is set as the number of categories of the classifier, and the classifier is trained using fault training data labeled with categories to obtain the target classifier.

[0071] Fault training data can be obtained from wind power corpus databases or other sources; there are no limitations on this. A training algorithm can be used to train the classifier so that the number of categories is equal to the target number of categories. The training algorithm used to train the classifier is not limited; for example, BERT (Bidirectional Encoder Representations from Transformers) or the sigmode function can be used. The target classifier can be used to classify the input fault feature words into the target number of categories.

[0072] In step S1052, the target classifier is used to classify the second fault feature words associated with the wind turbine generator in the fault text to be classified, and the classified second fault feature words are obtained.

[0073] Specifically, a second fault feature word associated with the wind turbine generator can be extracted from the fault text to be classified. The second fault feature word is the fault feature word associated with the wind turbine generator extracted from the fault text to be classified. The extraction method of the second fault feature word can refer to the extraction method of the first fault feature word in the above embodiment, and will not be repeated here.

[0074] The number of classifications in the target classifier is the target number of classifications. The target number of classifications is selected from multiple candidate number of classifications based on the classification results of the first fault feature word under the condition of multiple candidate number of classifications. This target number of classifications is a better classification result for the first fault feature word, which improves the accuracy of the classification of the first fault feature word. The second fault feature word is associated with the wind turbine generator set and is also extracted from the fault text. It has similar characteristics to the first fault feature word. Therefore, using a classifier with the number of classifications as the target number of classifications to classify the second fault feature word can also improve the accuracy of classifying the content associated with the wind turbine generator set in the fault text.

[0075] Figure 3 A flowchart of yet another embodiment of the fault text-based classification method provided in this application. Figure 3 and Figure 1 The difference is that, Figure 3 The classification method based on fault text shown may further include steps S106 to S109. Figure 1 Step S105 can be further refined as follows: Figure 3 Step S1053 in the process.

[0076] In step S106, a fault text test set is obtained from the wind power corpus database.

[0077] The fault text test set includes fault texts. At least some fault texts in the fault text test set differ from those in the fault text training set. That is, the fault text test set and the fault text training set are not entirely identical. In some examples, to ensure the effectiveness of the test, the intersection of the fault text test set and the fault text training set is an empty set.

[0078] In step S107, a third fault feature word associated with the wind turbine generator is extracted from the fault texts in the acquired fault text test set.

[0079] The third fault feature word is a fault feature word associated with the wind turbine generator set extracted from the fault texts in the obtained fault text test set. The extraction method of the third fault feature word can be referred to the extraction method of the first fault feature word in the above embodiment, and will not be repeated here.

[0080] In step S108, the third fault feature words are classified based on the number of target categories.

[0081] In some examples, a target classifier trained to classify fault feature words into categories of the target number of categories can be used to classify the third fault feature word, which is not limited here.

[0082] In step S109, the classification accuracy of the third fault feature word after classification is obtained.

[0083] Each category of third fault feature words after classification can be compared with the third fault feature words after accurate classification using other methods, thereby obtaining the classification accuracy of the third fault feature words using the target number of classifications.

[0084] In step S1053, if the classification accuracy of the third fault feature word is higher than the preset test standard accuracy, the second fault feature word in the fault text to be classified is classified based on the target classification number, and the classified second fault feature word is obtained.

[0085] The test standard accuracy is used to determine whether the accuracy of the above classification method meets the expected standard. The specific setting can be determined according to the scenario, requirements, experience, etc., and is not limited here. If the classification accuracy of the third fault feature word is higher than or equal to the preset test standard accuracy, it means that the accuracy of the above classification method has met the expected standard and can be put into use, that is, the classification of the second fault feature word in the fault text to be classified.

[0086] If the classification accuracy of the third fault feature word is less than the preset test standard accuracy, the classifier, device, or equipment can be updated or retrained until the classification accuracy of the third fault feature word is greater than or equal to the preset test standard accuracy.

[0087] A second aspect of this application provides a classification device based on fault text. Figure 4 This is a schematic diagram of an embodiment of the fault text-based classification device provided in this application. Figure 4 As shown, the fault text-based classification device 200 may include an acquisition module 201, an extraction module 202, a traversal processing module 203, a classification number determination module 204, and a classification module 205.

[0088] The acquisition module 201 can be used to obtain fault texts from wind farms as a fault text training set.

[0089] The fault text training set is stored in the wind power corpus database.

[0090] The extraction module 202 can be used to extract the first fault feature words associated with the wind turbine generator from the fault text in the fault text training set.

[0091] The traversal processing module 203 can be used to classify the first fault feature word according to a preset number of candidate categories, and obtain the classification result corresponding to each number of candidate categories.

[0092] In some examples, the number of candidate categories includes one or more of the following: 2, 3, 4, 5.

[0093] The classification number determination module 204 can be used to determine the number of candidate classifications with the greatest difference in the first fault feature words of different categories as the target classification number based on the classification results.

[0094] The classification module 205 can be used to classify the second fault feature words associated with the wind turbine generator in the fault text to be classified based on the target classification number, and obtain the classified second fault feature words.

[0095] In this embodiment, fault texts obtained from wind farms to form a fault text training set are used to classify the first fault feature words extracted from the fault texts in the fault text training set according to a preset number of candidate classifications, obtaining classification results corresponding to each candidate classification number. Based on the classification results, the candidate classification number that maximizes the difference between the first fault feature words of different categories after classification is selected as the target classification number. The target classification number is then used to classify the fault feature words associated with wind turbine generators in the fault text to be classified. The greater the difference between the first fault feature words of different categories, the higher the accuracy of the classification of the first fault feature words using that candidate classification number. By selecting the candidate classification number with the greatest difference between the first fault feature words of different categories to classify the fault feature words associated with wind turbine generators in the fault text to be classified, the accuracy of classifying the content in the fault text is improved.

[0096] In some embodiments, the classification module 205 can also be used to control the classification of the second fault feature words to be displayed on the display device for fault analysis of the wind turbine generator set.

[0097] In some embodiments, the fault text classification device may further include a control module. The control module can be used to control the wind turbine generator set based on the results of fault analysis.

[0098] In some embodiments, the classification result includes the probability of the first fault feature word being assigned to each category.

[0099] The above-mentioned classification number determination module 204 can be used to: calculate the absolute value of the probability difference under any candidate classification number condition, the absolute value of the probability difference includes the absolute value of the difference in the probability of the first fault feature word being classified into different categories; determine the absolute value of the probability difference under a candidate classification number condition as the difference degree of the first fault feature word in different categories; and determine the candidate classification number corresponding to the largest absolute value of the probability difference as the target classification number.

[0100] In some embodiments, the classification result includes the probability of the first fault feature word being assigned to each category.

[0101] The aforementioned traversal processing module 203 can be used to: for each of the multiple candidate categories, use a linear discriminant analysis algorithm to classify the first fault feature word into categories equal to the number of candidate categories, and obtain the probability of the first fault feature word being classified into each category.

[0102] The classification module 205 described above can be used to: set the target number of classifications as the number of classifications of the classifier, train the classifier using fault training data labeled with categories to obtain the target classifier, and use the target classifier to classify the second fault feature words associated with the wind turbine generator in the fault text to be classified.

[0103] In some embodiments, the extraction module 202 can be used to: segment the fault texts in the acquired fault text training set to obtain the feature words of the fault texts; remove invalid characters from the feature words; and extract the first fault feature word from the feature words from which invalid characters have been removed.

[0104] Figure 5 A schematic diagram of another embodiment of the fault text-based classification device provided in this application. Figure 5 and Figure 4 The difference is that, Figure 5 The fault text-based classification device 200 shown may also include a test set generation module 206 and an accuracy determination module 207.

[0105] The test set generation module 206 can be used to obtain a fault text test set from the wind power corpus database. The fault text test set includes fault text.

[0106] The extraction module 202 described above can also be used to extract third fault feature words associated with wind turbine generators from the fault texts in the acquired fault text test set.

[0107] The aforementioned classification module 205 can also be used to classify third fault feature words based on the number of target categories.

[0108] The accuracy determination module 207 can be used to obtain the classification accuracy of the third fault feature word after classification.

[0109] The aforementioned classification module 205 can be used to classify the second fault feature words in the fault text to be classified based on the target number of classifications, when the classification accuracy of the third fault feature word is higher than or equal to the preset test standard accuracy.

[0110] A third aspect of this application also provides a classification device based on fault text. Figure 6 This is a schematic diagram of the structure of an embodiment of the fault text-based classification device provided in this application. Figure 6As shown, the fault text-based classification device 300 includes a memory 301, a processor 302, and a computer program stored on the memory 301 and executable on the processor 302.

[0111] In one example, the processor 302 described above may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits that may be configured to implement the embodiments of this application.

[0112] Memory 301 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the fault text-based classification method according to embodiments of this application.

[0113] The processor 302 runs a computer program corresponding to the executable program code by reading the executable program code stored in the memory 301, in order to implement the fault text-based classification method in the above embodiments.

[0114] In one example, the fault text-based classification device 300 may further include a communication interface 303 and a bus 304. Wherein, as Figure 6 As shown, the memory 301, processor 302, and communication interface 303 are connected through bus 304 and complete communication with each other.

[0115] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application. Input devices and / or output devices can also be connected through the communication interface 303.

[0116] Bus 304 includes hardware, software, or both, that couples components of fault text-based classification device 300 together. For example, and not limitingly, bus 304 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-E) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 304 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0117] A fourth aspect of this application also provides a computer-readable storage medium storing computer program instructions. When executed by a processor, these computer program instructions can implement the fault text-based classification method described in the above embodiments and achieve the same technical effect. To avoid repetition, further details are omitted here. The aforementioned computer-readable storage medium may include non-transitory computer-readable storage media, such as read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, etc., and is not limited thereto.

[0118] It should be clarified that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. For the device embodiments, equipment embodiments, and computer-readable storage medium embodiments, the relevant parts can be referred to the description section of the method embodiments. This application is not limited to the specific steps and structures described above and shown in the figures. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application. Furthermore, for the sake of brevity, detailed descriptions of known methods and techniques are omitted here.

[0119] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0120] Those skilled in the art will understand that the above embodiments are exemplary and not restrictive. Different technical features appearing in different embodiments can be combined to achieve beneficial effects. Based on a study of the drawings, specification, and claims, those skilled in the art should be able to understand and implement other variations of the disclosed embodiments. In the claims, the term "comprising" does not exclude other means or steps; the quantifier "a" does not exclude a plurality; the terms "first" and "second" are used to identify names and not to indicate any particular order. No reference numerals in the claims should be construed as limiting the scope of protection. The functionality of multiple parts appearing in the claims can be implemented by a single hardware or software module. The appearance of certain technical features in different dependent claims does not mean that these technical features cannot be combined to achieve beneficial effects.

Claims

1. A classification method based on fault text, characterized in that, include: The fault texts obtained from the wind farm are used as the fault text training set, which is stored in the wind power corpus database. Extract the first fault feature words associated with the wind turbine generator from the fault texts in the fault text training set; According to the preset number of candidate categories, the first fault feature words are classified respectively to obtain the classification results corresponding to each number of candidate categories; Based on the classification results, the number of candidate categories with the largest difference in the first fault feature words of different categories is determined as the target number of categories; Based on the target number of classifications, the second fault feature words associated with wind turbine generators in the fault text to be classified are classified to obtain the classified second fault feature words; The classification result includes the probability of the first fault feature word being assigned to each category; The step of classifying the first fault feature word according to a preset number of candidate classifications, and obtaining the classification result corresponding to each of the preset number of candidate classifications, includes: For each of the multiple candidate categories, a linear discriminant analysis algorithm is used to classify the first fault feature word into categories equal to the number of candidate categories, and the probability of the first fault feature word being classified into each category is obtained. The classification of second fault feature words associated with wind turbine generators in the fault text to be classified, based on the target classification number, includes: The target number of categories is set as the number of categories of the classifier. The classifier is trained using fault training data labeled with categories to obtain the target classifier. The target classifier is used to classify the second fault feature words associated with wind turbine generators in the fault text to be classified.

2. The method according to claim 1, characterized in that, Following the second fault feature word obtained after classification, the following is also included: The second fault feature word after control classification is displayed on the display device for fault analysis of the wind turbine generator set.

3. The method according to claim 1, characterized in that, The classification result includes the probability of the first fault feature word being assigned to each category; The step of determining the number of candidate categories with the highest difference in the first fault feature words of different categories as the target number of categories based on the classification results includes: Under any given number of candidate categories, the absolute value of the probability difference is calculated, whereby the absolute value of the probability difference between the first fault feature word and the probability of the first fault feature word being classified into different categories is the absolute value of the probability difference. The absolute value of the probability difference under the condition of one number of candidate categories is determined as the degree of difference of the first fault feature words in different categories; The number of candidate categories corresponding to the largest absolute value of the probability difference is determined as the number of target categories.

4. The method according to claim 1, characterized in that, The number of candidate categories includes at least two of the following: 2、3、4、5。 5. The method according to claim 1, characterized in that, Before classifying the second fault feature words associated with wind turbine generators in the fault text to be classified based on the target classification number, the method further includes: Obtain a fault text test set from the wind power corpus database, the fault text test set including fault text; Extract third fault feature words associated with wind turbine generators from the fault texts in the obtained fault text test set; The third fault feature word is classified based on the number of target categories; Obtain the classification accuracy of the third fault feature word after classification; The classification of the second fault feature words in the fault text to be classified based on the target classification number includes: If the classification accuracy of the third fault feature word is higher than or equal to the preset test standard accuracy, the second fault feature word in the fault text to be classified is classified based on the target number of classifications.

6. The method according to claim 1, characterized in that, The step of extracting the first fault feature word associated with the wind turbine generator from the fault text in the fault text training set includes: The fault texts in the acquired fault text training set are segmented to obtain the feature words of the fault texts; Remove invalid characters from the feature words; Extract the first fault feature word from the feature words after removing the invalid characters.

7. The method according to claim 2, characterized in that, Also includes: Based on the results of fault analysis of the wind turbine generator set, the wind turbine generator set is controlled.

8. A classification device based on faulty text, characterized in that, include: The acquisition module is used to obtain fault texts from wind farms as a fault text training set, which is stored in a wind power corpus database. The extraction module is used to extract the first fault feature word associated with the wind turbine generator from the fault text in the fault text training set. The traversal processing module is used to classify the first fault feature words according to a preset number of candidate categories, and obtain the classification results corresponding to each number of candidate categories. The classification number determination module is used to determine the number of candidate classifications with the largest difference between the first fault feature words of different categories as the target number of classifications based on the classification results. The classification module is used to classify the second fault feature words associated with the wind turbine generator in the fault text to be classified based on the target number of classifications, and obtain the classified second fault feature words. The classification result includes the probability of the first fault feature word being assigned to each category; The traversal processing module is used to: for each of the multiple candidate classification numbers, use a linear discriminant analysis algorithm to classify the first fault feature word into categories with a number equal to the number of candidate classification numbers, and obtain the probability of the first fault feature word being classified into each category; The classification module is used to: set the target number of classifications as the number of classifications of the classifier, train the classifier using fault training data labeled with categories to obtain the target classifier, and use the target classifier to classify the second fault feature words associated with the wind turbine generator in the fault text to be classified.

9. A classification device based on fault text, characterized in that, include: Processor and memory storing computer program instructions; When the processor executes the computer program instructions, it implements the fault text-based classification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the fault text-based classification method as described in any one of claims 1 to 7.

Images