Wind power fault information extraction method and device

Abstract

A method and device for extracting wind power fault information are provided, the method comprising: obtaining text data about wind power faults; performing entity noun extraction processing on the obtained text data; and determining first extraction information and second extraction information of the text data based on the obtained text data and the entity noun extraction processing result, wherein the first extraction information is used to characterize the fault point where the fault occurs, and the second extraction information is used to characterize the fault-related description for the fault point. The method and device for extracting wind power fault information according to the exemplary embodiment of the present invention effectively improve the text mining effect of the wind power industry.

Description

Technical Field

[0001] The present invention generally relates to the technical field of wind power generation, and more specifically, to a method and device for extracting wind power fault information. Background Art

[0002] After experiencing the baptism of the big data era, the wind power industry has accumulated a huge amount of text data. However, these text data are generally messy and have no unified format, making it difficult to mine the value of text data.

[0003] At present, text data in the wind power industry is spread across various stages, such as product development, supply chain, marketing, and service. There are a large number of contracts for various models and batches in the supply chain. Due to factors such as time, tasks, and human factors, it is difficult to accurately and efficiently extract key information. In the service link, service work orders are messy and the recording methods are diverse, making manual review difficult. Historical work order failure data is difficult to trace and its value is difficult to mine.

[0004] The text data in the wind power industry is different from structured data such as SCADA. The text data in the wind power industry is unstructured and messy. At present, when business personnel want to use historical text data, they often use manual search. This manual search method is not only inefficient, but also often finds incomplete information. Summary of the invention

[0005] An exemplary embodiment of the present invention aims to provide a method and device for extracting wind power fault information to overcome at least one of the above-mentioned drawbacks.

[0006] In a general aspect, a method for extracting wind power fault information is provided, the method comprising: acquiring text data about wind power faults; performing entity noun extraction processing on the acquired text data; and determining first extraction information and second extraction information of the text data based on the acquired text data and the entity noun extraction processing result, wherein the first extraction information is used to characterize a fault point where a fault occurs, and the second extraction information is used to characterize a fault-related description of the fault point.

[0007] Optionally, the step of performing entity noun extraction processing on the acquired text data may include: extracting entity nouns from the acquired text data based on a named entity recognition (NER) algorithm.

[0008] Optionally, the entity nouns obtained through the extraction process may include at least one first entity noun for indicating a fault point and at least one second entity noun for describing a fault, wherein the step of determining first extraction information and second extraction information of the text data based on the acquired text data and the entity noun extraction process results may include: referring to the text features of wind power faults, and according to the acquired text data, splicing based on the at least one first entity noun to obtain first extraction information, and splicing based on the at least one second entity noun to obtain second extraction information.

[0009] Optionally, the first extraction information can be obtained by one of the following methods: concatenating two first entity nouns that are adjacent in position in the text data to obtain the first extraction information; concatenating two first entity nouns that satisfy a noun-conjunction-noun phrase structure and a conjunction between the two first entity nouns to obtain the first extraction information.

[0010] Optionally, the step of obtaining first extraction information based on splicing of at least one first entity noun may include: determining words in the text data that are adjacent to any first entity noun among the at least one first entity noun; determining the category of the adjacent words; and splicing any first entity noun based on the category of the adjacent words to obtain first extraction information.

[0011] Optionally, the first extraction information can be obtained in the following manner: segmenting the part of the text data except for the extracted at least one first entity noun and the at least one second entity noun to obtain at least one participle; if it is determined that the category of the adjacent word is another first entity noun, concatenating any one of the first entity nouns with the other first entity noun to obtain the first extraction information; if it is determined that the category of the adjacent word is a participle, determining the part of speech of the participle; if it is determined that the part of speech of the participle is a conjunction, determining whether another word adjacent to the participle is another first entity noun or a noun; if the other word adjacent to the participle is another first entity noun or a noun, concatenating any one of the first entity nouns with the participle and another word adjacent to the participle to obtain the first extraction information.

[0012] Optionally, the second extraction information can be obtained by one of the following methods: concatenating two second entity nouns that are adjacent in position in the text data to obtain the second extraction information; concatenating the second entity noun that satisfies the verb-entity noun phrase structure and the verb that precedes the second entity noun to obtain the second extraction information.

[0013] Optionally, the step of obtaining second extraction information based on splicing of at least one second entity noun may include: determining words in the text data that are adjacent to any second entity noun in the at least one second entity noun; determining the category of the adjacent words; and splicing any second entity noun based on the category of the adjacent words to obtain second extraction information.

[0014] Optionally, the second extraction information can be obtained in the following manner: segmenting the part of the text data except for the extracted at least one first entity noun and the at least one second entity noun to obtain at least one participle; if it is determined that the category of the adjacent word is another second entity noun, concatenating any second entity noun with the other second entity noun to obtain the second extraction information; if it is determined that the category of the adjacent word is a participle, determining the part of speech of the participle; if it is determined that the part of speech of the participle is a verb, and the participle is located before any second entity noun, concatenating any second entity noun with the participle to obtain the second extraction information.

[0015] Optionally, the wind power fault information extraction method may further include extracting at least one of the following items from the acquired text data: event type, name of the faulty wind turbine set, name of the faulty system, fault cause, fault handling method, and fault handling result.

[0016] In another general aspect, a wind power fault information extraction device is provided, the wind power fault information extraction device comprising: a text data acquisition module, which acquires text data about wind power faults; a word extraction processing module, which performs entity noun extraction processing on the acquired text data; an extraction information determination module, which determines first extraction information and second extraction information of the text data based on the acquired text data and the entity noun extraction processing result, wherein the first extraction information is used to characterize a fault point where a fault occurs, and the second extraction information is used to characterize a fault-related description of the fault point.

[0017] In another general aspect, a controller is provided, comprising: a processor; and a memory for storing a computer program, wherein the computer program implements the above-mentioned method for extracting wind power fault information when executed by the processor.

[0018] In another general aspect, a computer-readable storage medium storing a computer program is provided, and when the computer program is executed by a processor, the method for extracting wind power fault information as described above is implemented.

[0019] The wind power fault information extraction method and device according to the exemplary embodiment of the present invention effectively improves the text mining effect in the wind power industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other objects, features and advantages of exemplary embodiments of the present invention will become more apparent from the following detailed description taken in conjunction with the accompanying drawings which exemplarily illustrate exemplary embodiments.

[0021] Figure 1 A flow chart showing a method for extracting wind power fault information according to an exemplary embodiment of the present invention;

[0022] Figure 2A and Figure 2B An example diagram showing a trouble ticket according to an exemplary embodiment of the present invention;

[0023] Figure 3 An example diagram showing a first entity noun according to an exemplary embodiment of the present invention;

[0024] Figure 4 A flowchart showing steps of obtaining first extraction information according to an exemplary embodiment of the present invention;

[0025] Figure 5 A flowchart showing the steps of concatenating a first entity noun according to an exemplary embodiment of the present invention;

[0026] Figure 6 A flowchart showing steps of obtaining second extraction information according to an exemplary embodiment of the present invention;

[0027] Figure 7 A flowchart showing the steps of concatenating the second entity noun according to an exemplary embodiment of the present invention;

[0028] Figure 8 An example diagram showing wind power fault information extraction according to an exemplary embodiment of the present invention;

[0029] Fig. 9 A schematic diagram showing a comparison between the extraction results based on the NER algorithm for different text data according to an exemplary embodiment of the present invention and the extraction results based on the wind power fault information extraction method according to an exemplary embodiment of the present invention;

[0030] Fig.10 A block diagram showing a wind power fault information extraction device according to an exemplary embodiment of the present invention is shown;

[0031] Fig.11 A block diagram of a controller according to an exemplary embodiment of the present invention is shown. DETAILED DESCRIPTION

[0032] Various example embodiments will now be described more fully with reference to the accompanying drawings, in which some example embodiments are shown.

[0033] Figure 1A flow chart of a method for extracting wind power fault information according to an exemplary embodiment of the present invention is shown.

[0034] Reference Figure 1 , in step S10, text data about wind power faults is obtained.

[0035] For example, text data about wind power failures may be obtained from a corpus of the wind power industry. As an example, the corpus may be formed based on failure work orders, technical contracts, business contracts, etc. of the wind power industry.

[0036] The wind power fault information extraction method of the exemplary embodiment of the present invention can extract fault information from the text data of the fault work order. In one example, the fault work order can be as follows: Figure 2A and Figure 2B As shown in Table 1-1 and Table 1-2.

[0037] by Figure 2A and Figure 2B As shown in the example, the text data about the wind power failure may include the fault description of the wind power failure (such as the description of the cause of the failure and the fault phenomenon) or the handling measures and handling results of the wind power failure.

[0038] That is, the fault work order includes multiple fault records (such as a row of data in Table 1). In an exemplary embodiment of the present invention, fault information extraction can be performed for a fault description or a processing measure in a fault record of the fault work order.

[0039] In step S20, entity noun extraction processing is performed on the acquired text data.

[0040] Here, various algorithms in the field of natural language processing (NLP) can be used to extract entity nouns from text data. That is, in an exemplary embodiment of the present invention, natural language processing technology is applied to the wind power industry to mine data value from unstructured data such as text data.

[0041] Information extraction mainly includes entity extraction and relationship extraction. Entity extraction refers to the extraction of proper nouns in the wind power industry, and relationship extraction is the various relationships defined on the basis of entities. There are many entity extraction algorithms in the NLP field, such as HMM, CRF, and LSTM+CRF. Table 2 shows the comparison of the accuracy of various entity extraction algorithms.

[0042] Table 2

[0043] algorithm Accuracy (%) Combination of HMM,Maxent etc. (Florian et al., 2003) 89.76 MaxEnt classifier(Chieu.,2003) 88.31 Semi-supervised model combination(Ando and Zhang.,2005) 89.31 Conv-CRF (Collobert et al., 2011) 81.47 Conv-CRF(Senna+Gazetteer)(Collobert et al., 2011) 89.59 CRF with Lexicon Infused Embeddings (Passos et al., 2014) 90.90 BI-LSTM-CRF(ours) 84.26 BI-LSTM-CRF(Senna+Gazetteer)(ours) 90.10

[0044] In an optional example, entity nouns may be extracted from the acquired text data based on a named entity recognition (NER) algorithm, that is, the output of entity words may be obtained through the NER algorithm.

[0045] Here, the extracted entity nouns refer to entities with specific meanings or strong referentiality in the text data. In an exemplary embodiment of the present invention, the entity nouns may include customized proper nouns in the wind power industry.

[0046] In a preferred example, the BERT (Bidirectional Encoder Representation from Transformers) algorithm may be used to implement the NER function, that is, entity nouns are obtained based on BERT-NER.

[0047] Table 3 shows a comparison of the experimental results of different NER algorithms.

[0048] Table 3

[0049] Tool Classification epoch Training duration Test set F1 value Model loading speed Prediction speed BERTNER 30 5h 49m 96.18 2.8s 80ms NeuroNER 30 14h 19m 91.93 23s 2s NeuroNER 100 2d 7h 10m 92.33 23s 2s

[0050] In Table 3, epoch means period, d means day, h means hour, m means minute, s means second, and ms means millisecond. As shown in Table 3, after the two algorithms are trained with the same number of iterations, the accuracy of NER information extraction using the BERT algorithm can be as high as 96.18%, and the F1 value of the test set is about 4 percentage points higher than that of NeuroNER. Even if the NeuroNER iteration epoch is increased to 100, BertNER still has better recognition effect.

[0051] The BERTNER algorithm has great advantages in training time, model loading speed, and prediction speed, reaching industrial level, and is more suitable for application in the production environment of the wind power industry. Based on this, in a preferred example of the present invention, the BERT algorithm can be used to implement the NER function to extract entity nouns from the acquired text data.

[0052] As an example, the entity nouns obtained through the extraction process may include at least one first entity noun for indicating a fault point and at least one second entity noun for fault description.

[0053] In one example, proper nouns that need to be identified can be manually extracted from fault work orders in the wind power industry and marked with custom parts of speech. For example, the first entity noun can be marked as PA, and the second entity noun can be marked as INFO. Figure 3 Table 4 shown is an example of at least one first entity noun PA used to indicate a fault point.

[0054] In this case, the fault ticket and the annotated PA and INFO can be used to train the NER algorithm to improve the accuracy of entity noun extraction.

[0055] For example, the BERTNER algorithm is used to train on the corpus of the wind power industry. The training results are as follows:

[0056] Input data: text = 'The electronic components inside the anemometer were found to be damaged'

[0057] Output data: BERTNER recognition result = [['Anemometer', 'PA'], ['Electronic components', 'PA'], ['Damage', 'INFO']]

[0058] In step S30, first extraction information and second extraction information of the acquired text data are determined based on the acquired text data and the entity noun extraction processing result.

[0059] Here, the first extracted information is used to characterize the fault point where the fault occurs, and the second extracted information is used to characterize the fault-related description for the fault point. For example, the fault-related description for the fault point may include a description of the fault cause or fault phenomenon for the fault point or a description of the treatment measures or treatment results for the fault point.

[0060] Although the above NER algorithm can extract entity nouns very well, the extracted entity nouns are fragmented information, and a lot of precise information cannot be accurately extracted, so the application effect in the wind power industry is poor.

[0061] For example, for the text data "damage was found in the electronic components inside the anemometer", according to information extraction theory, effective information should contain accurate component information of the wind turbine and fault description information, that is, "electronic components inside the anemometer" should be a whole, but in the NER algorithms listed above, "anemometer" and "electronic components" are identified as two independent entity nouns, that is, the extraction results of the NER algorithm are biased.

[0062] In an exemplary embodiment of the present invention, the CFG+NER algorithm is used to solve the above problem in combination with the text features of the wind power industry and the grammatical structure of the information text. That is, on the basis of the end-to-end model, the CFG algorithm is used to accurately extract fault information in combination with the text features of the wind power industry and the grammatical structure of the information text.

[0063] Syntactic analysis is one of the key underlying technologies in natural language processing. Its basic task is to determine the syntactic structure of a sentence or the dependency relationship between words in a sentence.

[0064] Syntactic analysis is divided into syntactic structure parsing and dependency parsing. Its basic task is to determine the syntactic structure of the entire sentence or the complete phrase structure, that is, the dependency relationship between words in a sentence, which is called constituent structure parsing or phrase structure parsing.

[0065] The CFG algorithm is a grammatical structure analysis algorithm, which mainly extracts noun phrases, verb phrases, etc. In an exemplary embodiment of the present invention, with the help of the CFG algorithm idea, referring to the text features of wind power faults, according to the acquired text data, splicing based on at least one first entity noun to obtain first extraction information. And / or, referring to the text features of wind power faults, according to the acquired text data, splicing based on at least one second entity noun to obtain second extraction information.

[0066] For example, the first extraction information can be obtained by one of the following methods: concatenating two first entity nouns that are adjacent to each other in the text data to obtain the first extraction information; concatenating two first entity nouns that satisfy the noun-conjunction-noun phrase structure and a conjunction between the two first entity nouns to obtain the first extraction information.

[0067] Refer to the following Figure 4 The step of obtaining first extracted information based on at least one first entity noun is introduced. It should be understood that Figure 4 The illustrated method of obtaining the first extraction information is only an example, and the present invention is not limited thereto, and the first extraction information may also be obtained in other ways.

[0068] Figure 4 A flow chart showing steps of obtaining first extraction information according to an exemplary embodiment of the present invention.

[0069] Reference Figure 4 In step S301, a word adjacent to any first entity noun in at least one first entity noun in the acquired text data is determined.

[0070] In step S302, the category of the word adjacent to any first entity noun is determined.

[0071] In step S303, any first entity noun is concatenated based on the categories of words adjacent to any first entity noun to obtain first extraction information.

[0072] In an exemplary embodiment of the present invention, the text feature of wind power fault indicates the text expression habit about the fault point in the field of wind power. For example, based on the text feature of wind power fault, it can be determined that nouns (such as component names, component models, component manufacturer names, etc.) are usually related to the fault point. Based on this, any first entity noun can be spliced ​​according to the category of the words adjacent to any first entity noun to obtain the first extraction information.

[0073] Figure 5 A flowchart showing the steps of concatenating the first entity noun according to an exemplary embodiment of the present invention is shown. It should be understood that Figure 5 The manner of concatenating the first entity nouns shown is only an example, and the present invention is not limited thereto, and the first entity nouns may also be concatenated in other manners.

[0074] Reference Figure 5 In step S31, the part of the acquired text data except the extracted at least one first entity noun and at least one second entity noun is segmented to obtain at least one segmented word.

[0075] Here, various word segmentation methods may be used to segment text data to obtain at least one word segment.

[0076] In an exemplary embodiment of the present invention, entity noun extraction processing results PA and INFO are taken as a part of the grammatical structure, at the same level as verbs and nouns.

[0077] Taking the text data "Anemometer internal electronic components malfunction" as an example, combined with the entity noun extraction processing results, "Anemometer internal electronic components malfunction" can be divided into: "PA: anemometer, adj: internal, PA: electronic components, v: occurrence, INFO: malfunction" (such as Figure 6 shown).

[0078] In step S32, it is determined whether the category of a word adjacent to any first entity noun is another first entity noun.

[0079] As an example, the word category may include entity nouns and participles. Here, based on the text features of wind power failure, it is found that nouns are usually related to the fault point. Therefore, any first entity noun can be spliced ​​by finding whether there is a noun that can be spliced ​​in the words adjacent to any first entity noun.

[0080] If it is determined that the category of the word adjacent to any first entity noun is another first entity noun, step S33 is executed: any first entity noun is concatenated with another first entity noun to obtain first extraction information.

[0081] If it is determined that the category of the word adjacent to any first entity noun is not another first entity noun, that is, it is determined that the category of the adjacent word is a participle, step S34 is executed: determining the part of speech of the participle.

[0082] As examples, parts of speech may include, but are not limited to, nouns, verbs, adjectives, quantifiers, pronouns, adverbs, prepositions, conjunctions, auxiliary words, interjections, onomatopoeias, and the like.

[0083] In step S35, it is determined whether the part of speech of the participle is a conjunction.

[0084] If it is determined that the part of speech of the participle is a conjunction, step S36 is executed: determining whether another word adjacent to the participle is another first entity noun or a noun.

[0085] Here, a word adjacent to the participle is any first entity noun, and at this time it can be determined whether another word adjacent to the participle is another first entity noun or a noun.

[0086] In one example, it is further determined whether the part of speech of the participle is a noun. If it is determined that the part of speech of the participle is a noun, any first entity noun can be directly concatenated with the noun. If it is determined that the part of speech of the participle is not a noun, step S35 is executed to determine whether the part of speech of the participle is a conjunction.

[0087] If the other word adjacent to the participle is another first entity noun or a noun, step S37 is executed: any first entity noun is concatenated with the participle and the other word adjacent to the participle to obtain first extraction information.

[0088] If it is determined that the part of speech of the participle is not a conjunction, or it is determined that another word adjacent to the participle is not another first entity noun or noun, the first entity noun is not spliced. At this time, at least one first entity noun used to indicate a fault point can be determined as the first extraction information.

[0089] With the above Figure 6 The text data shown takes "Anemometer's internal electronic components malfunctioned" as an example. For the segmentation result: "PA: Anemometer, adj: Internal, PA: Electronic components", assuming that any first entity noun is "PA: Anemometer", since the category of the word adjacent to it is a participle ("adj: Internal"), the participle is a conjunction, and the other word adjacent to the participle is another first entity noun ("PA: Electronic components"), therefore, the three can be spliced ​​together, and the final output first extracted information is PA: Anemometer's internal electrical components.

[0090] For example, the second extraction information can be obtained by one of the following methods: concatenating two second entity nouns that are adjacent to each other in the text data to obtain the second extraction information; concatenating the second entity noun that satisfies the verb-entity noun phrase structure and the verb that precedes the second entity noun to obtain the second extraction information.

[0091] Refer to the following Figure 7 The step of obtaining the second extracted information based on at least one second entity noun is introduced. It should be understood that Figure 7 The illustrated method of obtaining the second extraction information is only an example, and the present invention is not limited thereto, and the second extraction information may also be obtained in other ways.

[0092] Figure 7 A flowchart showing steps of obtaining second extraction information according to an exemplary embodiment of the present invention.

[0093] Reference Figure 7 In step S310, a word adjacent to any second entity noun in the at least one second entity noun in the acquired text data is determined.

[0094] In step S320 , the category of the word adjacent to any second entity noun is determined.

[0095] In step S330, any second entity noun is concatenated based on the category of words adjacent to any second entity noun to obtain second extraction information.

[0096] In an exemplary embodiment of the present invention, the text features of wind power faults may further indicate text expression habits regarding fault triggers (which may also be fault causes, fault phenomena, and fault handling) in the field of wind power. For example, based on the text features of wind power faults, it can be determined that verbs are usually related to fault triggers. Based on this, any second entity noun can be spliced ​​according to the category of words adjacent to any second entity noun to obtain second extraction information.

[0097] Figure 8 A flowchart showing the steps of concatenating the second entity noun according to an exemplary embodiment of the present invention is shown. It should be understood that Figure 8 The manner of concatenating the second entity nouns shown is only an example, and the present invention is not limited thereto, and the second entity nouns may also be concatenated in other manners.

[0098] Reference Figure 8 In step S311, the part of the acquired text data other than the extracted at least one first entity noun and at least one second entity noun is segmented to obtain at least one segmented word.

[0099] In step S322, it is determined whether the category of a word adjacent to any second entity noun is another second entity noun.

[0100] If it is determined that the category of the word adjacent to any second entity noun is another second entity noun, step S333 is executed: any second entity noun is concatenated with another second entity noun to obtain second extraction information.

[0101] If it is determined that the category of the word adjacent to any second entity noun is not another second entity noun, that is, it is determined that the category of the adjacent word is a participle, step S344 is executed: determining the part of speech of the participle.

[0102] In step S355, it is determined whether the part of speech of the participle is a verb.

[0103] If it is determined that the part of speech of the participle is a verb, the relative position of the participle and any second entity noun is determined. If the participle is located before any second entity noun, step S366 is executed: the participle is concatenated with any second entity noun to obtain second extraction information.

[0104] If it is determined that the part of speech of the participle is not a verb, or it is determined that the participle is located after any second entity noun, the second entity noun is not concatenated. At this time, at least one second entity noun can be determined as the second extracted information.

[0105] With the above Figure 6 The text data shown takes "Anemometer internal electronic components malfunction" as an example. For the segmentation result: "v: occur, INFO: malfunction", assuming that any second entity noun is "INFO: malfunction", since the category of the word adjacent to it is a participle ("v: occur"), and the participle is a verb, the two can be concatenated, and the final output second extracted information is INFO: malfunction.

[0106] Here, when there are two words adjacent to any entity noun, the above steps can be used to determine whether to concatenate any entity noun for each adjacent word.

[0107] In a preferred example, the wind power fault information extraction method according to an exemplary embodiment of the present invention, in addition to extracting the above-mentioned first extraction information and second extraction information from text data, can also extract at least one of the following items from the acquired text data: event type, name of the faulty wind turbine set, name of the faulty system, cause of the fault, fault handling method, and fault handling result.

[0108] That is to say, based on the wind power fault information extraction method of the exemplary embodiment of the present invention, the text data is converted into structured data.

[0109] In one example, assume that the text data is: "Unit 6# reported a high temperature warning for the 1# variable pitch cabinet. Inspection found that one of the AC2 cooling fans had a blade missing, causing the temperature of the variable pitch cabinet to be high. After on-site replacement, the unit resumed normal operation."

[0110] As shown in Table 5 below, through information extraction, the unstructured data is converted into structured data, and the extracted information is stored as follows:

[0111] Table 5

[0112] Event Type Fault Unit Name Unit 6# system Pitch system part 1# pitch cabinet Fault Information High temperature warning reason One of the AC2 cooling fans lost a blade, causing the temperature of the pitch cabinet to be high. measure Field Replacement result The unit resumed normal operation

[0113] Fig. 9 A schematic diagram comparing extraction results based on the NER algorithm for different text data according to an exemplary embodiment of the present invention and extraction results based on the wind power fault information extraction method (CFG+NER algorithm) according to an exemplary embodiment of the present invention is shown.

[0114] As shown in Table 6 above, it can be seen that after using the CRF+NER algorithm, the industrial information can be accurately integrated together to form an independent PA information or INFO information, for example, Chinese phrases + component models (such as Beckhoff module KL3404, sensor N95), description information of specific components (yaw relay, pressure relay), etc.

[0115] Fig.10 A block diagram of a wind power fault information extraction device according to an exemplary embodiment of the present invention is shown.

[0116] like Fig.10 As shown, a wind power fault information extraction device 100 according to an exemplary embodiment of the present invention includes: a text data acquisition module 101, a word extraction processing module 102 and an extracted information determination module 103.

[0117] Specifically, the text data acquisition module 101 acquires text data about wind power faults.

[0118] For example, the text data acquisition module 101 may acquire text data about wind power failures from a corpus of the wind power industry. As an example, the corpus may be formed based on fault work orders, technical contracts, business contracts, etc. of the wind power industry.

[0119] The word extraction processing module 102 performs entity noun extraction processing on the acquired text data.

[0120] In an optional example, the word extraction processing module 102 can extract entity nouns from the acquired text data based on a named entity recognition (NER) algorithm, that is, output of entity words is obtained through the NER algorithm. In a preferred example, the BERT algorithm can be used to implement the NER function.

[0121] As an example, the entity nouns obtained through the extraction process may include at least one first entity noun for indicating a fault point and at least one second entity noun for fault description.

[0122] The extraction information determination module 103 determines the first extraction information and the second extraction information of the text data based on the acquired text data and the entity noun extraction processing result. Here, the first extraction information is used to characterize the fault point where the fault occurs, and the second extraction information is used to characterize the fault-related description for the fault point.

[0123] For example, the extraction information determination module 103 can obtain the first extraction information in one of the following ways: concatenating two first entity nouns that are adjacent to each other in the text data to obtain the first extraction information; concatenating two first entity nouns that satisfy the noun-conjunction-noun phrase structure and the conjunction between the two first entity nouns to obtain the first extraction information.

[0124] The following describes a process in which the extracted information determination module 103 obtains first extracted information based on at least one first entity noun.

[0125] For example, the extraction information determination module 103 can determine the words adjacent to any first entity noun in at least one first entity noun in the acquired text data, determine the category of the words adjacent to any first entity noun, splice any first entity noun based on the category of the words adjacent to any first entity noun, and obtain the first extraction information.

[0126] In an exemplary embodiment of the present invention, based on the text features of wind power faults, it can be determined that nouns are usually related to fault points. Based on this, the extracted information determination module 103 can splice any first entity noun according to the category of the words adjacent to any first entity noun to obtain the first extracted information.

[0127] In one example, the extraction information determination module 103 may concatenate the first entity nouns in the following manner: segment the part of the acquired text data except for the extracted at least one first entity noun and at least one second entity noun to obtain at least one segmented word, and determine whether the category of the word adjacent to any first entity noun is another first entity noun. If it is determined that the category of the word adjacent to any first entity noun is another first entity noun, concatenate any first entity noun with the other first entity noun to obtain the first extraction information.

[0128] If it is determined that the category of the word adjacent to any first entity noun is not another first entity noun, that is, it is determined that the category of the adjacent word is a participle, then the part of speech of the participle is determined. Determine whether the part of speech of the participle is a conjunction. If the part of speech of the participle is determined to be a conjunction, then determine whether the other word adjacent to the participle is another first entity noun or a noun. If the other word adjacent to the participle is another first entity noun or a noun, any first entity noun is spliced ​​with the participle and the other word adjacent to the participle to obtain the first extraction information. If it is determined that the part of speech of the participle is not a conjunction, or it is determined that the other word adjacent to the participle is not another first entity noun or a noun, the first entity noun is not spliced. At this time, at least one first entity noun used to indicate a fault point can be determined as the first extraction information.

[0129] For example, the extraction information determination module 103 can obtain the second extraction information in one of the following ways: concatenating two second entity nouns that are adjacent to each other in the text data to obtain the second extraction information; concatenating the second entity noun that satisfies the verb-entity noun phrase structure and the verb that precedes the second entity noun to obtain the second extraction information.

[0130] The following describes a process in which the extracted information determination module 103 obtains the second extracted information based on at least one second entity noun.

[0131] For example, the extraction information determination module 103 can determine the words adjacent to any second entity noun in at least one second entity noun in the acquired text data, determine the category of the words adjacent to any second entity noun, splice any second entity noun based on the category of the words adjacent to any second entity noun, and obtain second extraction information.

[0132] In an exemplary embodiment of the present invention, based on the text features of wind power faults, it can be determined that verbs are usually related to fault triggering. Based on this, any second entity noun is concatenated according to the category of the words adjacent to any second entity noun to obtain second extraction information.

[0133] In one example, the extracted information determination module 103 can concatenate the second entity nouns in the following manner: segment the part of the acquired text data except for at least one extracted first entity noun and at least one second entity noun, obtain at least one participle, and determine whether the category of the word adjacent to any second entity noun is another second entity noun.

[0134] If it is determined that the category of the word adjacent to any second entity noun is another second entity noun, then any second entity noun is concatenated with another second entity noun to obtain second extraction information. If it is determined that the category of the word adjacent to any second entity noun is not another second entity noun, that is, it is determined that the category of the adjacent word is a participle, then the part of speech of the participle is determined. It is determined whether the part of speech of the participle is a verb. If it is determined that the part of speech of the participle is a verb, then the relative position of the participle and any second entity noun is determined.

[0135] If the participle is located before any second entity noun, the participle is concatenated with any second entity noun to obtain the second extracted information. If it is determined that the part of speech of the participle is not a verb, or it is determined that the participle is located after any second entity noun, the second entity noun is not concatenated, and at this time, at least one second entity noun can be determined as the second extracted information.

[0136] In a preferred example, in addition to extracting the above-mentioned first extracted information and second extracted information from the text data, the extracted information determination module 103 may also extract at least one of the following items from the acquired text data: event type, name of the faulty wind turbine, name of the faulty system, cause of the fault, fault handling method, and fault handling result.

[0137] Fig.11 A block diagram of a controller according to an exemplary embodiment of the present invention is shown.

[0138] like Fig.11 As shown, the controller 200 according to the exemplary embodiment of the present invention includes: a processor 201 and a memory 202 .

[0139] Specifically, the memory 202 is used to store a computer program, and when the computer program is executed by the processor 201 , the computer program implements the above-mentioned method for extracting wind power fault information.

[0140] here, Figure 1 The wind power fault information extraction method shown in Fig.11 The processor 201 is shown. That is, Fig.10 Each module shown can be implemented by a general hardware processor such as a digital signal processor, a field programmable gate array, etc., can also be implemented by a dedicated hardware processor such as a dedicated chip, or can be completely implemented in software through a computer program. For example, it can be implemented as Fig.11 The various modules in the processor 201 shown in FIG.

[0141] According to an exemplary embodiment of the present invention, there is also provided a computer-readable storage medium storing a computer program. The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to execute the above-mentioned wind power fault information extraction method. The computer-readable recording medium is any data storage device that can store data read out by a computer system. Examples of computer-readable recording media include: read-only memory, random access memory, read-only optical disk, magnetic tape, floppy disk, optical data storage device, and carrier wave (such as data transmission through the Internet via a wired or wireless transmission path).

[0142] In an exemplary embodiment of the present invention, a method and device for extracting fault information based on NLP technology in the wind power industry are proposed.

[0143] According to the wind power fault information extraction method and device of the exemplary embodiment of the present invention, wind power fault information can be automatically extracted from text data, breaking through various troubles caused by manual constraints, not only reducing costs and increasing efficiency, but also accurately mining more valuable information.

[0144] According to the wind power fault information extraction method and device of the exemplary embodiment of the present invention, firstly, a corpus of text data is collected, then an information extraction algorithm is trained in the collected corpus, and finally, the accuracy of the information extraction algorithm is further improved according to the information text characteristics of the wind power industry.

[0145] In addition, the wind power fault information extraction method and device according to the exemplary embodiment of the present invention can achieve the data management purpose of the fault work order and map the expressions of different descriptions into a unified knowledge base.

[0146] In addition, an expert question-and-answer system for the wind power industry can also be constructed based on the above-mentioned wind power fault information extraction method and device to reduce fault diagnosis errors caused by lack of experience and improve work efficiency;

[0147] In addition, the method and device for extracting wind power fault information according to the exemplary embodiments of the present invention effectively improve the text data mining effect of the wind power industry and can also mine the wealth of expert experience.

[0148] While the invention has been particularly shown and described with reference to exemplary embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the claims.

Claims

1. A method for extracting wind power fault information, characterized in that: The wind power fault information extraction method comprises: Get text data about wind power failures; Performing entity noun extraction processing on the acquired text data, wherein the entity nouns obtained through the extraction processing include at least one first entity noun for indicating a fault point and at least one second entity noun for describing the fault; Based on the acquired text data and the entity noun extraction processing result, determining first extraction information and second extraction information of the text data, The first extracted information is used to characterize the fault point where the fault occurs, and the second extracted information is used to characterize the fault-related description for the fault point. The step of extracting entity nouns from the acquired text data includes: Based on the named entity recognition (NER) algorithm, entity nouns are extracted from the acquired text data. The named entity recognition (NER) algorithm is trained based on text data on wind power failures obtained from a corpus of the wind power industry and the self-defined parts of speech of the wind power industry that are annotated. Wherein, based on the acquired text data and the entity noun extraction processing result, the step of determining the first extraction information and the second extraction information of the text data includes: Based on a context-free grammar (CFG) algorithm, referring to a text feature of a wind power fault indicating a text expression habit about a fault point in the field of wind power, and concatenating the acquired text data based on the at least one first entity noun to obtain first extracted information; Based on the CFG algorithm, referring to the text features of wind power faults indicating the text expression habits regarding fault triggering, fault causes, fault phenomena or fault processing in the wind power field, according to the acquired text data, splicing is performed based on the at least one second entity noun to obtain second extracted information, The first extraction information is obtained by one of the following methods: concatenating two first entity nouns adjacent to each other in the text data to obtain the first extraction information; concatenating two first entity nouns satisfying a noun-conjunction-noun phrase structure and a conjunction between the two first entity nouns to obtain the first extraction information, Among them, the second extraction information is obtained by one of the following methods: concatenating two second entity nouns that are adjacent in position in the text data to obtain the second extraction information; concatenating the second entity noun that satisfies the verb-entity noun phrase structure and the verb that precedes the second entity noun to obtain the second extraction information.

2. The method for extracting wind power fault information according to claim 1, characterized in that: The step of obtaining first extracted information by concatenating the at least one first entity noun comprises: Determining a word in the text data that is adjacent to any first entity noun of the at least one first entity noun; Determining the categories of the adjacent words; Based on the categories of the adjacent words, any of the first entity nouns are concatenated to obtain first extraction information.

3. The method for extracting wind power fault information according to claim 2, characterized in that: The first extraction information is obtained by: Performing word segmentation on a portion of the text data except for the extracted at least one first entity noun and the at least one second entity noun to obtain at least one word segmentation; If it is determined that the category of the adjacent word is another first entity noun, concatenating any one of the first entity nouns with the other first entity noun to obtain first extraction information; If it is determined that the category of the adjacent word is a participle, then determining the part of speech of the participle; If it is determined that the part of speech of the participle is a conjunction, determining whether another word adjacent to the participle is another first entity noun or a noun; If another word adjacent to the participle is another first entity noun or a noun, any first entity noun is concatenated with the participle and another word adjacent to the participle to obtain first extraction information.

4. The method for extracting wind power fault information according to claim 1, characterized in that: The step of obtaining second extracted information by concatenating the at least one second entity noun comprises: determining a word in the text data that is adjacent to any second entity noun in the at least one second entity noun; Determining the categories of the adjacent words; Based on the categories of the adjacent words, any second entity noun is concatenated to obtain second extraction information.

5. The method for extracting wind power fault information according to claim 4, characterized in that: The second extraction information is obtained by: Performing word segmentation on a portion of the text data except for the extracted at least one first entity noun and the at least one second entity noun to obtain at least one word segmentation; If it is determined that the category of the adjacent word is another second entity noun, concatenating any second entity noun with the another second entity noun to obtain second extraction information; If it is determined that the category of the adjacent word is a participle, then determining the part of speech of the participle; If it is determined that the part of speech of the participle is a verb, and the participle is located before any second entity noun, the any second entity noun is concatenated with the participle to obtain second extraction information.

6. The method for extracting wind power fault information according to claim 1, characterized in that: The wind power fault information extraction method further includes: extracting at least one of the following items from the acquired text data: Event type, name of the failed wind turbine, name of the failed system, cause of the failure, fault handling method, and fault handling result.

7. A wind power fault information extraction device, characterized in that: The wind power fault information extraction device comprises: A text data acquisition module, which acquires text data about wind power faults; A word extraction processing module performs entity noun extraction processing on the acquired text data, wherein the entity nouns obtained by the extraction processing include at least one first entity noun for indicating a fault point and at least one second entity noun for describing the fault; The extraction information determination module determines the first extraction information and the second extraction information of the text data based on the acquired text data and the entity noun extraction processing result. The first extracted information is used to characterize the fault point where the fault occurs, and the second extracted information is used to characterize the fault-related description for the fault point. The word extraction processing module extracts entity nouns from the acquired text data based on the named entity recognition (NER) algorithm. The named entity recognition (NER) algorithm is trained based on text data on wind power failures obtained from a corpus of the wind power industry and the self-defined parts of speech of the wind power industry that are annotated. The extracted information determination module is based on a context-free grammar (CFG) algorithm, with reference to a text feature of a wind power failure indicating a text expression habit regarding a fault point in the wind power field, and based on the acquired text data, splices based on at least one first entity noun to obtain first extracted information; based on the CFG algorithm, with reference to a text feature of a wind power failure indicating a text expression habit regarding a fault trigger, a fault cause, a fault phenomenon, or a fault processing in the wind power field, and based on the acquired text data, splices based on at least one second entity noun to obtain second extracted information, The extraction information determination module obtains the first extraction information by one of the following methods: concatenating two first entity nouns adjacent to each other in the text data to obtain the first extraction information; concatenating two first entity nouns satisfying a noun-conjunction-noun phrase structure and a conjunction between the two first entity nouns to obtain the first extraction information; Among them, the extraction information determination module obtains the second extraction information by one of the following methods: splicing two second entity nouns that are adjacent in position in the text data to obtain the second extraction information; splicing the second entity noun that satisfies the verb-entity noun phrase structure and the verb located before the second entity noun to obtain the second extraction information.

8. A controller, characterized in that: include: processor; A memory for storing a computer program, wherein the computer program, when executed by the processor, implements the method for extracting wind power fault information as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for extracting wind power fault information as claimed in any one of claims 1 to 6 is implemented.

Images