Transformer substation alarm event extraction method and system

Abstract

The invention discloses a transformer substation alarm event extraction method and system. The method comprises the following steps: preprocessing original monitoring alarm information; calculating a word vector for the preprocessed original monitoring alarm information by using a CBOW model, converting the word vector into a sentence vector, and obtaining distributed vector representation of the monitoring alarm information with the same word vector; identifying keyword information in the sentence vector by adopting a TF-IDF text mining algorithm according to the entity type of the alarm information and the alarm action; and a BM25F algorithm is adopted, according to the keyword information, alarm information with the highest correlation with the keyword information is obtained and serves as key alarm information in the whole alarm information set, and the problems that when a power grid and equipment break down, a large number of alarms can cause a monitor to omit important alarms, and the processing time is delayed are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of substation alarms, and in particular to a method and system for extracting substation alarm events. Background Art

[0002] With the continuous advancement of large-scale power grid construction, the number of substations has increased rapidly, and traditional substations are gradually developing in the direction of high automation, digitization, and intelligence. But at the same time, independent systems will also generate a large amount of alarm information. Traditional monitoring methods are mainly based on signal monitoring of a single system, which brings huge information monitoring pressure to monitoring personnel, especially when power grids and equipment fail. A large number of alarms will cause the monitor to miss important alarms and delay processing time. Monitoring personnel rely solely on manual experience to judge the relevance of signals and possible equipment failures or anomalies, and lack the corresponding technical means to intuitively understand the impact that equipment failures or anomalies may have on the power grid.

[0003] In the prior art, the patent publication number is a method for analyzing customer satisfaction of e-commerce products based on machine learning, including obtaining e-commerce product review text, performing data preprocessing such as word segmentation and part-of-speech tagging; selecting Chinese block markers to manually annotate the word segmentation results; training the model based on the Lib-SVM tool, and then obtaining the noun Chinese block as the candidate product feature, calculating the TF-IDF value to filter the feature; constructing a sentiment dictionary, calculating the sentiment score of each feature of the product; training the word vector language model to obtain the vector representation of the product feature; based on the word vector similarity, clustering the customer satisfaction of the product features, and calculating the total score. The analysis method of the prior art is not suitable for applying substation alarm event extraction. The prior art is to extract the most frequent keywords from multiple pieces of information. However, the most frequent keywords of substation alarms may not be important alarm information. They may be due to multiple reasons, generating common key hot words, such as many alarm messages will generate red indicator lights. If the analysis method of the prior art is used, the red indicator light is the keyword. If so, there are still many alarm messages that meet the conditions, which will cause the monitor to miss important alarms. Summary of the invention

[0004] The technical problem to be solved by the present invention is to solve the problem that when a power grid or equipment fails, a large number of alarms may cause the monitor to miss important alarms, thus delaying the processing time.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] A method for extracting alarm events of a substation, comprising:

[0007] S10, pre-processing the original monitoring alarm information;

[0008] S20, using the CBOW model, calculating the word vector for the preprocessed original monitoring alarm information, and converting the word vector into a sentence vector to obtain a distributed vector representation of the monitoring alarm information with the same word vector;

[0009] S30, uses the TF-IDF text mining algorithm to identify keyword information in the sentence vector according to the entity type and alarm action of the alarm information; and uses the BM25F algorithm to obtain the alarm information with the highest correlation with the keyword information according to the keyword information, as the key alarm information in the entire alarm information set.

[0010] In one embodiment of the present invention, the original monitoring alarm information is pre-processed, including:

[0011] Import the substation names and line names derived from the original monitoring alarm information into the vocabulary and perform word segmentation as a power dictionary; use the Jieba word segmentation tool to start word segmentation and generate a time-series monitoring alarm information consisting of a series of Chinese phrases;

[0012] Establish a stop word list to remove meaningless words in the time series monitoring alarm information.

[0013] In one embodiment of the present invention, the preprocessed original monitoring alarm information is used as a training sample. The training sample is assumed to consist of the central word w currently required to be calculated and other c fields in the context (context(w), w), and the distributed vector representation of each central word w is obtained through iterative training.

[0014] In one embodiment of the present invention, the CBOW model calculates the word vector, which is obtained by the following formula:

[0015]

[0016] In the formula, x w is the center word vector, x i is the word vector of the context.

[0017] In one embodiment of the present invention, the distributed vector representation of monitoring alarm information with the same word vector is obtained by the following formula:

[0018]

[0019] Where d represents a monitoring alarm message; word_num represents the number of words in d; q represents the words in the monitoring alarm message; vec(t) represents the vector of t; vec_sum(d) represents the distributed vector representation of the monitoring alarm message.

[0020] In one embodiment of the present invention, keyword information is identified in a sentence vector by the following formula:

[0021] TF-IDF(t,d)=TF(t,d)·IDF(t);

[0022] Where t represents a keyword, d represents a monitoring alarm message, TF(t,d) represents the frequency of keyword t appearing in alarm message d, IDF(t) represents the inverse document frequency; TF-IDF(t,d) represents the importance value of a keyword.

[0023] In one embodiment of the present invention, the monitoring alarm information with the highest correlation with the keyword information is obtained according to the keyword information through the following formula:

[0024]

[0025] In the formula, Q i represents the weight of field i, c represents the number of fields, k i , b i is the control factor, and k is taken i The value is 2, b i The value is 0.75, L i It represents the sum of the lengths of all monitoring alarm information containing field i, l d,i Indicates the length of field i in monitoring alarm information d, f d,t,i is the frequency of keyword t appearing in field i in monitoring alarm information d, and score(T,d) is the relevance score of monitoring alarm information d in the alarm information set.

[0026] In one embodiment of the present invention, the TF-IDF(t,d) values ​​of the keywords in the monitoring alarm information are combined to find the top j keywords with the highest TF-IDF(t,d) values; the relevance scores of the top j keywords are calculated; the monitoring alarm information with the highest relevance is screened out, extracted, and judged.

[0027] The present invention also provides a system using the above-mentioned substation alarm event extraction method, comprising:

[0028] Preprocessing module, preprocessing the original monitoring alarm information;

[0029] The distributed vector representation module uses the CBOW model to calculate the word vectors for the preprocessed original monitoring alarm information, and converts the word vectors into sentence vectors to obtain the distributed vector representation of the monitoring alarm information with the same word vectors;

[0030] The extraction module uses the TF-IDF text mining algorithm to identify keyword information in the sentence vector according to the entity type and alarm action of the alarm information; and uses the BM25F algorithm to obtain the alarm information with the highest correlation with the keyword information based on the keyword information as the key alarm information in the entire alarm information set.

[0031] In one embodiment of the present invention, keyword information is identified in a sentence vector by the following formula:

[0032] TF-IDF(t,d)=TF(t,d)·IDF(t);

[0033] In the formula, t represents a keyword, d represents a monitoring alarm message, TF(t,d) represents the frequency of keyword t appearing in alarm message d, IDF(t) represents the inverse document frequency; TF-IDF(t,d) represents the importance value of a keyword;

[0034] The following formula is used to obtain the alarm information with the highest correlation with the keyword information based on the keyword information:

[0035]

[0036] In the formula, Q i represents the weight of field i, c represents the number of fields, k i , b i is the control factor, and k is taken i The value is 2, b i The value is 0.75, L i It represents the sum of the lengths of all monitoring alarm information containing field i, l d,i Indicates the length of field i in monitoring alarm information d, f d,t,i is the frequency of keyword t appearing in field i in monitoring alarm information d, and score(T,d) is the relevance score of monitoring alarm information d in the alarm information set.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] When preprocessing the original monitoring alarm information, the substation name and line name are exported to generate time-series monitoring alarm information, so that the alarm information is linked to the substation name and line name. Then, through the CBOW model, the monitoring alarm information with the same word vector is distributedly represented, and the alarm information is linked to the specific substation name and line name alarm information.

[0039] Based on the sentence vector, the alarm information is identified through the TF-IDF text mining algorithm to identify keywords. At this time, the word vectors with the same words have the same keywords and the same attributes, and the word vectors with other attributes are screened out.

[0040] Then, the BM25F algorithm is used to calculate the keywords, and the monitoring alarm information with the highest degree of relevance to the keywords is screened out from the vector distribution data with the same attributes. Secondary screening is performed, and the information is extracted for judgment.

[0041] The present invention can quickly and effectively classify and handle complex faults, multiple faults and large-scale faults caused by severe weather conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The present invention is a flowchart of a method for extracting substation alarm events according to an embodiment of the present invention.

[0043] Figure 2 The present invention is a block diagram of a method for extracting substation alarm events according to an embodiment of the present invention.

[0044] Figure 3 The present invention is a block diagram of a substation alarm event extraction system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0046] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0047] See also Figure 1 and Figure 2 As shown, the present invention provides a method for extracting substation alarm events, comprising:

[0048] S10, pre-processing the original monitoring alarm information.

[0049] In one embodiment of the present invention, the pre-processing process includes:

[0050] Word segmentation: import the substation name and line name derived from the original monitoring alarm information into the vocabulary and perform word segmentation as a power dictionary. Use the Jieba word segmentation tool to start word segmentation and generate a time-series monitoring alarm information consisting of a series of Chinese phrases.

[0051] Filter stop words, establish a stop word list, remove meaningless words in the time series monitoring alarm information, achieve data cleaning, and improve the training effect.

[0052] S20, using the CBOW model, calculates the word vector for the preprocessed original monitoring alarm information, and converts the word vector into a sentence vector to obtain a distributed vector representation of the monitoring alarm information with the same word vector.

[0053] In one embodiment of the present invention, the preprocessed original monitoring alarm information is used as a training sample. The training sample is assumed to consist of the central word w currently required to be calculated and other c fields in the context (context(w), w), and the distributed vector representation of each central word w is obtained through iterative training.

[0054] Among them, the CBOW model calculates the word vector and obtains it through the following formula:

[0055]

[0056] In the formula, x w is the center word vector, x i is the word vector of the context.

[0057] In this embodiment, when an alarm event occurs, the monitoring alarm information is represented in the form of a sentence, where a sentence may contain one or more features of the event. The vectors of all words in the monitoring alarm information are averaged to obtain a distributed vector representation of the monitoring alarm information with the same word vector. Specifically, the distributed vector representation of the monitoring alarm information with the same word vector is obtained by the following formula:

[0058]

[0059] Where d represents a monitoring alarm message; word_num represents the number of words in d; q represents the words in the monitoring alarm message; vec(t) represents the vector of t; vec_sum(d) represents the distributed vector representation of the monitoring alarm message.

[0060] S30, uses the TF-IDF text mining algorithm to identify keyword information in the sentence vector according to the entity type and alarm action of the alarm information; and uses the BM25F algorithm to obtain the alarm information with the highest correlation with the keyword information according to the keyword information, as the key alarm information in the entire alarm information set.

[0061] TF-IDF (Term Frequency-inverse Document Frequency) stands for Term Frequency-Inverse Document Frequency, and is a weighting scheme commonly used in information retrieval and text mining. TF-IDF is a method for evaluating the importance of a term to a document or corpus, which considers both the frequency of a term in a document (Term Frequency, TF) and the frequency of the term in the entire corpus (Inverse Document Frequency, IDF). The idea behind TF-IDF is that terms that are rare and specific to a document or a subset of documents are more discriminative than common terms that are widely used in the entire corpus. TF-IDF can be used as a statistical measure of term importance.

[0062] In one embodiment of the present invention, the TF-IDF text mining algorithm is used to identify keyword information in the alarm information set according to the entity type and alarm action of the alarm information. Then, the BM25F algorithm is used to obtain the alarm information with the highest correlation with the keyword according to the keyword information as the key alarm information in the entire alarm information set.

[0063] In this embodiment, the TF-IDF algorithm is used to identify keyword information in the alarm information set as the keyword representing the document. The importance of a keyword is not only proportional to its frequency of occurrence in the document, but also inversely proportional to the number of documents containing it. Specifically, the keyword information is identified in the sentence vector using the following formula:

[0064] TF-IDF(t,d)=TF(t,d)·IDF(t);

[0065] In the formula, t represents a keyword, d represents a monitoring alarm message, TF(t,d) represents the frequency of keyword t appearing in alarm message d, IDF(t) represents the inverse document frequency, and the larger the IDF, the lower the importance of keyword t in the entire corpus; TF-IDF(t,d) represents the importance value of a keyword. The higher the TF-IDF(t,d) value, the more important the term is in the current monitoring alarm message and it is relatively rare in the entire corpus, and has better distinguishing ability. By finding the words with the highest TF-DIF(t,d) values, they are output as alarm message keywords.

[0066] In one embodiment of the present invention, the BM25F algorithm is an improved algorithm of the BM25 model. In the BM25 algorithm relevance retrieval model, the text is considered as a whole when calculating the text relevance. However, with the rapid development of retrieval technology, structured data has gradually replaced text data, and each text has been cut into multiple independent domains, for example, a web page is cut into domains such as title, keyword, and content. The content in different fields has different contributions to the topic and different weights, and the scores of each word in each field are weighted and summed.

[0067] Using the TF-IDF algorithm alone can only take into account the frequency of a certain word in the alarm information, but cannot obtain the complete information that the alarm information wants to convey. Therefore, the BM25F algorithm is used in conjunction. By first using the TF-IDF algorithm to search for keywords, and then using the BM25F algorithm to match the most relevant alarm information through keywords, the alarm information extracted in this way can more effectively convey the core information of all alarm information, helping staff quickly locate equipment failures or abnormal information.

[0068] In this embodiment, the BM25F algorithm is used to calculate the relevance score of the monitoring alarm information d in the alarm information set. In the BM25F algorithm, each monitoring alarm information is regarded as a combination of multiple fields. Let D be the alarm information set, specifically the original monitoring alarm information after preprocessing, T be the keyword set, d∈D be one of the alarm information, t∈T be one of the keywords, then obtain the monitoring alarm information with the highest relevance to the keyword information, the formula is as follows:

[0069]

[0070] In the formula, Q i represents the weight of field i, c represents the number of fields, k i , b i is the control factor, and k is taken i The value is 2, b i The value is 0.75, L i It represents the sum of the lengths of all monitoring alarm information containing field i, l d,i Indicates the length of field i in monitoring alarm information d, f d,t,i is the frequency of keyword t appearing in field i in monitoring alarm information d, and score(T,d) is the relevance score of monitoring alarm information d in the alarm information set.

[0071] In this embodiment, the top j keywords with the highest TF-IDF(t, d) values ​​are searched in combination with the TF-IDF(t, d) values ​​of the keywords in the monitoring alarm information. The relevance scores of the top j keywords are calculated; the monitoring alarm information with the highest relevance is screened out and extracted for judgment. The top few keywords are manually set. The monitoring alarm information with the highest relevance is extracted to complete the alarm information extraction task, and then manually interpreted.

[0072] See also Figure 3 As shown, the present invention also provides a system applying the above-mentioned substation alarm event extraction method, comprising:

[0073] The preprocessing module preprocesses the original monitoring alarm information.

[0074] The distributed vector representation module uses the CBOW model to calculate the word vectors for the preprocessed original monitoring alarm information, and converts the word vectors into sentence vectors to obtain the distributed vector representation of the monitoring alarm information with the same word vectors.

[0075] The extraction module uses the TF-IDF text mining algorithm to identify keyword information in the sentence vector according to the entity type and alarm action of the alarm information; and uses the BM25F algorithm to obtain the alarm information with the highest correlation with the keyword information based on the keyword information as the key alarm information in the entire alarm information set.

[0076] In one embodiment of the present invention, keyword information is identified in a sentence vector by the following formula:

[0077] TF-IDF(t,d)=TF(t,d)·IDF(t);

[0078] In the formula, t represents a keyword, d represents a monitoring alarm message, TF(t,d) represents the frequency of keyword t appearing in alarm message d, IDF(t) represents the inverse document frequency; TF-IDF(t,d) represents the importance value of a keyword;

[0079] The following formula is used to obtain the alarm information with the highest correlation with the keyword information based on the keyword information:

[0080]

[0081] In the formula, Q i represents the weight of field i, c represents the number of fields, k i , b i is the control factor, and k is taken i The value is 2, b i The value is 0.75, L iIt represents the sum of the lengths of all monitoring alarm information containing field i, l d,i Indicates the length of field i in monitoring alarm information d, f d,t,i is the frequency of keyword t appearing in field i in monitoring alarm information d, and score(T,d) is the relevance score of monitoring alarm information d in the alarm information set.

[0082] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0083] The above-described embodiments merely represent implementation methods of the invention. The protection scope of the present invention is not limited to the above-described embodiments. For those skilled in the art, several modifications and improvements may be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for extracting substation alarm events, characterized in that: include: S10, pre-processing the original monitoring alarm information; S20, using the CBOW model, calculating the word vector for the preprocessed original monitoring alarm information, and converting the word vector into a sentence vector to obtain a distributed vector representation of the monitoring alarm information with the same word vector; S30, uses the TF-IDF text mining algorithm to identify keyword information in the sentence vector according to the entity type and alarm action of the alarm information; and uses the BM25F algorithm to obtain the alarm information with the highest correlation with the keyword information according to the keyword information, as the key alarm information in the entire alarm information set.

2. The method for extracting substation alarm events according to claim 1, characterized in that: Pre-process the original monitoring alarm information, including: Import the substation names and line names derived from the original monitoring alarm information into the vocabulary and perform word segmentation as a power dictionary; use the Jieba word segmentation tool to start word segmentation and generate a time-series monitoring alarm information consisting of a series of Chinese phrases; Establish a stop word list to remove meaningless words in the time series monitoring alarm information.

3. The method for extracting substation alarm events according to claim 1, characterized in that: The preprocessed original monitoring alarm information is used as the training sample. The training sample is assumed to consist of the central word w currently required to be calculated and other c fields in the context (context(w), w). The distributed vector representation of each central word w is obtained through iterative training.

4. The method for extracting substation alarm events according to claim 3, characterized in that: The CBOW model calculates word vectors and obtains them through the following formula: In the formula, x w is the center word vector, x i is the word vector of the context.

5. The method for extracting substation alarm events according to claim 4, characterized in that: Get the distributed vector representation of monitoring alarm information with the same word vector through the following formula: Where d represents a monitoring alarm message; word_num represents the number of words in d; q represents the words in the monitoring alarm message; vec(t) represents the vector of t; vec_sum(d) represents the distributed vector representation of the monitoring alarm message.

6. The method for extracting substation alarm events according to claim 1, characterized in that: The following formula is used to identify keyword information in the sentence vector: TF-IDF(t,d)=TF(t,d)·IDF(t); Where t represents a keyword, d represents a monitoring alarm message, TF(t,d) represents the frequency of keyword t appearing in alarm message d, IDF(t) represents the inverse document frequency; TF-IDF(t,d) represents the importance value of a keyword.

7. The method for extracting substation alarm events according to claim 6, characterized in that: The following formula is used to obtain the monitoring alarm information with the highest correlation with the keyword information based on the keyword information: In the formula, Q i represents the weight of field i, c represents the number of fields, k i , b i is the control factor, and k is taken i The value is 2, b i The value is 0.75, L i It represents the sum of the lengths of all monitoring alarm information containing field i, l d,i Indicates the length of field i in monitoring alarm information d, f d,t,i is the frequency of keyword t appearing in field i in monitoring alarm information d, and score(T,d) is the relevance score of monitoring alarm information d in the alarm information set.

8. The method for extracting substation alarm events according to claim 7, characterized in that: Combined with the TF-IDF(t,d) values ​​of the keywords in the monitoring alarm information, find the top j keywords with the TF-IDF(t,d) values; calculate the relevance scores of the top j keywords; filter out the monitoring alarm information with the highest relevance, extract it, and make a judgment.

9. A system using the substation alarm event extraction method according to any one of claims 1 to 8, characterized in that: include: Preprocessing module, preprocessing the original monitoring alarm information; The distributed vector representation module uses the CBOW model to calculate the word vectors for the preprocessed original monitoring alarm information, and converts the word vectors into sentence vectors to obtain the distributed vector representation of the monitoring alarm information with the same word vectors; The extraction module uses the TF-IDF text mining algorithm to identify keyword information in the sentence vector according to the entity type and alarm action of the alarm information; and uses the BM25F algorithm to obtain the alarm information with the highest correlation with the keyword information based on the keyword information as the key alarm information in the entire alarm information set.

10. A substation alarm event extraction system according to claim 9, characterized in that: The following formula is used to identify keyword information in the sentence vector: TF-IDF(t,d)=TF(t,d)·IDF(t); In the formula, t represents a keyword, d represents a monitoring alarm message, TF(t,d) represents the frequency of keyword t appearing in alarm message d, IDF(t) represents the inverse document frequency; TF-IDF(t,d) represents the importance value of a keyword; The following formula is used to obtain the alarm information with the highest correlation with the keyword information based on the keyword information: In the formula, Q i represents the weight of field i, c represents the number of fields, k i , b i is the control factor, and k is taken i The value is 2, b i The value is 0.75, L i It represents the sum of the lengths of all monitoring alarm information containing field i, l d,i Indicates the length of field i in monitoring alarm information d, f d,t,i is the frequency of keyword t appearing in field i in monitoring alarm information d, and score(T,d) is the relevance score of monitoring alarm information d in the alarm information set.

Images