Configuration instantiation method for smart substation

Abstract

The present invention relates to the field of digital data processing, and the present invention relates to a method for instantiating a configuration of a smart substation, wherein a first vector representing a text is obtained by obtaining a model based on word granularity and a word granularity model based on core words that describe entity text, and then a second vector representing a text is obtained by obtaining a word granularity model based on keywords and a word granularity model based on keywords, and the vector weight is determined by calculating information gain and using a regression model, thereby finally realizing the instantiation and configuration of a configuration model of a smart substation. The present invention can accurately match configuration information points with information to be matched through multi-level vector matching and fine weight allocation, thereby improving the configuration accuracy and efficiency of substation equipment and signals, and providing strong support for the acceptance of smart substations.

Description

Technical Field

[0001] The present invention relates to the field of digital data processing, and in particular to a configuration instantiation method for an intelligent substation. Background Art

[0002] Smart substation configuration model instantiation refers to the automatic and intelligent configuration and instantiation of equipment, signals and configuration parameters in the smart substation according to the predefined configuration model. Through this process, the rapid configuration and accurate mapping of substation equipment and signals can be achieved, and the automation and intelligence level of the substation can be improved. The development of this technology helps to improve the operating efficiency and reliability of the power system, reduce the workload and error rate of manual configuration, and play a key role in the acceptance of large substations. In the process of smart substation configuration model instantiation, there are some technical problems that need to be solved. For example, the definitions of equipment and signals in the configuration model are relatively complex, and a large number of text descriptions need to be accurately parsed and matched. Traditional methods usually rely on manual operations, which are inefficient and prone to errors. In addition, the differences between different manufacturers and standards make the mapping process of signals and equipment more complicated, making it difficult to ensure the accuracy and consistency of the instantiation process. These problems are particularly prominent in large-scale applications, and an efficient and reliable automation solution is urgently needed.

[0003] The prior art has some shortcomings and deficiencies when dealing with the instantiation problem of the configuration model of the smart substation. Traditional methods mostly rely on manual configuration, which is cumbersome and error-prone, and is difficult to adapt to the needs of large-scale applications. Although some automated tools can improve efficiency, they often show limitations when dealing with complex text parsing and semantic matching. For example, a Chinese patent document with announcement number CN108600317B discloses a dispatching master station and a method for interaction between the dispatching master station and the substation, which realizes preliminary automated configuration by parsing the engineering configuration file, but there are still some technical defects, such as inaccurate port matching and low matching efficiency. In addition, the prior art lacks flexibility when dealing with standardized and non-standardized signal mappings, and is difficult to adapt to diverse application scenarios. Summary of the invention

[0004] In view of the above-mentioned problems such as inaccurate semantic matching and insufficient model generalization ability, the present invention proposes a method for instantiating intelligent substation configuration, including: obtaining a corpus, and obtaining a word granularity vector model based on the corpus; obtaining a representation vector of an entity text based on the word granularity vector model; the corpus is all text information contained in the substation configuration project; obtaining a dictionary containing proprietary vocabulary in the power field, and extracting core words in the corpus according to the dictionary, wherein the core words are proprietary vocabulary that appears more than a set number of times in the corpus; obtaining a first vector of the entity text based on the core words, wherein the first vector represents the superposition of semantic expressions of the characters and core words contained in the corpus at the word granularity level and the word granularity level; screening keywords in the corpus, wherein the keywords are related to the substation configuration project. ; Obtain keywords based on the context of the keywords and the dictionary; Obtain a second vector of the entity text based on the keywords; The second vector represents the superposition of the keywords contained in the corpus and the semantic expressions of the keywords at the character granularity level and the word granularity level; Collect historical data of the substation and divide it into multiple sub-scenarios based on the historical data; Obtain information gain based on the core words and keywords, and build a regression model through the information gain and sub-scenarios; Determine the weights of the first vector and the second vector based on the regression model; Weight the first vector and the second vector according to the weights to obtain a comprehensive vector; Obtain the comprehensive vector corresponding to the entity text of the configuration information point and the information to be matched; Calculate the similarity of the entity text of the configuration information point and the information to be matched based on the comprehensive vector to complete information mapping and matching.

[0005] The smart substation configuration instantiation method provided by the present invention obtains the character granularity vector model and the word granularity vector model describing the entity text, screens keywords and obtains the semantic vector model based on the keyword and keyword granularity, and constructs a regression model to determine the vector weight, thereby finally achieving the precise matching of the short text and the configuration information point. Compared with the prior art, the present invention can more accurately capture the semantic information of the text, improve the efficiency and accuracy of the instantiation process, and reduce manual intervention and configuration errors.

[0006] Furthermore, obtaining a word granularity vector model according to the corpus also includes: acquiring the corpus and using the skip-gram mode in the word2vec algorithm for training to obtain the word granularity vector model.

[0007] By using the skip-gram mode in the word2vec algorithm for training, the contextual relationship describing the text words can be captured more effectively and high-quality semantic expression vectors can be generated. This method can significantly improve the accuracy and expressiveness of word embedding when processing a large number of power system documents, which is helpful for subsequent semantic analysis and matching.

[0008] Furthermore, obtaining a dictionary containing specialized vocabulary in the power field also includes: using natural language processing technology to collect and process a large number of power system documents, technical reports, academic papers and online forum discussions; using machine learning algorithms to automatically extract and use clustering algorithms to cluster frequently occurring professional terms and key phrases, and finally generating a dictionary containing these vocabulary.

[0009] The dictionary with custom vocabulary generated by natural language processing technology and machine learning algorithm can automatically extract and cluster frequently occurring professional terms and key phrases to build a high-quality professional dictionary. This dictionary can better meet the specific needs of the power system field, improve the accuracy of vocabulary matching, and reduce semantic ambiguity.

[0010] Furthermore, a machine learning algorithm is used to automatically extract and a clustering algorithm is used to cluster frequently occurring professional terms and key phrases, which also includes: the machine learning algorithm is a BERT algorithm; and the clustering algorithm is a DBSCAN algorithm.

[0011] Furthermore, obtaining a first vector of the entity text according to the core word also includes: obtaining a core word granularity vector model according to the core word; obtaining a representation vector of the entity text based on the core word granularity vector model; connecting the representation vector of the entity text based on the character granularity vector model and the representation vector based on the core word granularity vector model to obtain a first vector;

[0012] Furthermore, screening the keywords in the corpus also includes: obtaining the corpus, using the TF-IDF algorithm to measure the importance of each word in the corpus; sorting all the words from high to low according to the TF-IDF value, and taking the top words. Generate keyword set In the present invention Get 50 experience points.

[0013] By using the TF-IDF algorithm to measure the importance of words in the corpus and filter out high-frequency and important keyword sets, the expressiveness of the semantic vector model can be significantly improved. The filtered keywords can more accurately represent the core content of the text, which is helpful for subsequent semantic vector generation and similarity calculation.

[0014] Furthermore, obtaining keywords based on the context of the keywords and the dictionary also includes: using context window technology to search for phrases in the context of the keywords as alternative keywords; matching the alternative keywords with the phrases in the dictionary, and if the same phrase also exists in the dictionary, then using the alternative keywords as keywords.

[0015] Furthermore, obtaining a second vector of the entity text according to the keyword also includes: obtaining a keyword granularity vector model according to the keyword; obtaining a keyword granularity vector model according to the keyword; using a PCA principal component analysis algorithm to integrate the entity text based on the keyword granularity vector model and the representation vector obtained based on the keyword granularity vector model to obtain a second vector.

[0016] Furthermore, determining the weights of the first vector and the second vector according to the information gain also includes: constructing a linear regression model based on the information gain and the sub-scenario; and determining the weights of the first vector and the second vector according to the linear regression model.

[0017] The linear regression model is adopted and the mean square error index is used to evaluate the prediction accuracy of the model, which can effectively determine the weights of the first vector and the second vector. Through the optimization of the regression model, the applicability and stability of the model in different sub-scenarios are ensured, and the accuracy and reliability of the instantiation process are improved.

[0018] Furthermore, calculating the entity text similarity between the configuration information point and the information to be matched based on the comprehensive vector also includes: the similarity of the entity text vocabulary is cosine similarity.

[0019] The technical effects of the present invention are:

[0020] The present invention can accurately match configuration information points with information to be matched through multi-level vector matching and fine weight allocation, thereby improving the configuration accuracy and efficiency of substation equipment and signals, and providing strong support for the acceptance of smart substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0022] Figure 1 is a flowchart schematically illustrating a method for instantiating a configuration of a smart substation based on image processing according to an embodiment of the present invention;

[0023] Figure 2 It is a logic block diagram schematically showing some steps of an embodiment of the present invention;

[0024] Figure 3 The figure is a schematic diagram schematically showing the signal point mapping of the standardized acceptance intermediate information base in the embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0026] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0027] Smart substation configuration instantiation method embodiment:

[0028] like Figure 1 As shown, the smart substation configuration instantiation method of the present invention includes:

[0029] S1. Obtain a corpus; obtain a representation vector of the entity text based on a word granularity vector model.

[0030] The configuration model instantiation method of smart substation aims to improve the automation and intelligence level of substation. Through the research and optimization of the configuration model, the rapid configuration and accurate mapping of substation equipment and signals are realized. In the configuration modeling instantiation method studied in the present invention, it is necessary to map the test sub-item logical criteria, applied semaphores, intermediate variable elements and signal definitions in the actual engineering SCD file, and rely on the primary and secondary associations of the graph model configuration and the signal configuration to quickly instantiate the acceptance test sub-item, forming an instantiated acceptance test case that can be used in the acceptance test process of the relay protection equipment.

[0031] After completing the configuration of the intermediate information library, the signal points in the intermediate information library need to be mapped with the signal points in the actual project SCD file. The mapping process is the matching process of text information and reference addresses, which can also be understood as semantic similarity matching, such as Figure 3 As shown, in the present invention, the matching process of the above text information and reference address is described as the matching process of the entity text of the information to be matched and the entity text of the configuration information point.

[0032] There are two main categories of text semantic similarity calculation: knowledge-based and corpus-based. In one embodiment: the knowledge-based similarity calculation method is to quantify the degree of semantic association between two texts by using the information obtained from the knowledge base. WordNet is a method for solving the text similarity calculation problem through a dictionary. With the help of the WordNet idea, the automatic mapping of information can be achieved directly by establishing a signal index and its corresponding dictionary interpretation relationship.

[0033] The rules for signal indexing in the intermediate information library can be defined based on the standardized specifications of the Southern Power Grid. The 2018 version of the General Technical Specifications for Relay Protection of the Southern Power Grid has established a standardized model for virtual terminals, signals and descriptions. Through a simple combination of morpheme expressions and a weighted algorithm for each morpheme weight, the object with the highest global similarity is calculated as the association mapping object; for non-standardized models, the unique feature information is strengthened through models, references, and descriptions, and an iterative intermediate information library index is established, which can also solve the problem of automatic signal association mapping and instantiation.

[0034] In another embodiment: all text information involved in the configuration model instantiation work project is collected to form a corpus. In the research on the configuration model instantiation work, the signal that needs to be instantiated meets the characteristics of short texts, and the word2vec semantic similarity algorithm can better realize the short text similarity calculation. This project improves the word2vec algorithm through research and adopts the skip-gram mode of word2vec for learning and training. The skip-gram mode predicts the probability of the current central word appearing in the context through the central vocabulary. The following content is an example of the training process of the skip-gram model:

[0035] The training process is to obtain the word vector of the central word and use it to predict the context words, that is, the similar background words. entity vocabulary, and set the final output embedded representation dimension of each word to Dimension, for each word , Exists in sentence sequence ,enter One-hot vector of vocabulary, with central word Similar background words .

[0036] After the input layer weight matrix Projection to hidden layer , the input layer weight matrix The size is , thus the hidden layer obtains dimensional center word input word vector The hidden layer to the output layer needs to pass through the output layer weight matrix , the size of the output layer weight matrix is , output through the softmax function, and normalize the probability of each element being a similar background word to obtain the word representation vector of the similar background word , the normalization process is as follows:

[0037]

[0038] in Indicates the central word A collection of; Indicates the central word Corresponding similar background words A collection of; Indicates that given the central word In the case of The conditional probability of Word representation vectors representing similar background words; Represents the center word input word vector; The word representation vectors representing all entity vocabulary; the symbol Represents the transposed matrix. The smaller the difference, the greater the cross entropy. In order to make the prediction more accurate, it is necessary to maximize , loss function as follows:

[0039]

[0040] in Indicates that given the central word In the case of The joint probability of simultaneous occurrence is is the last background word in the background word set; the input layer weight matrix is ​​continuously updated through softmax gradient descent and the output layer weight matrix , to maximize the maximum likelihood probability and minimize the loss function, and obtain the optimized semantic expression vector describing the text vocabulary, that is, In the present invention, the entity text representation vector obtained based on the word granularity word2vec model is recorded as .

[0041] By using the skip-gram mode in the word2vec algorithm for training, the contextual relationship describing the text words can be captured more effectively and high-quality semantic expression vectors can be generated. This method can significantly improve the accuracy and expressiveness of word embedding when processing a large number of power system documents, which is helpful for subsequent semantic analysis and matching.

[0042] S2. Obtain a representation vector of the entity text based on a core word vector model; obtain a first vector.

[0043] like Figure 2As shown in the dashed box S2, considering that the semantic model based on word vector has certain limitations for extracting vocabulary features in the power field, lacks certain prior knowledge and background information, there may be certain deviations in the representation of proprietary vocabulary. The present invention proposes to fuse the word representation vector of the secondary test proprietary vocabulary on the basis of word vector, so as to enrich the feature representation information describing the short text.

[0044] In one embodiment, natural language processing technology can be used to collect and analyze a large number of power system documents, technical reports, academic papers and online forum discussions, and use machine learning algorithms to automatically extract and cluster frequently occurring professional terms and key phrases, and finally generate a dictionary containing these custom words. The machine learning algorithm can use the BERT algorithm; the clustering algorithm can use the DBSCAN clustering algorithm.

[0045] The dictionary with custom vocabulary generated by natural language processing technology and machine learning algorithm can automatically extract and cluster frequently occurring professional terms and key phrases to build a high-quality professional dictionary. This dictionary can better meet the specific needs of the power system field, improve the accuracy of vocabulary matching, and reduce semantic ambiguity.

[0046] First, we use the customized secondary test vocabulary dictionary to extract the core words in the corpus and get the core word sequence ,in Indicates the first Then The text sequence is input into the word2vec model for training. Different from the model training based on word vectors, this training does not require word segmentation and is trained on word-granular text. The other training processes are consistent with the word2vec model based on word vectors. Finally, a semantic vector representation model of the word-granular word2vec model based on core words can be obtained.

[0047] The representation vector of the entity text obtained in S1 based on the word-granularity word2vec model is Here, the entity text is represented by the core word word2vec model. , the two representation vectors are connected in parallel, and the secondary device configuration short text distributed representation vector is finally obtained by combining the word vector and the core word vector. For the convenience of representation, the present invention records this vector as the first vector. The first vector mentioned anywhere in this article refers to this vector, which has:

[0048]

[0049] in represents the first vector; It is the representation vector of entity text based on the word-granularity word2vec model; The entity text is represented by the vector obtained based on the core word word2vec model; Represents the vector concatenation operator symbol.

[0050] S3, filter keywords and obtain the representation vector of the entity text based on the keyword vector model; extract keywords and obtain the representation vector of the entity text based on the keyword vector model; obtain the second vector.

[0051] like Figure 2 As shown in the dashed box S3, the range of words collected in the semantic expression model based on word vectors in S2 is relatively large, which not only increases the amount of model data, but also may include some unnecessary non-professional words. Therefore, on this basis, keywords can be screened and keywords can be extracted again based on the keywords to obtain the second vector.

[0052] S301, filter keywords and obtain representation vectors of entity texts based on a keyword vector model.

[0053] For a corpus, which contains multiple documents in a project file, the TF-IDF algorithm is used to measure the importance of a word in the corpus. First, the frequency of each word in its document, that is, the term frequency, is calculated; then the inverse of the frequency of each word in the entire corpus, that is, the inverse document frequency, is calculated; then the term frequency is multiplied by the inverse document frequency to obtain the TF-IDF value of each word. A high TF-IDF value indicates that the word has a high importance in the current document; the TF-IDF value of each word in all documents in the corpus is calculated to obtain the TF-IDF mean of each word; then all words are sorted from high to low according to the TF-IDF mean, and the top words are taken. Generate keyword set ,in The value of needs to be selected in combination with the power system scenario, taking into account the document length and the number of domain-specific terms. For longer documents, the value can be appropriately increased. Value, in the present invention Take the empirical value of 50, and implementers can adjust it according to actual conditions to ensure that the extracted keywords can fully cover the core content of the document; finally, the keyword set Input into the word2vec model for training, obtain the keyword-based word2vec model, and record the representation vector of the entity text based on the keyword-based word2vec model as .

[0054] By using the TF-IDF algorithm to measure the importance of words in the corpus and filter out high-frequency and important keyword sets, the expressiveness of the semantic vector model can be significantly improved. The filtered keywords can more accurately represent the core content of the text, which is helpful for subsequent semantic vector generation and similarity calculation.

[0055] S302: extract keywords and obtain representation vectors of entity texts based on a keyword vector model.

[0056] In S2, a custom vocabulary dictionary is created, which contains some frequently used professional terms and key phrases in the power system. However, the dictionary may still include some words that are not frequently used in actual scenarios. Therefore, the dictionary is combined with the keyword set obtained in S301. Extract the words with the highest frequency in this scenario and use them to generate keyword vectors.

[0057] First, use the context window technique to search for phrases in the context of the keywords. For example, in the keyword set The keyword "change" is included in the text containing "change". If there are phrases such as "transformer" or "frequency conversion" in the corpus, then these two phrases are used as candidate keywords. The size of the context window in the present invention is set to an empirical value of 2. The implementer can adjust it according to the actual power system-related vocabulary to make the keyword more accurate; then the candidate keywords are matched with the phrases in the secondary test proprietary vocabulary dictionary. If the two phrases also exist in the dictionary, then these two phrases are used as keywords; then the keyword set is matched All keywords in S2 are searched for as above, and all keywords are arranged in the order in which they appear to obtain a keyword sequence. Finally, the keyword sequence is input into the word2vec model as a text sequence for training. The process is the same as the core word sequence training process in S2, and will not be repeated here. The representation vector obtained by the entity text based on the keyword word2vec model is recorded as .

[0058] S303: Obtain a second vector.

[0059] In the above content, the entity text representation vectors obtained based on the keyword word granularity word2vec model are obtained respectively And the representation vector obtained based on the keyword granularity word2vec model ; When integrating the two vectors, since there is a certain correlation between the two vectors and the comprehensive dimension of the two vectors is large, the present invention uses PCA principal component analysis to integrate the above two vectors, which can effectively reduce the dimension and retain the main features. This method belongs to the known technology, and the specific process will not be repeated. The vector With vector The vector after PCA dimensionality reduction of the concatenated matrix is ​​recorded as , which is the second vector.

[0060] S4. Determine the vector weight according to the information gain and the regression model.

[0061] For the first vector obtained, it covers a wider range of vocabulary, but only relies on single words and vocabulary, and may ignore the information of context and combined phrases; for the second vector, it covers fewer vocabulary, but captures more semantic associations, and improves the ability to understand complex texts. In the present invention, the weights of the above two vectors are determined according to information gain and regression model.

[0062] S401. Calculate information gain.

[0063] First, collect the operation data in the substation system, including equipment status, operation records, and fault information. Extract relevant historical data to provide a basis for model training and analysis; then preprocess the collected data to remove noise data and ensure data quality; then divide the substation operation scenario into multiple sub-scenarios based on the operation data, such as normal operation can be accepted, minor faults need to be adjusted, and serious faults cannot be accepted; further calculate the information gain of the first vector and the second vector in each sub-scenario, where the information gain is obtained by the initial information entropy and conditional entropy, which belongs to the known technology and will not be repeated in the present invention.

[0064] In one embodiment, the core word sequence corresponding to the first vector can be As a data set, suppose there is a core word "current" in the data set, which appears 40 times in the sub-scenario "normal operation can be accepted"; 50 times in the sub-scenario "minor fault needs adjustment"; and 70 times in the sub-scenario "serious fault cannot be accepted". Then the information entropy and conditional entropy can be calculated according to the frequency of its appearance in each sub-scenario, so as to obtain the information gain of the core word "current", which is recorded as ; Assume that there are core words, and calculate the information gain of each core word, which are: , Represents the last core word of the data set, and finally the information gain of the data set The calculation process is:

[0065]

[0066] in is the total number of core words in the dataset; Indicates the data set core words; the information gain of this data set As the information gain of the first vector. Although the process of obtaining the first vector also includes the model training process based on word granularity, the amount of information covered by words is small, the data scale is large, and it has little reference significance for the judgment of sub-scenes, so here only the information gain of vocabulary is calculated.

[0067] For the keyword sequence corresponding to the second vector, the process of obtaining the information gain is the same as the process of obtaining the information gain of the core word sequence mentioned above, which will not be repeated here. The information gain corresponding to the keyword sequence is taken as the information gain of the second vector and recorded as .

[0068] S402: Determine vector weights using a regression model.

[0069] In S401, the information gain of the first vector and the second vector is obtained. and , which is used as the input data set of the regression model; the target variable is the system state, that is, the sub-scenario set. The linear regression model can be used for analysis, and the regression equation is:

[0070]

[0071] in represents the regression target variable; and represent regression coefficients respectively; and Represent the regression independent variables respectively; is the error term.

[0072] Use historical data sets for training, divide the data into training sets and test sets, 80% of the data is used for training and 20% of the data is used for testing, perform linear regression analysis on the training set, and obtain the coefficients of the regression equation and , i.e., the weights of the first vector and the second vector, and the model is verified using a test set, and the prediction accuracy of the model can be evaluated using a mean square error index. The above linear regression analysis is a well-known technique, and the specific analysis content will not be repeated here.

[0073] The linear regression model is adopted and the mean square error index is used to evaluate the prediction accuracy of the model, which can effectively determine the weights of the first vector and the second vector. Through the optimization of the regression model, the applicability and stability of the model in different sub-scenarios are ensured, and the accuracy and reliability of the instantiation process are improved.

[0074] S5. Calculate the comprehensive similarity to complete model optimization.

[0075] Based on the first vector, the second vector and the vector weight obtained above, the most widely used cosine similarity can be used to measure the comprehensive similarity of entity texts. The entity text to be matched , the similarity calculation process is as follows:

[0076]

[0077]

[0078] in represents the comprehensive vector; represents the weight of the first vector; represents the first vector; represents the second vector; represents the second vector; Entity text representing a configuration information point The corresponding comprehensive vector representation; Entity text representing the information to be matched The corresponding comprehensive vector representation. When the cosine similarity of the vectors corresponding to the two entity texts is greater, the distance is smaller. Specifically, the system will select the information to be matched with the smallest distance from the given configuration information point, which means that the entity text description of the information to be matched with the smallest distance is the most consistent or relevant configuration information description, thereby completing the mapping of the signal points in the intermediate information library and the signal points in the actual engineering SCD file.

[0079] After the above optimization and adjustment, the present invention proposes a method for instantiating a configuration model of a smart substation based on multi-level vector matching, which obtains the semantic expression vector describing the text vocabulary and the semantic vector model based on word granularity, screens keywords and obtains the semantic vector model based on keywords and keyword granularity, and constructs a regression model to determine the vector weight, and finally achieves the precise matching of short text and configuration information points. Compared with the prior art, the present invention can more accurately capture the semantic information of the text, improve the efficiency and accuracy of the instantiation process, and reduce manual intervention and configuration errors.

[0080] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0081] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for instantiating a configuration of a smart substation, characterized in that: The method comprises: Acquire a corpus, and obtain a word granularity vector model according to the corpus; obtain a representation vector of an entity text based on the word granularity vector model; the corpus is all text information contained in a substation configuration project; Obtain a dictionary containing proprietary words in the electric power field, extract core words in the corpus according to the dictionary, and the core words are proprietary words that appear more than a set number of times in the corpus; obtain a first vector of the entity text according to the core words, including: Acquire a core word granularity vector model according to the core word; Obtaining a representation vector of the entity text based on the core word granularity vector model; Connecting the representation vector of the entity text based on the character granularity vector model and the representation vector based on the core word granularity vector model to obtain a first vector; The first vector represents the superposition of semantic expressions of characters and core words contained in the corpus at the character granularity level and the word granularity level; Filtering keywords in the corpus, the keywords are related to the substation configuration project; obtaining keywords based on the context of the keywords and the dictionary; obtaining a second vector of the entity text according to the keywords; the second vector is expressed at the character granularity level and the word granularity level, based on the keywords contained in the corpus and the superposition of the semantic expressions of the keywords; Dividing a plurality of sub-scenarios according to the historical data of the substation; obtaining information gain based on the core words and the keywords, and determining the weights of the first vector and the second vector according to the information gain, including: Constructing a linear regression model based on the information gain and the sub-scenario; Determine the weights of the first vector and the second vector according to the linear regression model, and perform weighting to obtain a comprehensive vector; Obtaining comprehensive vectors corresponding to the configuration information point and the entity text of the information to be matched; and calculating the similarity between the configuration information point and the information to be matched based on the comprehensive vector to complete the matching.

2. The method for instantiating a smart substation configuration according to claim 1, characterized in that: Obtaining a word granularity vector model according to the corpus includes: A corpus is obtained and trained using the skip-gram mode in the word2vec algorithm to obtain the word granularity vector model.

3. The method for instantiating a configuration of a smart substation according to claim 1, characterized in that: Get a dictionary of terms specific to the power sector, including: Use natural language processing technology to collect and process a large number of power system documents, technical reports, academic papers and online forum discussions; Use machine learning algorithms to automatically extract and cluster frequently occurring professional terms and key phrases using clustering algorithms, and ultimately generate a dictionary containing these terms.

4. The method for instantiating a configuration of a smart substation according to claim 3, characterized in that: Use machine learning algorithms to automatically extract and cluster frequently occurring professional terms and key phrases, including: The machine learning algorithm is a BERT algorithm; The clustering algorithm is the DBSCAN algorithm.

5. The method for instantiating a configuration of a smart substation according to claim 1, characterized in that: The keywords in the corpus are screened, including: Use the TF-IDF algorithm to measure the importance of each word in the corpus; Sort all words by TF-IDF value from high to low, and take the top Generate keyword set In the present invention Get 50 experience points.

6. The method for instantiating a configuration of a smart substation according to claim 1, characterized in that: Acquiring a keyword based on the context of the keyword and the dictionary includes: Use context window technology to search for phrases in the context of keywords as candidate keywords; The candidate keyword is matched with the phrase in the dictionary. If the same phrase also exists in the dictionary, the candidate keyword is used as the keyword.

7. The method for instantiating a configuration of a smart substation according to claim 1, characterized in that: Obtaining a second vector of the entity text according to the keyword includes: Acquire a keyword granularity vector model according to the keyword; Acquire a keyword granularity vector model according to the keyword; The PCA principal component analysis algorithm is used to integrate the entity text based on the keyword granularity vector model and the representation vector obtained based on the keyword granularity vector model to obtain the second vector.

8. The method for instantiating a configuration of a smart substation according to claim 1, characterized in that: Calculating the similarity between the configuration information point and the information to be matched based on the comprehensive vector to complete the matching includes: The similarity between the configuration information point and the information to be matched is cosine similarity.

Images