Method for analyzing and judging the correlation degree of innovative technologies in different fields of power Internet of Things

Abstract

The present invention relates to a method for analyzing and judging the degree of association of innovative technologies in different fields of the electric power Internet of Things, and belongs to the technical field of data processing methods for management. The method divides the electric power Internet of Things into 8 sub-fields, retrieves and obtains documents, and extracts titles, abstracts, keywords, and publication years from the documents as document data; extracts sentences containing keywords in the abstract as inputs of the Spacy tool and trains to obtain an entity recognition model, and then traverses each sentence in the abstract for entity recognition to obtain key technical terms of the electric power Internet of Things; uses a word embedding model to map Chinese and English document data to a Chinese and English bilingual word embedding matrix, constructs a co-occurrence matrix of key technical terms and sub-fields, calculates the two-dimensional mutual information of any two of the sub-fields, and finally judges the strength of association between innovative technologies in any two sub-fields based on the two-dimensional mutual information. This method can provide a reliable data source for judging the degree of association between innovative technologies in the electric power Internet of Things in different fields.

Description

Technical Field

[0001] The present invention relates to a method for analyzing and judging the mutual synergistic relationship between innovative technologies in different sub-fields of the electric power Internet of Things, and belongs to the technical field of data processing methods suitable for management. Background Art

[0002] The power Internet of Things is an information-physical fusion system, and its construction process is also an innovative application process of Internet of Things related technologies in the power system. Researching the technical coupling points and collaborative innovation relationships between Internet of Things related technologies and power systems is conducive to finding breakthroughs in key technologies of the power Internet of Things and developing efficient innovation paths.

[0003] At present, the research on the coupling and collaboration of innovative technologies between power systems and the Internet of Things focuses on the technological development trend of the Internet of Things itself. However, since the power Internet of Things is a physical-information fusion system, its technological innovation includes two aspects: power system and Internet of Things construction. Therefore, the currently known research methods on the coupling and collaboration of innovative technologies between power systems and the Internet of Things cannot provide an effective and reliable analytical basis for judging the development direction of innovative technologies of the power Internet of Things. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an effective and reliable data basis for judging the development direction of innovative technologies in the power Internet of Things.

[0005] The technical solution proposed by the present invention to solve the above technical problems is: a method for analyzing and judging the degree of correlation of innovative technologies in different fields of the power Internet of Things, comprising the following steps:

[0006] Step 1: Division of the power Internet of Things field and collection of literature data, as follows:

[0007] The power Internet of Things is divided into eight sub-fields, namely, the power source end, the grid end, the load end, the storage end, and the Internet of Things perception layer, network layer, computing layer, and application layer; Chinese and English literature search formulas related to the sub-fields are constructed according to the definitions of the sub-fields, and each search formula contains several search terms; the technical sub-fields are searched from HowNet and Web of Science core databases according to the search formulas and Chinese and English literature are obtained respectively, and the titles, abstracts, keywords, and publication years of the Chinese and English literature are extracted as Chinese literature data and English literature data respectively, which together constitute the Chinese and English literature data;

[0008] Step 2: Obtain the key technical terms of the power Internet of Things, as follows:

[0009] Step 2.1 extract the sentences in the abstract of each document that contain the keywords corresponding to the document, use the extracted sentences as input of the Spacy tool and train to obtain an entity recognition model;

[0010] Step 2.2 uses the entity recognition model to traverse each sentence in all abstracts in the Chinese and English literature data for entity recognition. If the recognized entity is in the same sentence as the keyword of the document, the entity is used as the key technical term of the power Internet of Things, and the number of occurrences of all key technical terms of the power Internet of Things in the Chinese and English literature data is counted;

[0011] Step 3: The Chinese and English literature data of the power Internet of Things are uniformly vectorized and the word embedding model is used to map the Chinese and English literature data to the Chinese and English bilingual word embedding matrix, as follows:

[0012] Step 3.1 customizes a Chinese-English translation anchor file, wherein the Chinese-English translation anchor file defines a one-to-one correspondence between Chinese and English translations of common words, obtains an additional Chinese-English correspondence by calling the search terms in the Chinese search formula of Baidu Translate and the Chinese translation of the key technical terms obtained in step 2.2 into English, and then adds the additional Chinese-English correspondence to the Chinese-English translation anchor file;

[0013] Step 3.2: Use Chinese and English word segmentation tools to perform word segmentation on the Chinese and English document data, respectively, to obtain Chinese word sequences of the Chinese document data and English word sequences of the English document data, respectively, and use word2vec to train the Chinese word sequences and English word sequences of the document data to obtain word vectors for each word, and the word vectors constitute Chinese document word embedding matrices and English document word embedding matrices, respectively, and the dimension of each matrix is ​​the number of words in the corresponding Chinese document or English document data multiplied by the same word vector dimension d;

[0014] Step 3.3 builds a Chinese to English bilingual word vector mapping model, as shown in formula (1).

[0015]

[0016] Among them, d represents the dimension of word vector, M d (R) represents a d×d real matrix defined on the real number field R, S and T represent the Chinese embedding matrix and the English literature word embedding matrix respectively, W is the weight matrix, and argmin represents minimizing the distance between the Chinese literature word embedding matrix S and the English literature word embedding matrix T ||WS-T|| F ,|| || F Represents the Frobenius norm, and the calculation result is the optimal weight matrix W *, the Chinese and English bilingual word embedding matrix is ​​obtained through the bilingual word vector mapping model.

[0017] Step 4 constructs a co-occurrence matrix of the key technical terms and the sub-fields, as follows:

[0018] Step 4.1: Divide the search terms into 8 categories according to the 8 sub-fields, and extract the word vector corresponding to the search term according to the bilingual word embedding matrix of Chinese and English literature obtained in step 3 as the word vector v of the search term;

[0019] Step 4.2: Select the word vector corresponding to the key technical term from the bilingual word embedding matrix of Chinese and English literature obtained in step 3 as the word vector u of the key technical term, and calculate the similarity D(u,v) between the word vector u of the key technical term and the word vector v of the search term according to the cosine similarity. Assuming the similarity threshold is 0.3, when D(u,v)>0.3, the key technical term belongs to the sub-field corresponding to the word vector v of the search term;

[0020] Step 4.3: According to step 4.2, the subordinate relationship between the key technical terms and the sub-fields is obtained, and the sum of the times of occurrence of all the key technical terms corresponding to the sub-fields in the literature data is taken as the co-occurrence number of the key technical terms and the sub-fields, and the co-occurrence matrix of the key technical terms and the sub-fields is constructed according to the publication year of the literature;

[0021] Step 5 calculates the mutual information of any two sub-fields, as follows:

[0022] Step 5.1 For any two subdomains x 1 and x 2 According to formula (2), the one-dimensional information entropy H(x 1 ) and H(x 2 ).

[0023]

[0024] Among them, x is the sub-domain, c i The number of co-occurrences of key technical terms in the i-th (i=1,2,...,8) sub-field;

[0025] Step 5.1 Calculate the two sub-domains x according to formula (3) 1 and x 2 The two-dimensional information entropy H(x 1 ,x 2 ),

[0026]

[0027] Among them, c1 and c 2 Any two sub-domains x 1 and x 2 The number of co-occurrences of key technical terms,

[0028] Then these two sub-domains x 1 and x 2 The two-dimensional mutual information of is calculated by formula (4):

[0029] H(x 1 )+H(x 2 )-H(x 1 ,x 2 ) (4),

[0030] According to the two-dimensional mutual information, any two sub-domains x are judged 1 and x 2 The degree of correlation between innovative technologies.

[0031] The beneficial effects of the present invention are as follows: since the power Internet of Things includes two sub-fields, power and Internet of Things, the existing power Internet of Things innovative technology research based on scientific literature is mostly based on literature statistical metrology methods, and therefore lacks data analysis of the key technologies of the power Internet of Things involved in the content of scientific literature and the relationship between the key technologies and related sub-fields; the present invention starts from the perspective of power Internet of Things literature text data analysis, by mining the key technical terms contained in the power Internet of Things sub-field literature text, establishing the subordinate relationship between technical terms and sub-fields, and counting the number of co-occurrences of key technical terms of the power Internet of Things and the power Internet of Things sub-fields, to provide a more reliable data source for judging the degree of synergy between innovative technologies of the power Internet of Things in different fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The following is a further explanation of the method for analyzing and judging the degree of correlation of innovative technologies in different fields of the power Internet of Things of the present invention in conjunction with the accompanying drawings.

[0033] Figure 1 It is the distribution diagram of the Chinese literature word embedding matrix in the two-dimensional plane.

[0034] Figure 2 It is the distribution diagram of the word embedding matrix of English literature in the two-dimensional plane.

[0035] Figure 3 It is the distribution diagram of the Chinese and English bilingual word embedding matrix in the two-dimensional plane.

[0036] Figure 4 It is a relationship diagram of the mutual information between the three pairs of sub-fields: source-load, source-storage, and network-storage. DETAILED DESCRIPTION

[0037] Example

[0038] The method for analyzing and judging the degree of correlation of innovative technologies in different fields of the power Internet of Things of this embodiment includes the following steps:

[0039] Step 1: Division of the power Internet of Things field and collection of literature data, as follows:

[0040] The power Internet of Things is divided into 8 sub-fields, namely, the power source end, the network end, the load end, the storage end, and the Internet of Things perception layer, network layer, computing layer, and application layer; Chinese and English literature retrieval formulas related to the sub-fields are constructed according to the definitions of the sub-fields, and each retrieval formula contains several search terms; the technical sub-fields are searched from HowNet and Web of Science core databases according to the retrieval formulas and Chinese and English literature are obtained respectively, and the titles, abstracts, keywords, and publication years of these documents (including Chinese and English documents) are extracted as literature data (including Chinese and English literature data); the Chinese and English literature retrieval formulas of this embodiment are shown in Table 1 below.

[0041] Table 1

[0042]

[0043]

[0044]

[0045] The number of documents retrieved in this embodiment is shown in Table 2.

[0046] Table 2

[0047]

[0048] Step 2: Obtain the key technical terms of the power Internet of Things, as follows:

[0049] Step 2.1 extract the sentences in the abstract of each document that contain the keywords corresponding to the document, use the extracted sentences as the input of the Spacy tool and train to obtain an entity recognition model. Spacy is an open source tool designed for NLP word segmentation, entity recognition, and part-of-speech tagging, and supports custom training of entity recognition models;

[0050] Step 2.2 uses the entity recognition model to traverse each sentence in all the abstracts in the literature data for entity recognition. If the recognized entity is in the same sentence as the keyword of the literature, the entity is used as the key technical term of the power Internet of Things, and the number of occurrences of all key technical terms of the power Internet of Things in the literature data is counted. The key technical terms with higher occurrence frequency are shown in Table 3.

[0051] Table 3

[0052]

[0053] Step 3: Unify the vectorization of Chinese and English literature data of the power Internet of Things, and use the word embedding model to map the Chinese and English literature data to the Chinese and English bilingual word embedding matrix. In order to avoid the influence of the difference in Chinese and English literature data on the determination of the sub-field affiliation of the key technical terms of the power Internet of Things, the multilingual natural language processing word embedding technology is used to vectorize the Chinese and English literature data, and obtain the Chinese and English literature word embedding matrix distributed in the same vector space, which is convenient for constructing the subordinate relationship between the key technical terms of the power Internet of Things and the sub-fields, as follows:

[0054] Step 3.1 customizes a Chinese-English translation anchor file, wherein the Chinese-English translation anchor file defines a one-to-one correspondence between Chinese and English translations of common words, obtains an additional Chinese-English correspondence by calling the search terms in the Chinese search formula of Baidu Translate and the Chinese translation of the key technical terms obtained in step 2.3 into English, and then adds the additional Chinese-English correspondence to the Chinese-English translation anchor file;

[0055] Step 3.2 Use Chinese and English word segmentation tools (such as jieba, nltk) to perform word segmentation on the document data (including Chinese document data and English document data), respectively, and obtain Chinese word sequences and English word sequences of these document data, respectively, and use word2vec to train the Chinese word sequence and the English word sequence to obtain the word vector of each word (the word2vec model can represent words as multidimensional vectors, thereby mapping the text to a word embedding matrix composed of multidimensional vectors), and the word vectors constitute the Chinese document word embedding matrix and the English document word embedding matrix, respectively, and the dimension of each matrix is ​​the number of words in the corresponding document data (Chinese document data or English document data) multiplied by the same word vector dimension d; the word vector represents each word as a vector, and the dimension of the word vector represents the number of elements contained in this vector. In this embodiment, the word vector dimension is set to 300. Figure 1 and Figure 2 The distribution of the Chinese literature word embedding matrix and the English literature word embedding matrix in the two-dimensional plane are shown respectively.

[0056] Step 3.3: Construct a bilingual word vector mapping model from Chinese to English, as shown in formula (1):

[0057]

[0058] Among them, d represents the dimension of word vector, M d(R) represents a d×d real matrix defined on the real number field R, S and T represent the Chinese embedding matrix and the English literature word embedding matrix respectively, W is the weight matrix, and argmin represents minimizing the distance between the Chinese literature word embedding matrix S and the English literature word embedding matrix T ||WS-T|| F ,|| || F Represents the Frobenius norm, and the calculation result is the optimal weight matrix W * ; The Chinese and English bilingual word embedding matrix is ​​obtained through the bilingual word vector mapping model.

[0059] The optimization goal of this model is to find the weight matrix W so that the distance between the Chinese literature word embedding matrix and the English literature word embedding matrix is ​​||WS-T|| F The shortest, thus unifying the vector space where the Chinese literature word embedding matrix and the English literature word embedding matrix are located; the model can be transformed into a Procrustes problem and the singular value decomposition and gradient descent method are used to iteratively solve the optimal W * , where the translation anchor file provides a one-to-one correspondence between some Chinese and English reference words. The distance between any two Chinese and English word vectors in the Chinese and English literature word embedding matrix can be indirectly calculated by solving the distance between each word vector and the word vector of the same language reference word. Therefore, through the weight matrix W * The Chinese literature word embedding matrix can be mapped to the same vector space as the English literature word embedding matrix, so that the word vectors in the word embedding matrix can be compared with each other, and together they form a bilingual word embedding matrix of Chinese and English literature. This word embedding matrix contains the word vectors corresponding to all word sequences of Chinese and English literature data, such as Figure 3 The figure shows the distribution of the Chinese and English bilingual word embedding matrix in a two-dimensional plane.

[0060] Step 4 constructs a co-occurrence matrix of the key technical terms and the sub-fields, as follows:

[0061] Step 4.1: Take the search terms in the search formula in Table 1 and divide them into 8 categories according to 8 sub-fields. Take the word vector corresponding to the search term according to the bilingual word embedding matrix of Chinese and English literature obtained in step 3 as the word vector v of the search term.

[0062] Step 4.2: Select the word vector corresponding to each key technical term from the word vectors obtained in step 4.1 as the word vector u of each key technical term. Calculate the similarity D(u,v) between the word vector u of each key technical term and the word vector v of the search term based on the cosine similarity. Assume that the similarity threshold is 0.3. When D(u,v)>0.3, the key technical term belongs to the subfield corresponding to the word vector v of the search term.

[0063] Step 4.3: According to the calculation process of step 4.2, the subordinate relationship between each key technical term and the eight sub-fields can be obtained. The sum of the number of times all key technical terms corresponding to the eight sub-fields appear in the literature data is taken as the co-occurrence number of key technical terms and sub-fields (referred to as term-fields). According to the publication year of the literature to which each key technical term belongs, a co-occurrence matrix between each key technical term and the eight sub-fields is constructed; as shown in Table 4.

[0064] Table 4

[0065]

[0066]

[0067] Step 5: Calculate the mutual information between any two sub-domains.

[0068] Step 5.1: For any two subdomains x 1 and x 2 According to formula (2), the one-dimensional information entropy H(x 1 ) and H(x 2 ),

[0069]

[0070] Among them, x is the sub-domain, c i The number of co-occurrences of key technical terms in the i-th (i=1,2,...,8) sub-field; for example, the one-dimensional information entropy of the power source terminal field in 2010 is

[0071] Step 5.1: Calculate the two sub-domains x according to formula (3) 1 and x 2 The two-dimensional information entropy H(x 1 ,x 2 ),

[0072]

[0073] Among them, c 1 and c 2 Any two sub-domains x 1 and x 2 The number of co-occurrences of key technical terms,

[0074] Then these two sub-domains x 1 and x 2 The two-dimensional mutual information of is calculated by formula (4):

[0075] H(x 1 )+H(x 2 )-H(x1 ,x 2 ) (4),

[0076] Figure 4 The calculation results of the mutual information between three pairs of sub-fields, source-load, source-storage, and network-storage, are shown. The two-dimensional mutual information of any two sub-fields from 2010 to 2019 is added and averaged to obtain the two-dimensional average mutual information of 8 sub-fields, as shown in Table 5.

[0077] Table 5 (mbit)

[0078]

[0079] According to the two-dimensional mutual information calculated above, we can judge whether any two sub-domains x 1 and x 2 The degree of correlation between innovative technologies.

[0080] The above description is only a preferred embodiment of the present invention, but the present invention is not limited thereto. For example, all equivalent replacements or equivalent changes of the concepts and technical solutions according to the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing and judging the correlation degree of innovative technologies in different fields of the power Internet of Things. Features The following steps are involved: Step 1: Division of the power Internet of Things field and collection of literature data, as follows: The power Internet of Things is divided into eight sub-fields, namely, the power source end, the grid end, the load end, the storage end, and the Internet of Things perception layer, network layer, computing layer, and application layer; Chinese and English literature search formulas related to the sub-fields are constructed according to the definitions of the sub-fields, and each search formula contains several search terms; the sub-fields are searched from HowNet and Web of Science core databases according to the search formulas and Chinese and English literature are obtained respectively, and the titles, abstracts, keywords, and publication years of the Chinese and English literature are extracted as Chinese literature data and English literature data respectively, which together constitute the Chinese and English literature data; Step 2: Obtain the key technical terms of the power Internet of Things, as follows: Step 2.1 extract the sentences in the abstract of each document that contain the keywords corresponding to the document, use the extracted sentences as input of the Spacy tool and train to obtain an entity recognition model; Step 2.2 uses the entity recognition model to traverse each sentence in all abstracts in the Chinese and English literature data to perform entity recognition. If the recognized entity and the keyword of the document are in the same sentence, the entity is used as the key technical term of the power Internet of Things; Step 3: The Chinese and English literature data of the power Internet of Things are uniformly vectorized and the word embedding model is used to map the Chinese and English literature data to the Chinese and English bilingual word embedding matrix, as follows: Step 3.1 customizes a Chinese-English translation anchor file, wherein the Chinese-English translation anchor file defines a one-to-one correspondence between Chinese and English translations of common words, obtains an additional Chinese-English correspondence by calling the search terms in the Chinese search formula of Baidu Translate and the Chinese translation of the key technical terms obtained in step 2.2 into English, and then adds the additional Chinese-English correspondence to the Chinese-English translation anchor file; Step 3.2: Use Chinese and English word segmentation tools to perform word segmentation on the Chinese and English document data, respectively, to obtain Chinese word sequences of the Chinese document data and English word sequences of the English document data, respectively, and use word2vec to train the Chinese word sequences and English word sequences of the document data to obtain word vectors for each word, and the word vectors constitute Chinese document word embedding matrices and English document word embedding matrices, respectively, and the dimension of each matrix is ​​the number of words in the corresponding Chinese document or English document data multiplied by the same word vector dimension d; Step 3.3: Construct a bilingual word vector mapping model from Chinese to English, as shown in formula (1): Among them, d represents the dimension of word vector, M d (R) represents a d×d real matrix defined on the real number field R, S and T represent the Chinese embedding matrix and the English literature word embedding matrix respectively, W is the weight matrix, and argmin represents minimizing the distance between the Chinese literature word embedding matrix S and the English literature word embedding matrix T ||WS-T|| F ,|| || F Represents the Frobenius norm, and the calculation result is the optimal weight matrix W * , obtaining a Chinese-English bilingual word embedding matrix through the bilingual word vector mapping model; Step 4 constructs a co-occurrence matrix of the key technical terms and the sub-fields, as follows: Step 4.1: Divide the search terms into 8 categories according to the 8 sub-fields, and extract the word vector corresponding to the search term according to the bilingual word embedding matrix of Chinese and English literature obtained in step 3 as the word vector v of the search term; Step 4.2: Select the word vector corresponding to the key technical term from the bilingual word embedding matrix of Chinese and English literature obtained in step 3 as the word vector u of the key technical term, and calculate the similarity D(u,v) between the word vector u of the key technical term and the word vector v of the search term according to the cosine similarity. Assuming the similarity threshold is 0.3, when D(u,v)>0.3, the key technical term belongs to the sub-field corresponding to the word vector v of the search term; Step 4.3: According to step 4.2, the subordinate relationship between the key technical terms and the sub-fields is obtained, and the sum of the times of occurrence of all the key technical terms corresponding to the sub-fields in the literature data is taken as the co-occurrence number of the key technical terms and the sub-fields, and the co-occurrence matrix of the key technical terms and the sub-fields is constructed according to the publication year of the literature; Step 5 calculates the mutual information of any two sub-fields, as follows: Step 5.1 For any two subdomains x 1 and x 2 According to formula (2), the one-dimensional information entropy H(x 1 ) and H(x 2 ), Among them, x is the sub-domain, c i The number of co-occurrences of key technical terms in the i-th (i=1,2,...,8) sub-field; Step 5.2 Calculate the two sub-domains x according to formula (3) 1 and x 2 The two-dimensional information entropy H(x 1 ,x 2 ), Among them, c 1 and c 2 Any two sub-domains x 1 and x 2 The co-occurrence of key technical terms in these two sub-fields x 1 and x 2 The two-dimensional mutual information of is calculated by formula (4): H(x 1 )+H(x 2 )-H(x 1 ,x 2 ) (4), According to the two-dimensional mutual information, any two sub-domains x are judged 1 and x 2 The degree of correlation between innovative technologies.

Images