Method and system for constructing AI semantic large model of power distribution network standard knowledge base

Abstract

The invention relates to the technical field of smart power grids, and discloses an AI semantic large model construction method and system for a power distribution network standard knowledge base, and the method comprises the steps: carrying out the knowledge extraction of power distribution network data obtained in advance, obtaining knowledge data, carrying out the knowledge fusion of the knowledge data, and obtaining a knowledge graph; performing entity keyword extraction on the knowledge data to obtain keywords, performing semantic recognition combination on the keywords to obtain key statements of the keywords, and performing classification recognition on the key statements to obtain category labels of the key statements; performing large model text conversion storage on the category labels, the knowledge graph and a pre-acquired power distribution network operation rule to obtain a standard knowledge base, and constructing an AI semantic large model based on the standard knowledge base; and obtaining a real-time power distribution network problem of the user, and inputting the real-time power distribution network problem into the AI semantic large model to obtain an answer to the real-time power distribution network problem. According to the method, the problem processing capability of the AI semantic large model of the power distribution network standard knowledge base can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid, in particular to an AI semantic large model construction method and system of power distribution network standard knowledge base. BACKGROUND

[0002] At present, the demand for intelligentization of power distribution network management is increasing with each passing day. The power distribution network data is large in scale and complex in structure, covering equipment parameters, operation status, user information and other aspects. Constructing an AI semantic large model of power distribution network standard knowledge base has become a key technical means to realize efficient and intelligent power distribution network management, which can quickly process and analyze massive data to provide accurate support for operation and maintenance decision-making, fault diagnosis and other aspects.

[0003] However, the existing construction method of AI semantic large model of power distribution network standard knowledge base has many defects. On the one hand, in the knowledge extraction and fusion link, the traditional method has insufficient processing capacity for noise and redundant information in the data, resulting in poor accuracy of extracted knowledge and imperfect construction of knowledge graph, which cannot fully and accurately reflect the real situation of the power distribution network. On the other hand, in the model training and application stage, due to the lack of effective semantic understanding and logical reasoning ability, the constructed model has weak processing capacity for complex power distribution network problems. When encountering complex scenarios such as concurrent multiple faults and cross-region fault diagnosis, the model cannot quickly and accurately provide solutions, which seriously affects the operation efficiency and power supply reliability of the power distribution network. SUMMARY

[0004] The present application provides an AI semantic large model construction method and system of power distribution network standard knowledge base, which mainly aims to solve the problem of weak processing capacity of AI semantic large model of power distribution network standard knowledge base.

[0005] To achieve the above purpose, the present application provides an AI semantic large model construction method of power distribution network standard knowledge base, which comprises: S1, knowledge extraction is performed on the pre-acquired power distribution network data to obtain knowledge data of the power distribution network, and knowledge fusion is performed on the knowledge data to obtain a knowledge graph of the power distribution network; S2, entity keyword extraction is performed on the knowledge data to obtain keywords of the power distribution network, semantic recognition and combination are performed on the keywords to obtain key sentences of the keywords, and classification recognition is performed on the key sentences to obtain class labels of the key sentences; S3, large model text conversion storage is performed on the class labels, the knowledge graph and pre-acquired power distribution network operation rules to obtain a standard knowledge base of the power distribution network, and an AI semantic large model is constructed based on the standard knowledge base; S4, obtaining a real-time power distribution network problem of a user, inputting the real-time power distribution network problem into the AI semantic large model to obtain an answer to the real-time power distribution network problem.

[0006] In a preferred embodiment, the knowledge extraction on the pre-acquired power distribution network data to obtain knowledge data of the power distribution network comprises: The pre-acquired power distribution network data is subjected to noise cleaning to obtain standardized power distribution network data; The standardized power distribution network data is subjected to entity relationship separation to obtain entity relationships of the standardized power distribution network data; The pre-acquired power distribution network data is subjected to knowledge extraction based on the entity relationships to obtain knowledge data of the power distribution network.

[0007] In a preferred embodiment, the knowledge fusion on the knowledge data to obtain the knowledge graph of the power distribution network comprises: The knowledge data is subjected to entity alignment and disambiguation to obtain knowledge entities; The knowledge entities are subjected to multi-modal knowledge fusion to obtain the knowledge graph of the power distribution network.

[0008] In a preferred embodiment, the entity keyword extraction on the knowledge data to obtain the keywords of the power distribution network comprises: The knowledge data is subjected to word segmentation and part-of-speech tagging to obtain a vocabulary of the knowledge data; The vocabulary is subjected to stop word filtering to obtain a filtered vocabulary of the vocabulary; The knowledge data is subjected to feature word and sentence extraction based on the TF-IDF algorithm and the filtered vocabulary to obtain feature words and sentences of the knowledge data; The knowledge data is subjected to key segmentation based on the TextRank algorithm and the filtered vocabulary to obtain segmented sections of the knowledge data; The feature words and the segmented sections are subjected to fusion, disambiguation and extraction to obtain the keywords of the power distribution network.

[0009] In a preferred embodiment, the semantic recognition and combination on the keywords to obtain key sentences of the keywords comprises: The keywords are subjected to context semantic analysis using a BERT model to obtain context semantic feature vectors of the keywords; The context semantic feature vectors are subjected to local semantic feature extraction using a TextCNN model to obtain local semantic feature vectors of the keywords; The local semantic feature vectors are subjected to semantic recognition and combination using a Softmax function to obtain key sentences of the keywords.

[0010] In a preferred embodiment, the classifying and identifying the key sentence to obtain the class label of the key sentence comprises: character splitting the key sentence by using an embedding layer in the Bi-LSTM-CRF model to obtain a word vector of the key sentence; semantic extraction of the word vector by using a long short-term memory neural network layer in the Bi-LSTM-CRF model to obtain a semantic expression vector of the word vector; classifying and identifying the semantic expression vector by using a conditional random field layer in the Bi-LSTM-CRF model to obtain the class label of the key sentence.

[0011] In a preferred embodiment, the large model text conversion and storage of the class label, the knowledge graph and the pre-acquired power distribution network operation rule to obtain the standard knowledge base of the power distribution network comprises: collecting the class label, the knowledge graph and the pre-acquired power distribution network operation rule as a data set of the power distribution network; large model text conversion of the data set to obtain text data of the power distribution network; distributed storage of the text data to obtain the standard knowledge base of the power distribution network.

[0012] In a preferred embodiment, the construction of the AI semantic large model based on the standard knowledge base comprises: mapping relationship of the standard knowledge base to a SQL database; SQL conversion of the standard knowledge base based on the mapping relationship and a Text2SQL.AI tool to obtain a SQL statement of the standard knowledge base; semantic training of the SQL statement by using a Seq2Seq model to obtain an AI semantic large model of the standard knowledge base.

[0013] In a preferred embodiment, the real-time power distribution network problem comprises repair information, work order information, power distribution monitoring information, main line fault, branch line fault and substation monitoring.

[0014] To solve the above problems, the application further provides an AI semantic large model construction system of a power distribution network standard knowledge base, which comprises: a knowledge graph generation module configured to perform knowledge extraction on pre-acquired power distribution network data to obtain knowledge data of the power distribution network, and perform knowledge fusion on the knowledge data to obtain a knowledge graph of the power distribution network; The category label generation module is configured to perform entity keyword extraction on the knowledge data to obtain keywords of the power distribution network, perform semantic recognition and combination on the keywords to obtain key sentences of the keywords, and perform classification recognition on the key sentences to obtain category labels of the key sentences. The AI semantic large model construction module is configured to perform large model text conversion and storage on the category labels, the knowledge graph, and pre-acquired power distribution network operation rules to obtain a standard knowledge base of the power distribution network, and construct an AI semantic large model based on the standard knowledge base. The AI semantic large model application module is configured to acquire a real-time power distribution network problem of a user, input the real-time power distribution network problem into the AI semantic large model, and obtain an answer to the real-time power distribution network problem.

[0015] Compared with the prior art, the present application has the following beneficial effects: 1. The AI semantic large model constructed by the present application can significantly improve the problem processing capability of the power distribution network. In the data processing and knowledge construction link, accurate knowledge graphs are obtained by comprehensively extracting and fusing knowledge from power distribution network data. At the same time, key words are extracted from knowledge data, semantic recognition and classification are performed, and multi-source information is integrated into a standard knowledge base. This process effectively integrates complex power distribution network information, provides a rich and accurate knowledge base for the model, and enables the model to quickly locate relevant knowledge when facing various power distribution network problems, thereby laying a solid foundation for accurate problem solving.

[0016] 2. The AI semantic large model constructed by the present application can efficiently handle various real-time power distribution network problems due to its strong knowledge reserve and semantic understanding capability. Whether it is for repair information, work order information processing, power distribution monitoring, fault diagnosis, or other aspects, it can more deeply understand the semantic connotation of the problem, accurately match the relevant knowledge in the knowledge base, greatly improve the efficiency and accuracy of handling complex problems, and effectively improve the intelligent level and operation and maintenance efficiency of power distribution network management. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of an AI semantic large model construction method of a power distribution network standard knowledge base according to an embodiment of the present application is provided. Figure 2 A functional module diagram of an AI semantic large model construction system of a power distribution network standard knowledge base according to an embodiment of the present application is provided. The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0018] It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0019] The embodiment of the application provides an AI semantic large model construction method of a power distribution network standard knowledge base. The execution subject of the AI semantic large model construction method of the power distribution network standard knowledge base includes but is not limited to at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the application. In other words, the AI semantic large model construction method of the power distribution network standard knowledge base can be executed by software or hardware installed in a terminal device or a server device. The server includes but is not limited to a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be a stand-alone server, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0020] Referring to Figure 1 FIG. 1 is a flowchart of an AI semantic large model construction method of a power distribution network standard knowledge base provided by an embodiment of the application. In this embodiment, the AI semantic large model construction method of the power distribution network standard knowledge base includes the following steps. S1, knowledge extraction is performed on pre-acquired power distribution network data to obtain knowledge data of the power distribution network, and knowledge fusion is performed on the knowledge data to obtain a knowledge graph of the power distribution network; In the embodiment of the application, the knowledge extraction performed on the pre-acquired power distribution network data to obtain the knowledge data of the power distribution network includes the following steps. Noise points in the pre-acquired power distribution network data are cleaned to obtain standardized power distribution network data; Entity relationships in the standardized power distribution network data are separated to obtain entity relationships of the standardized power distribution network data; Knowledge extraction is performed on the pre-acquired power distribution network data based on the entity relationships to obtain the knowledge data of the power distribution network.

[0021] The knowledge fusion performed on the knowledge data to obtain the knowledge graph of the power distribution network includes the following steps. Entity alignment and disambiguation are performed on the knowledge data to obtain knowledge entities; Multi-modal knowledge fusion is performed on the knowledge entities to obtain the knowledge graph of the power distribution network.

[0022] Specifically, the power distribution network data is divided into unstructured data, semi-structured data and structured data, and mainly includes text data such as power distribution network engineering construction specifications, design specifications, operation standards, unmanned aerial vehicle patrol standards, patrol pictures, graphic model files and expert experience, etc.

[0023] Further, structured data refers to relational databases in the power grid, including real-time and historical operation data of the power grid, device asset information databases, properties and topology of power grid devices, as well as meteorological data, environment around the power grid, user load, and other information.

[0024] Further, semi-structured data includes power knowledge encyclopedia or power professional lexicon related to distribution data; unstructured data mainly refers to text data, including distribution operation management regulations, evaluation guidelines, distribution operation and maintenance cases, etc.

[0025] Specifically, clean up errors, missing values, and outliers in the data to ensure data quality.

[0026] Further, unify data formats, units, and other data in different data sources for better comparison and analysis.

[0027] Further, determine the association between different data sources and establish a data model for matching.

[0028] Further, use appropriate methods to integrate information from different data sources, such as database connection, data association operation, etc.

[0029] Further, convert data into a unified format to adapt to the overall data structure.

[0030] Further, combine data from different data sources into an integrated data storage, such as a data warehouse.

[0031] Further, establish metadata to describe the source, format, meaning, and other information of the data.

[0032] Further, track and maintain metadata to ensure data understandability and maintainability.

[0033] Further, design metrics to evaluate data quality and develop corresponding control strategies.

[0034] Further, implement an anomaly detection mechanism to promptly discover and handle data quality issues.

[0035] Further, take measures to ensure data security during integration, including encryption, access control, etc.

[0036] Further, ensure compliance with relevant privacy regulations when integrating data and take privacy protection measures such as data anonymization, desensitization, etc., to ultimately obtain standardized distribution grid data.

[0037] Further, establish data monitoring to regularly monitor the quality and performance of integrated data.

[0038] Further, a data update mechanism is designed to ensure that the data is always up-to-date.

[0039] Specifically, with the help of natural language processing and machine learning techniques, entities in the data are identified. A device name library is constructed, and methods such as keyword matching and regular expressions are used to extract the names and models of power distribution network devices. Geographic information system (GIS) data and address resolution algorithms are used to identify and locate geographic entities related to the power distribution network. A named entity recognition (NER) model can also be built, and deep learning algorithms are used to train the data to automatically identify various entities.

[0040] Further, after completing entity recognition, the relationship between entities is extracted. The syntax and semantics of the data are analyzed to explore the association between entities. By analyzing power topology data, the connection method and sequence of devices such as transformers, switch cabinets, and lines are sorted out to determine the device connection relationship. Based on GIS data, the geographical location relationship of different devices is clarified. Semantic analysis techniques are used to extract the ownership relationship between users and devices from text descriptions.

[0041] Further, after completing entity recognition, the relationship between entities is extracted. The syntax and semantics of the data are analyzed to explore the association between entities. By analyzing power topology data, the connection method and sequence of devices such as transformers, switch cabinets, and lines are sorted out to determine the device connection relationship. Based on GIS data, the geographical location relationship of different devices is clarified. Semantic analysis techniques are used to extract the ownership relationship between users and devices from text descriptions. Further, the extracted entity relationships are verified by comparing them with known power distribution network topology and business rules to find and correct incorrect relationships. The completeness of the relationship extraction is evaluated to see if important relationships are missing, and finally the entity relationships of the standardized power distribution network data are obtained.

[0042] Specifically, according to the business characteristics and analysis requirements of the power distribution network, knowledge extraction rules are formulated.

[0043] For example, for fault diagnosis knowledge extraction, when the device connection relationship is interrupted and the device state is abnormal, the fault type and fault location are extracted. For load forecasting, rules are set to extract user power consumption patterns and load variation rules based on user and device ownership relationships and historical power consumption data.

[0044] Further, according to the established knowledge extraction rules, the data after association and integration is processed. By writing programs or using data mining tools, knowledge that meets the rules is automatically extracted from the data. Taking device connection relationship data as an example, graph traversal algorithms are used to extract upstream and downstream relationships and power transmission paths between devices based on the topology of device connection relationships.

[0045] Further, the extracted knowledge data is quality checked to check the accuracy, completeness and consistency of the knowledge.

[0046] Further, the extracted knowledge is compared with the actual operation of the power distribution network and industry standards to verify the correctness of the knowledge.

[0047] Further, it is checked whether the knowledge is repeated or contradictory to ensure the reliability of the knowledge data.

[0048] Specifically, for each entity in the knowledge data, a unique identifier and index are established.

[0049] Further, based on key attributes such as device name, model, and geographic location, each entity is assigned a unique ID.

[0050] For example, for a transformer entity, a unique ID is generated by combining its name, manufacturer, and factory number. With these IDs, an inverted index is constructed to facilitate subsequent quick positioning and retrieval of entities, laying the foundation for entity alignment and disambiguation.

[0051] Further, the attributes and characteristics of the entity are extracted from the knowledge data, and algorithms such as string matching and cosine similarity calculation are used to compare entities from different data sources.

[0052] Further, by comparing attributes such as device name, model, and parameters, different records referring to the same entity are found.

[0053] Further, for attributes with differences, domain knowledge and actual business are combined to make reasonable judgments and mergers, eliminate entity redundancy, and ensure that each entity has uniqueness in the knowledge data.

[0054] Further, for entities with the same name but different actual objects, the true meaning of the entity is accurately determined through analysis of the context information of the entity and the use of semantic understanding technology, ultimately obtaining the knowledge entity.

[0055] For example, in power distribution network data, there may be devices with the same name in different regions. By analyzing the geographic location and connection relationship of the device, the object referred to by each device with the same name is determined, ambiguity is avoided, and the accuracy of the knowledge data is improved.

[0056] Specifically, the knowledge entity after entity alignment and disambiguation is integrated with the information extracted from multi-modal data.

[0057] Further, taking a device entity as an example, the image information and text description information of the device are integrated with the structured data such as device account and connection relationship to enrich the attribute and relationship information of the entity.

[0058] Further, by means of the graph structure of the knowledge graph, the knowledge of different modalities is organically connected to form a comprehensive and three-dimensional power distribution network knowledge graph.

[0059] Further, the constructed knowledge graph is verified to check the rationality of entity relationships, the integrity and consistency of knowledge in the graph.

[0060] Further, by comparing with the actual operation of the power distribution network and expert experience, errors and loopholes in the graph are found and corrected.

[0061] In general, the knowledge graph is a structured semantic knowledge base used to describe concepts and their interconnections in the physical world in symbolic form. Its basic component unit is a "entity-relation-entity" triple, and the entity and its related attributes are always connected to each other through the relationship, forming a network-like knowledge structure.

[0062] In general, the power distribution network data is large in scale and complex in structure. Through knowledge extraction, valuable information is extracted from the pre-acquired power distribution network data to obtain knowledge data.

[0063] In general, these knowledge data contain device parameters, operating status, user information and other aspects, providing rich raw data for subsequent knowledge fusion and model construction.

[0064] In general, knowledge fusion further integrates these knowledge data to construct a knowledge graph, presenting various entities and their relationships in the power distribution network in a structured form, making the data more orderly, easy to understand and use, and providing a clear and accurate knowledge base for AI semantic large models.

[0065] S2, entity keyword extraction is performed on the knowledge data to obtain keywords of the power distribution network, semantic recognition and combination are performed on the keywords to obtain key sentences of the keywords, and classification recognition is performed on the key sentences to obtain class labels of the key sentences. In the embodiment of the present application, the entity keyword extraction on the knowledge data to obtain the keywords of the power distribution network comprises: The knowledge data is segmented and tagged with parts of speech to obtain a vocabulary table of the knowledge data. The vocabulary table is filtered for stop words to obtain a filtered vocabulary table of the vocabulary table. The knowledge data is extracted for feature words and sentences based on the TF-IDF algorithm and the filtered vocabulary table to obtain feature words and sentences of the knowledge data. The knowledge data is segmented based on the TextRank algorithm and the filtered vocabulary table to obtain segmented sentences of the knowledge data. The characteristic words and sentences and the segmented segments are fused and disambiguated to obtain keywords of the distribution network.

[0066] The semantic recognition and combination of the keywords to obtain key sentences of the keywords includes: Performing contextual semantic analysis on the keyword using a BERT model to obtain a contextual semantic feature vector of the keyword; Using a TextCNN model to extract local semantic features from the contextual semantic feature vector to obtain a local semantic feature vector of the keyword; The Softmax function is used to perform semantic recognition and combination on the local semantic feature vectors to obtain the key sentences of the keywords.

[0067] The classifying and identifying the key sentences to obtain the category labels of the key sentences includes: Using the embedding layer in the Bi-LSTM-CRF model to perform character splitting on the key sentence to obtain the word vector of the key sentence; Using the long short-term memory neural network layer in the Bi-LSTM-CRF model to perform semantic extraction on the word vector to obtain a semantic expression vector of the word vector; The conditional random field layer in the Bi-LSTM-CRF model is used to classify and identify the semantic expression vector to obtain the category label of the key sentence.

[0068] Specifically, the knowledge data is segmented using the selected segmentation tool.

[0069] Furthermore, the word segmentation tool will divide the continuous text stream according to word boundaries based on preset algorithms and rules.

[0070] For example, for the sentence "The distribution network transformer is in good operating condition," the word segmentation tool will break it down into individual words, such as "distribution network," "transformer," "operating condition," and "good." During the word segmentation process, it's crucial to ensure the tool accurately identifies specialized terms in the distribution network field, such as "feeder" and "switchgear," to avoid incorrect segmentation and ensure the segmentation results are consistent with the distribution network context.

[0071] Furthermore, after completing the word segmentation, each word segmentation result is tagged with a part-of-speech tagging tool or function.

[0072] Furthermore, part-of-speech tagging will assign corresponding part-of-speech tags to words based on their grammatical functions and semantic features in the sentence. For example, "distribution network" may be marked as a noun, and "operation" as a verb.

[0073] Further, there are various common part-of-speech tagging sets, such as the Peking University part-of-speech tagging set for Chinese and the Penn Treebank part-of-speech tagging set for English.

[0074] Further, the tagging set matching the language type of the knowledge data is selected to ensure that the tagging result accurately reflects the grammatical properties of the words, providing more information for subsequent analysis and processing.

[0075] Further, the results of word segmentation and part-of-speech tagging are organized to construct a vocabulary.

[0076] Further, the vocabulary is presented in a list or dictionary data structure, and each entry contains the word segmentation result and its corresponding part-of-speech.

[0077] For example, the vocabulary may have entries such as "transformer - noun" and "monitor - verb". When generating the vocabulary, duplicate entries are removed to ensure the simplicity and uniqueness of the vocabulary, facilitating subsequent quick queries and statistical analysis of the vocabulary in the knowledge data.

[0078] Specifically, a suitable stop word list for power grid knowledge data needs to be selected or created.

[0079] Further, reference can be made to general stop word lists, such as Chinese stop word lists and English stop word lists. These stop word lists contain common words that have no actual meaning or contribute less to text analysis, such as Chinese words like "de" "le" "zai" and English words like "the" "and" "is".

[0080] Further, considering the characteristics of the power grid field, professional domain-specific stop words are supplemented, such as some professional abbreviations that only serve as connectors and have no substantive technical meaning.

[0081] Further, these words are organized and summarized to form the final stop word list for filtering.

[0082] Further, the determined stop word list is loaded into the processing program so that it can be quickly accessed in memory.

[0083] Further, the vocabulary to be processed is also loaded to ensure that subsequent comparisons between each word in the vocabulary and the stop word list can be made.

[0084] Further, whether using data structures in programming languages, such as lists and sets in Python, or other data storage methods, the loaded stop word list and vocabulary should be able to efficiently perform data retrieval and operations.

[0085] Further, each word in the vocabulary is compared with the words in the stop word list one by one.

[0086] Further, if a certain vocabulary in the vocabulary table has a matching item in the stop word table, the vocabulary is removed from the vocabulary table; if there is no match, the vocabulary is retained.

[0087] Further, during the comparison process, attention should be paid to the format problems such as case, full-width and half-width of the vocabulary, to ensure the accuracy of the match.

[0088] For example, when comparing English words, the words are converted to lowercase for comparison to avoid misjudgment due to case differences.

[0089] Further, after the stop word filtering operation, the remaining words constitute the filtered vocabulary table.

[0090] Further, these words are reorganized in a certain order, such as maintaining the order of the original vocabulary table, or sorting according to certain attributes of the words (such as alphabetical order, frequency of occurrence, etc.).

[0091] Further, the filtered vocabulary table is stored in a suitable data format, such as a text file, a database table, or a data structure in a programming language (such as a list, a dictionary, etc.), to facilitate subsequent use and processing, and to provide a more concise and effective data basis for further analysis and application based on the vocabulary table.

[0092] Specifically, the term frequency (TF) of each vocabulary in the filtered vocabulary table in an independent chapter is calculated, that is, the number of occurrences of the vocabulary in the chapter divided by the total number of words in the chapter.

[0093] Further, the inverse document frequency (IDF) is calculated, which is obtained by taking the logarithm of the total number of documents divided by the number of documents containing the vocabulary. Multiply TF and IDF to get the TF-IDF value of each vocabulary. The higher the value, the more important and representative the vocabulary is in the chapter.

[0094] Further, according to a pre-set threshold or selecting the top-ranked TF-IDF values, the filtered vocabulary table is filtered to select these words as the characteristic words of the independent chapter. These characteristic words can highlight the core content of the chapter.

[0095] Specifically, the sentence is regarded as a graph composed of vocabulary nodes, and the edges between the nodes represent the co-occurrence relationship between the words.

[0096] Further, based on the filtered vocabulary table, the similarity between the words is calculated to determine the weight of the edge.

[0097] Further, the TextRank algorithm is used to iteratively update the score of each node (word), which reflects the importance of the word in the sentence.

[0098] Further, according to the score, a suitable segmentation point is set to segment the sentence into multiple segments. The position of the word with a higher score can be used as a potential segmentation point, so that the segmented segments can reflect the key semantic information of the sentence, and the segmented segments of the knowledge data are obtained.

[0099] Specifically, the feature sentences of the independent chapters and the segmented segments of the sentences are merged to form a set to be processed.

[0100] Further, for the repeated or similar sentences that may exist in the set, disambiguation processing is performed.

[0101] Further, whether these sentences refer to the same concept is determined through semantic analysis, context understanding and the like, if yes, the sentences are merged; if not, the uniqueness of the sentences is preserved.

[0102] Further, considering the professionalism and importance of the sentences in the power distribution network field, sentences that can best represent the core content of the power distribution network and have high recognition and differentiation are further screened, and finally these sentences are determined as the keywords of the power distribution network.

[0103] Further, these keywords can accurately summarize the key information of the power distribution network related content, and provide strong support for subsequent knowledge mining, information retrieval and the like.

[0104] Specifically, the text segment containing the keywords is input into the BERT model. The BERT model based on the pre-trained language representation will deeply encode each word in the input text, fully considering the context relationship between the words.

[0105] Further, when processing the keywords, it captures the semantic information of the keywords in the context of the previous and subsequent texts through a multi-layer bidirectional Transformer structure to obtain the context semantic feature vector of the keywords.

[0106] Further, in the result output by the model, the vector corresponding to the position of the keyword is the context semantic feature vector of the keyword, and this vector integrates the semantic information of the entire text segment, which can more accurately reflect the meaning of the keyword in a specific context.

[0107] Specifically, the context semantic feature vector of the keyword output by the BERT model is input into the TextCNN model.

[0108] Further, the TextCNN model performs convolution operation on the input vector through the convolution layer, and different size convolution kernels slide on the vector to extract semantic features in different local ranges.

[0109] Further, these convolution kernels act like filters that capture semantic information of different length segments near the keywords.

[0110] Further, after the convolution operation, the features are compressed by the pooling layer to retain the most representative features.

[0111] Further, the result output by the model is the local semantic feature vector of the keyword, which focuses on the local semantics around the keyword and highlights the key features of the keyword in the local context.

[0112] Specifically, the local semantic feature vector obtained by the TextCNN model is input into the Softmax function.

[0113] Further, the Softmax function calculates the probability value corresponding to each semantic category, which is pre-set and covers the semantic range related to the power distribution network field.

[0114] Further, the local semantic feature vector is mapped to a probability distribution, and the category with the highest probability value is considered as the most likely semantic category of the keyword.

[0115] Further, according to these semantic categories, the keywords are combined to generate key sentences.

[0116] Further, through the processing of the Softmax function, the local semantic features of the keywords are converted into meaningful sentences, which can accurately express the core semantics of the keywords in the context of the power distribution network.

[0117] Specifically, the key sentences are input into the embedding layer of the Bi-LSTM-CRF model.

[0118] Further, the embedding layer will split the key sentences into individual character units, and then map these characters to a low-dimensional vector space to form corresponding word vectors.

[0119] Further, this process is like giving each character a unique numerical code, and these codes combined together can represent the semantic information of the character in a specific context.

[0120] Further, in this way, the original text information is converted into a vector form that can be processed by computers, preparing for subsequent semantic extraction operations.

[0121] Specifically, the word vectors output by the embedding layer are input into the Bi-LSTM-CRF model's long short-term memory neural network (Bi-LSTM) layer. The Bi-LSTM layer is a powerful sequence model that can handle long-term dependencies in sequence data.

[0122] Further, in this layer, both forward and backward LSTM units process the word vector sequence simultaneously, considering not only the information of the current word but also the information of its context, thus capturing the semantic association between word vectors more comprehensively.

[0123] Further, after processing by the Bi-LSTM layer, each word vector is transformed into a semantic representation vector containing rich semantic information, which can more accurately reflect the semantics of each word in the key sentence in the overall context.

[0124] Specifically, the semantic representation vector output by the Bi-LSTM layer is input into the Conditional Random Field (CRF) layer of the Bi-LSTM-CRF model.

[0125] Specifically, the main function of the CRF layer is to classify and identify each part of the key sentence based on the semantic representation vector.

[0126] Further, it considers the context relationship between labels and assigns each semantic representation vector a most appropriate class label by learning the transition probability of labels in the sequence.

[0127] For example, in the power distribution network field, these class labels may represent device names, fault types, operation instructions, etc.

[0128] Further, through the processing of the CRF layer, each part of the key sentence is assigned a clear class label, completing the classification task of the key sentence.

[0129] In summary, entity keyword extraction on knowledge data can extract key information from complex power distribution network knowledge data, simplify massive data into core vocabulary, and facilitate subsequent processing.

[0130] For example, extracting keywords such as "transformer" and "fault current" can accurately locate related knowledge and improve data retrieval and processing efficiency.

[0131] In summary, semantic recognition and combination of keywords generate key sentences, further clarify the logic between keywords, and convert isolated words into semantically coherent sentences, such as "transformer failure causes current anomaly", making knowledge expression clearer and more accurate, facilitating model understanding and use of knowledge.

[0132] In summary, BERT model analyzes the context semantics of keywords, and TextCNN model extracts local semantic features, enabling the model to deeply understand the semantic connotation of keywords and their meanings in different contexts, grasp complex semantic relationships, and improve the accuracy of understanding of power distribution network professional terms.

[0133] For example, "short circuit" has different meanings in different devices and scenarios, which can be accurately identified by the model. This process enhances the model's understanding and expression of power distribution network knowledge, providing support for accurately answering user questions and handling complex business.

[0134] In summary, the key sentence classification identification obtains a category label, which provides a basis for knowledge data classification management and retrieval.

[0135] In summary, using the Bi-LSTM-CRF model, the key sentence is classified into categories such as "fault diagnosis" and "device maintenance". When the model receives real-time power distribution network problems from users, it can quickly match the problem and the corresponding category knowledge, locate the relevant key sentences and answers, and improve the problem processing speed and accuracy. For example, when a user asks a fault question, the model quickly searches for the answer from the "fault diagnosis" category.

[0136] S3, the category label, the knowledge graph, and the pre-acquired power distribution network operation rules are converted and stored by a large model text, obtaining a standard knowledge base of the power distribution network, and an AI semantic large model is constructed based on the standard knowledge base; In the embodiments of the present application, the category label, the knowledge graph, and the pre-acquired power distribution network operation rules are converted and stored by a large model text, obtaining a standard knowledge base of the power distribution network, comprising: The category label, the knowledge graph, and the pre-acquired power distribution network operation rules are collected as a data set for the power distribution network; The data set is converted by a large model text to obtain text data of the power distribution network; The text data is stored in a distributed manner to obtain a standard knowledge base of the power distribution network.

[0137] The AI semantic large model is constructed based on the standard knowledge base, comprising: A mapping relationship between the standard knowledge base and a SQL database is constructed; Based on the mapping relationship and the Text2SQL.AI tool, the standard knowledge base is converted to SQL to obtain the SQL statements of the standard knowledge base; The Seq2Seq model is used to train the semantics of the SQL statements to obtain the AI semantic large model of the standard knowledge base.

[0138] Specifically, the category label, the knowledge graph, and the pre-acquired power distribution network operation rules obtained in the previous steps are integrated.

[0139] Further, the category label represents the category to which the key sentence belongs, the knowledge graph presents various entities and their relationships in the power distribution network, and the power distribution network operation rules include device operation specifications, safety criteria, etc. Centralize these data of different forms and sources, organize them according to a certain structure, and obtain the text data of the power distribution network.

[0140] For example, a table containing multiple fields or a specific data structure is constructed, each field respectively stores the category label, the node and edge information of the knowledge graph, the operation rule clause, etc., thereby forming the data set of the power distribution network and providing a unified data basis for subsequent processing.

[0141] Specifically, for the collected power distribution network data set, a suitable large model text conversion method or tool is selected.

[0142] Further, if it is a large model based on natural language processing, the non-text data (such as the graph structure information of the knowledge graph) in the data set needs to be converted into text form, for example, the nodes and edges in the knowledge graph are represented by descriptive sentences.

[0143] Further, the professional terms and logical expressions in the category label and the operation rules are adjusted according to the format that the large model can understand.

[0144] Further, using specific conversion algorithms or functions, the entire data set is converted into text data that meets the input requirements of the large model, making the data consistent and readable, and facilitating the learning and processing of the large model.

[0145] Specifically, the converted power distribution network text data is stored using distributed storage technology. A suitable distributed storage system such as Ceph, GlusterFS, etc. is selected.

[0146] Further, according to the characteristics and configuration requirements of the storage system, the text data is divided into multiple parts and stored on different storage nodes.

[0147] Further, metadata is added to each stored data block to record the source, category, etc. of the data, facilitating subsequent retrieval and management.

[0148] Further, through distributed storage, not only the reliability and scalability of data storage are improved, but also efficient data access is achieved.

[0149] Further, these distributed stored data collectively constitute the standard knowledge base of the power distribution network, providing stable and persistent data support for the construction and application of AI semantic large models.

[0150] Specifically, the data structure of the standard knowledge base and the table structure design principles of the SQL database are analyzed in depth.

[0151] Further, the correspondence between various data elements in the standard knowledge base, such as entities, relationships, and category labels in the knowledge graph, and tables and fields in the SQL database is explicitly defined.

[0152] For example, the device entity in the knowledge graph is mapped to the "device table" in the SQL database, and the entity's attributes (such as device name and model) are mapped to the corresponding fields in the "device table".

[0153] Further, detailed mapping rules are developed for different types of data, including data type conversion (such as converting text category labels to appropriate character type fields in the database), and setting primary and foreign keys to reflect the association between data.

[0154] Further, these rules are configured in programs or related tools to establish a stable mapping relationship from the standard knowledge base to the SQL database, laying the foundation for subsequent data conversion.

[0155] Specifically, after the mapping relationship is established, the Text2SQL.AI tool is used to convert the SQL statements of the standard knowledge base. The data in the standard knowledge base is sorted according to the established mapping relationship, and the knowledge data is converted into SQL syntax-compliant query statements.

[0156] For example, if you want to query the devices in a specific area that are in a fault state, the tool will extract information from the corresponding table and field according to the mapping relationship, and generate a SQL statement similar to "SELECT device name FROM device table WHERE geographic location ='specific area' AND running status = 'fault'".

[0157] Further, the Text2SQL.AI tool will generate a series of statements for operating the SQL database according to various data query requirements in the knowledge base, combining mapping rules, to realize the automatic conversion from knowledge data to SQL statements and improve data processing efficiency.

[0158] Specifically, the converted SQL statements are input into the Seq2Seq model as training data. The Seq2Seq model consists of an encoder and a decoder. The encoder encodes the input SQL statement into a fixed-length semantic vector, which contains the semantic information of the SQL statement.

[0159] Further, the decoder generates the target text according to the semantic vector, which in this scenario is the natural language description or question answer corresponding to the SQL statement semantics.

[0160] Further, during the training process, the model's parameters are constantly adjusted through a large number of SQL statement samples, allowing the model to learn the semantic association between SQL statements and natural language.

[0161] Further, after multiple rounds of training, the model gradually learns to accurately understand the meaning of SQL statements and generate reasonable natural language outputs, ultimately forming an AI semantic large model for the power distribution network standard knowledge base. This model has the ability to convert user natural language questions into SQL queries and then convert the query results into natural language answers, achieving intelligent knowledge Q&A.

[0162] In summary, category labels, knowledge graphs, and power distribution network operation rules contain important information about different aspects of the power distribution network.

[0163] In summary, category labels classify key sentences for quick retrieval of specific types of knowledge; knowledge graphs display power distribution network entities and their relationships in a structured format, reflecting the system architecture; and operation rules define the guidelines for power distribution network operation.

[0164] In summary, by gathering this information and converting it into text data through a large model text conversion, and then storing it in a distributed manner to form a standard knowledge base, complex and scattered knowledge can be integrated to build a comprehensive, systematic, and easy-to-manage and use knowledge resource library, providing a rich and standardized data foundation for subsequent model construction.

[0165] In summary, the AI semantic large model based on the standard knowledge base can handle complex power distribution network problems.

[0166] In summary, since the knowledge base integrates multi-source knowledge, the model can quickly retrieve and match relevant information from a wealth of knowledge when facing problems such as multi-device fault diagnosis and complex operation and maintenance scenario decision-making, allowing for comprehensive analysis and reasoning.

[0167] In summary, by combining device relationships in the knowledge graph and operation rules, the model can determine the scope of the fault and possible causes, and quickly locate relevant knowledge categories using category labels, improving the accuracy and efficiency of problem-solving and meeting the complex needs of intelligent power distribution network management.

[0168] S4, obtain a user's real-time power distribution network problem, input the real-time power distribution network problem into the AI semantic large model, and obtain the answer to the real-time power distribution network problem.

[0169] In summary, the user's real-time problem is the "test data" for the model, and the model's performance can be reflected by its performance in solving these problems.

[0170] Overall, if the model frequently gives wrong or inaccurate answers, it indicates that the model lacks in knowledge understanding, reasoning ability, etc.

[0171] Overall, by collecting and analyzing the model's answers, the problem can be found and the model can be optimized accordingly, such as adjusting the training data, improving the algorithm, etc., to improve the model's ability to handle complex problems and make the model more perfect.

[0172] In the embodiment of the present application, the real-time power distribution network problem includes repair information, work order information, power distribution monitoring information, main line fault, branch line fault and transformer area monitoring.

[0173] Compared with the prior art, the present application has the following beneficial effects: 1. The AI semantic large model constructed by the present application can significantly improve the power distribution network problem processing capability. In the data processing and knowledge construction link, through comprehensive knowledge extraction and fusion of power distribution network data, an accurate knowledge graph is obtained; at the same time, key words are extracted from knowledge data, semantic recognition and classification are performed, and multi-source information is integrated into a standard knowledge base, which effectively integrates complex power distribution network information, provides a rich and accurate knowledge base for the model, and enables the model to quickly locate relevant knowledge when facing various power distribution network problems, laying a solid foundation for accurate problem solving.

[0174] 2. The AI semantic large model constructed by the present application has strong knowledge reserve and semantic understanding ability, and can efficiently process various real-time power distribution network problems. Whether it is repair information, work order information processing, or power distribution monitoring, fault diagnosis, etc., it can more deeply understand the semantic connotation of the problem, accurately match the relevant knowledge in the knowledge base, greatly improve the efficiency and accuracy of processing complex problems, and effectively improve the intelligent level of power distribution network management and operation and maintenance efficiency.

[0175] As Figure 2 shown is a functional module diagram of an AI semantic large model construction system of a power distribution network standard knowledge base according to an embodiment of the present application.

[0176] The AI semantic large model construction system 100 of the power distribution network standard knowledge base according to the present application can be installed in an electronic device. According to the functions to be realized, the AI semantic large model construction system 100 of the power distribution network standard knowledge base can include a knowledge graph generation module 101, a category label generation module 102, an AI semantic large model construction module 103, and an AI semantic large model application module 104. The modules according to the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete a fixed function, and are stored in the memory of the electronic device.

[0177] In the embodiment, the functions of the modules / units are as follows: The knowledge graph generation module 101 is configured to perform knowledge extraction on the pre-acquired power distribution network data to obtain knowledge data of the power distribution network, and perform knowledge fusion on the knowledge data to obtain a knowledge graph of the power distribution network. The category label generation module 102 is configured to perform entity keyword extraction on the knowledge data to obtain keywords of the power distribution network, perform semantic recognition and combination on the keywords to obtain key sentences of the keywords, and perform category recognition on the key sentences to obtain category labels of the key sentences. The AI semantic large model construction module 103 is configured to perform large model text conversion and storage on the category labels, the knowledge graph, and pre-acquired power distribution network operation rules to obtain a standard knowledge base of the power distribution network, and construct an AI semantic large model based on the standard knowledge base. The AI semantic large model application module 104 is configured to acquire a real-time power distribution network problem of a user, input the real-time power distribution network problem into the AI semantic large model, and obtain an answer to the real-time power distribution network problem.

[0178] In the several embodiments provided in the present application, it should be understood that the disclosed method and system can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the modules is only a logical function division. In actual implementation, other division manners can be adopted.

[0179] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, i.e., they may be located in one place or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0180] In addition, each functional module in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of hardware plus software functional modules.

[0181] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0182] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. The artificial intelligence is a theory, method, technology and application system for simulating, extending and expanding human intelligence by using a digital computer or a machine controlled by a digital computer, perceiving an environment, acquiring knowledge and using the knowledge to obtain optimal results.

[0183] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for constructing an AI semantic large model of a distribution network standard knowledge base, characterized in that: The method comprises: S1. Performing knowledge extraction on pre-acquired distribution network data to obtain knowledge data of the distribution network, and performing knowledge fusion on the knowledge data to obtain a knowledge graph of the distribution network; S2. Extract entity keywords from the knowledge data to obtain keywords for the power distribution network, perform semantic recognition and combination on the keywords to obtain key sentences for the keywords, classify and recognize the key sentences to obtain category labels for the key sentences; S3. Convert and store the category labels, the knowledge graph, and the pre-acquired distribution network operation rules into a large model text to obtain a standard knowledge base of the distribution network, and construct an AI semantic large model based on the standard knowledge base; S4. Obtain the user's real-time power distribution network problem, input the real-time power distribution network problem into the AI ​​semantic big model, and obtain the answer to the real-time power distribution network problem.

2. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The method of extracting knowledge from the pre-acquired distribution network data to obtain the knowledge data of the distribution network includes: Cleaning the pre-acquired distribution network data to obtain standardized distribution network data; Performing entity relationship separation on the standardized power distribution network data to obtain entity relationships of the standardized power distribution network data; Knowledge extraction is performed on the pre-acquired distribution network data based on the entity relationship to obtain knowledge data of the distribution network.

3. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The performing knowledge fusion on the knowledge data to obtain the knowledge graph of the distribution network includes: Performing entity alignment and disambiguation on the knowledge data to obtain knowledge entities; Multimodal knowledge fusion is performed on the knowledge entities to obtain a knowledge graph of the distribution network.

4. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The extracting entity keywords from the knowledge data to obtain keywords of the distribution network includes: Performing word segmentation and part-of-speech tagging on the knowledge data to obtain a vocabulary of the knowledge data; performing stop word filtering on the vocabulary to obtain a filtered vocabulary of the vocabulary; Extracting characteristic words and sentences from the knowledge data based on the TF-IDF algorithm and the filtering vocabulary to obtain characteristic words and sentences of the knowledge data; Performing key segmentation on the knowledge data based on the TextRank algorithm and the filtering vocabulary to obtain segmented segments of the knowledge data; The characteristic words and sentences and the segmented segments are fused and disambiguated to obtain keywords of the distribution network.

5. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The semantic recognition and combination of the keywords to obtain key sentences of the keywords includes: Performing contextual semantic analysis on the keyword using a BERT model to obtain a contextual semantic feature vector of the keyword; Using a TextCNN model to extract local semantic features from the contextual semantic feature vector to obtain a local semantic feature vector of the keyword; The Softmax function is used to perform semantic recognition and combination on the local semantic feature vectors to obtain the key sentences of the keywords.

6. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The classifying and identifying the key sentences to obtain the category labels of the key sentences includes: Using the embedding layer in the Bi-LSTM-CRF model to perform character splitting on the key sentence to obtain the word vector of the key sentence; Using the long short-term memory neural network layer in the Bi-LSTM-CRF model to perform semantic extraction on the word vector to obtain a semantic expression vector of the word vector; The conditional random field layer in the Bi-LSTM-CRF model is used to classify and identify the semantic expression vector to obtain the category label of the key sentence.

7. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The class labels, the knowledge graph, and the pre-acquired distribution network operation rules are converted into large-model text and stored to obtain a standard knowledge base of the distribution network, including: Gathering the category labels, the knowledge graph, and pre-acquired distribution network operation rules into a data set of the distribution network; Performing large model text conversion on the data set to obtain text data of the distribution network; The text data is distributedly stored to obtain a standard knowledge base of the power distribution network.

8. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The step of constructing an AI semantic model based on the standard knowledge base includes: Constructing a mapping relationship between the standard knowledge base and the SQL database; Performing SQL conversion on the standard knowledge base based on the mapping relationship and the Text2SQL.AI tool to obtain the SQL statement of the standard knowledge base; The SQL statements are semantically trained using a Seq2Seq model to obtain an AI semantic model of the standard knowledge base.

9. The method for constructing an AI semantic large model of a distribution network standard knowledge base according to claim 1, characterized in that: The real-time distribution network problems include: emergency repair information, work order information, distribution monitoring information, main line faults, branch line faults and substation monitoring.

10. An AI semantic large model construction system for a distribution network standard knowledge base, characterized in that: The system comprises: A knowledge graph generation module is used to extract knowledge from pre-acquired distribution network data to obtain knowledge data of the distribution network, and to perform knowledge fusion on the knowledge data to obtain a knowledge graph of the distribution network; a category label generation module, configured to extract entity keywords from the knowledge data to obtain keywords for the power distribution network, perform semantic recognition and combination on the keywords to obtain key sentences of the keywords, classify and recognize the key sentences, and obtain category labels for the key sentences; An AI semantic big model construction module is used to convert and store the category labels, the knowledge graph, and the pre-acquired distribution network operation rules into big model text, obtain a standard knowledge base for the distribution network, and construct an AI semantic big model based on the standard knowledge base; The AI ​​semantic big model application module is used to obtain the user's real-time power distribution network problem, input the real-time power distribution network problem into the AI ​​semantic big model, and obtain the answer to the real-time power distribution network problem.

Images