Electric power knowledge semantic analysis system and method

Abstract

A power knowledge semantic analysis system and method thereof construct a power grid physical model graph, a power grid operation data graph and a dispatching semantic graph, and fuse the power grid physical model graph, the power grid operation data graph and the dispatching semantic graph to obtain a power grid knowledge graph; through the knowledge graph, various structured and unstructured texts are quickly processed to extract knowledge, and the knowledge is digested and sorted to form a reasonable knowledge structure; intelligent question and answer of power grid knowledge is realized based on the power grid knowledge graph, thereby reducing the cost of knowledge induction and learning.

Description

Technical Field

[0001] The present application relates to the field of intelligent analysis technology, and more specifically, to an electric power knowledge semantic analysis system and method thereof. Background Art

[0002] At present, the number of centralized new energy stations has reached 848 by the end of 2021, belonging to more than 400 different corporate entities; the number of distributed new energy stations has exceeded 80,000, belonging to more than 3,000 investment institutions (excluding individual household owners), and the new power system's generation, transmission, transformation, distribution, and use are coupled with each other, which will inevitably form a huge knowledge system, which directly leads to the difficulty in controlling and analyzing massive operating data.

[0003] The traditional method of increasing the training of dispatching and operating personnel to achieve the recognition and application of a huge knowledge system is time-consuming and economically costly, and the improvement of cognitive ability is limited, making it difficult to adapt to the development of the new situation. Summary of the invention

[0004] In order to solve the above technical problems, this application is proposed. The embodiment of this application provides a power knowledge semantic analysis system and method thereof, which constructs a power grid physical model map, a power grid operation data map and a scheduling semantic map, and integrates the power grid physical model map, the power grid operation data map and the scheduling semantic map to obtain a power grid knowledge map; through the knowledge map, various structured and unstructured texts are quickly processed, knowledge is extracted, and it is digested and sorted to form a reasonable knowledge structure; intelligent question and answer of power grid knowledge is realized based on the power grid knowledge map, thereby reducing the cost of knowledge induction and learning.

[0005] In a first aspect, a method for semantic analysis of electric power knowledge is provided, which includes:

[0006] Construct the power grid physical model map, power grid operation data map and dispatch semantic map;

[0007] The power grid physical model graph, the power grid operation data graph and the scheduling semantic graph are integrated to obtain a power grid knowledge graph;

[0008] Get electricity knowledge questions;

[0009] After performing sentence recognition and entity recognition on the power knowledge question, the question is input into a semantic parsing model to obtain a semantic understanding feature vector of the power knowledge question;

[0010] The power grid knowledge graph is passed through a graph neural network model to obtain a power grid semantic knowledge graph feature matrix;

[0011] Taking the power knowledge question semantic understanding feature vector as the query feature vector, calculating the product between the query feature vector and the power grid semantic knowledge graph feature matrix to obtain a classification feature vector;

[0012] Performing feature distribution optimization on the classification feature vector to obtain an optimized classification feature vector; and

[0013] The optimized classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to represent a sequence label of the candidate answer.

[0014] In the above-mentioned power knowledge semantic analysis method, the semantic parsing model is a Bert model including a word embedding layer.

[0015] In the above-mentioned electric power knowledge semantic analysis method, after sentence recognition and entity recognition are performed on the electric power knowledge question, the question is input into a semantic parsing model to obtain a semantic understanding feature vector of the electric power knowledge question, including: sentence recognition and entity recognition are performed on the electric power knowledge question to obtain multiple question components, wherein the question components include concepts, relations and attributes; the multiple question components are word-embedded using the word embedding layer of the semantic parsing model to obtain multiple question component word embedding vectors; and, the multiple question component word embedding vectors are contextually semantically encoded using the Bert model of the semantic parsing model to obtain the semantic understanding feature vector of the electric power knowledge question.

[0016] In the above-mentioned electric power knowledge semantic analysis method, the Bert model of the semantic parsing model is used to perform context semantic encoding on the multiple question component word embedding vectors to obtain the electric power knowledge question semantic understanding feature vector, including: one-dimensionally arranging the multiple question component word embedding vectors to obtain a global word embedding feature vector; calculating the product between the global word embedding feature vector and the transposed vector of each question component word embedding vector in the multiple question component word embedding vectors to obtain multiple self-attention association matrices; respectively normalizing each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; passing each standardized self-attention association matrix in the multiple standardized self-attention association matrices through the Softmax classification function to obtain multiple probability values; and, respectively weighting each question component word embedding vector in the multiple question component word embedding vectors using each probability value in the multiple probability values ​​as a weight to obtain the electric power knowledge question semantic understanding feature vector.

[0017] In the above-mentioned power knowledge semantic analysis method, the feature distribution of the classification feature vector is optimized to obtain an optimized classification feature vector, including: the feature distribution of the classification feature vector is optimized to obtain an optimized classification feature vector according to the following optimization formula; wherein the optimization formula is:

[0018] v′ i =(μσ)v i 2 +v i μ+(v i -σ)μ 2

[0019] Wherein, V represents the classification feature vector, μ and σ are the mean and standard deviation of the feature value set at all positions in the classification feature vector, and v i is the eigenvalue of the ith position of the classification feature vector, v′ i is the eigenvalue of the i-th position of the optimized classification feature vector.

[0020] In the above-mentioned electric power knowledge semantic analysis method, the optimized classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to represent a sequence label of an alternative answer, including: using multiple fully connected layers of the classifier to fully connect encode the optimized classification feature vector to obtain an encoded classification feature vector; and, passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0021] In a second aspect, a power knowledge semantic analysis system is provided, which includes:

[0022] Graph component module, used to construct power grid physical model graph, power grid operation data graph and dispatch semantic graph;

[0023] A graph fusion module, used for fusing the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph;

[0024] The power knowledge acquisition module is used to obtain power knowledge questions;

[0025] A semantic parsing module, which is used to perform sentence recognition and entity recognition on the power knowledge question and then input the semantic parsing model to obtain a semantic understanding feature vector of the power knowledge question;

[0026] A graph neural network module, used for passing the power grid knowledge graph through a graph neural network model to obtain a power grid semantic knowledge graph feature matrix;

[0027] A vector calculation module, used for taking the power knowledge question semantic understanding feature vector as a query feature vector, and calculating the product between the query feature vector and the power grid semantic knowledge graph feature matrix to obtain a classification feature vector;

[0028] an optimization module, configured to optimize the feature distribution of the classification feature vector to obtain an optimized classification feature vector; and

[0029] The module for generating sequence labels of candidate answers is used to pass the optimized classification feature vector through a classifier to obtain a classification result, and the classification result is used to represent the sequence labels of the candidate answers.

[0030] In the above-mentioned electric power knowledge semantic analysis system, the semantic parsing model is a Bert model including a word embedding layer.

[0031] In the above-mentioned electric power knowledge semantic analysis system, the semantic parsing module includes: a recognition unit, which is used to perform sentence recognition and entity recognition on the electric power knowledge question to obtain multiple question components, wherein the question components include concepts, relations and attributes; a word embedding unit, which is used to use the word embedding layer of the semantic parsing model to word embed the multiple question components to obtain multiple question component word embedding vectors; and a context encoding unit, which is used to perform context semantic encoding on the multiple question component word embedding vectors using the Bert model of the semantic parsing model to obtain the electric power knowledge question semantic understanding feature vector.

[0032] In the above-mentioned electric power knowledge semantic analysis system, the context encoding unit includes: a one-dimensional arrangement subunit, which is used to arrange the multiple question component word embedding vectors in one dimension to obtain a global word embedding feature vector; a self-attention subunit, which is used to calculate the product between the global word embedding feature vector and the transposed vector of each question component word embedding vector in the multiple question component word embedding vectors to obtain multiple self-attention association matrices; a normalization subunit, which is used to standardize each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; a probability value calculation subunit, which is used to pass each standardized self-attention association matrix in the multiple standardized self-attention association matrices through the Softmax classification function to obtain multiple probability values; and a weighting subunit, which is used to weight each question component word embedding vector in the multiple question component word embedding vectors with each probability value in the multiple probability values ​​as a weight to obtain the electric power knowledge question semantic understanding feature vector.

[0033] Compared with the prior art, the electric power knowledge semantic analysis system and method provided in the present application construct a power grid physical model graph, a power grid operation data graph and a scheduling semantic graph, and integrate the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph; through the knowledge graph, various structured and unstructured texts are quickly processed, knowledge is extracted, and it is digested and organized to form a reasonable knowledge structure; intelligent question and answer of power grid knowledge is realized based on the power grid knowledge graph, thereby reducing the cost of knowledge induction and learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 A schematic diagram for constructing a scheduling business knowledge graph according to an embodiment of the present application.

[0036] Figure 2 It is a schematic diagram of the fusion of the power grid ontology graph and the scheduling semantic graph according to an embodiment of the present application.

[0037] Figure 3 Schematic diagram of a question-and-answer service based on a scheduling knowledge graph according to an embodiment of the present application.

[0038] Figure 4 This is a schematic diagram of an intelligent operation reasoning service based on a scheduling knowledge graph according to an embodiment of the present application.

[0039] Figure 5 A schematic diagram of intelligent training according to an embodiment of the present application.

[0040] Figure 6 It is a control intelligent agent application according to an embodiment of the present application.

[0041] Figure 7 A schematic diagram of a scenario of a method for semantic analysis of electric power knowledge according to an embodiment of the present application.

[0042] Figure 8 Flow chart of a method for semantic analysis of electric power knowledge according to an embodiment of the present application.

[0043] Fig. 9 Schematic diagram of the architecture of the power knowledge semantic analysis method according to an embodiment of the present application.

[0044] Fig.10 It is a flowchart of the sub-steps of step 140 in the power knowledge semantic analysis method according to an embodiment of the present application.

[0045] Fig.11 It is a flowchart of the sub-steps of step 143 in the power knowledge semantic analysis method according to an embodiment of the present application.

[0046] Fig.12 It is a flowchart of the sub-steps of step 180 in the power knowledge semantic analysis method according to an embodiment of the present application.

[0047] Fig.13 4 is a block diagram of an electric power knowledge semantic analysis system according to an embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0049] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present application have the same meaning as those commonly understood by those skilled in the art of the present application. The terms used in this application are only for the purpose of describing specific embodiments and are not intended to limit the scope of this application.

[0050] In the embodiments of the present application, it should be noted that, unless otherwise specified and limited, the term "connection" should be understood in a broad sense. For example, it can be an electrical connection or a connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0051] It should be noted that the terms "first\second\third" involved in the embodiments of the present application are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that the specific order or sequence of "first\second\third" can be interchanged where permitted. It should be understood that the objects distinguished by "first\second\third" can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0052] The objectives of this application are: (1) to build a new type of dispatching AI cognitive service, realize the graph-based knowledge access, knowledge modeling and generation of 220kV and above power grid models and real-time data, and build a power grid ontology knowledge graph; to realize the graph-based generation of dispatching semantic documents, mainly including "Electric Power Dispatching Management Regulations", "Dispatching Operation Management Regulations", "High Voltage DC System Dispatching Operation Management Regulations" and other power grid operation procedures, and build a dispatching semantic knowledge graph; (2) to realize the integration of the power grid ontology knowledge graph and the dispatching semantic knowledge graph, and establish an update mechanism to realize the timely update of the power grid model and real-time operation data graph knowledge; (3) to build a machine reading and intelligent question-answering service for dispatching semantic documents, and build the pre-training capability, machine reading capability, automatic question generation capability and graph knowledge-based question-answering capability of power grid operation procedure documents such as "Electric Power Dispatching Management Regulations", "Dispatching Operation Management Regulations", and "High Voltage DC System Dispatching Operation Management Regulations". (4) Construct a reasoning service based on the dispatching knowledge graph. Based on the "one graph for power grid dispatching" that integrates the power grid ontology knowledge graph and the dispatching semantic knowledge graph, realize graph reasoning services such as operation rule graph generation, power grid operation status reasoning, and equipment connectivity reasoning analysis; (5) Construct a power grid operation reinforcement learning environment; realize dynamic topology analysis and AC / DC flow calculation of power grids of 220kV and above; realize the deep combination of optimization and solution of MIP / LP / NLP capabilities and reinforcement learning; (6) Establish a finite Markov decision process model for the interaction of "intelligent agent-power grid digital twin environment". Support changing custom training goals: It can support intelligent agent training for multiple scenarios such as real-time output dispatch of units, section over-limit recovery, reactive power optimization, medium- and long-term power safety verification, spot clearing unit combination, and maximum consumption of new energy based on historical data. Support system load balance, upper and lower limits of unit output, unit climbing constraints, line flow constraints, section flow constraints, etc. as constraints; support setting of line, unit, transformer faults and fault probability, and set disturbances in the training process through custom faults.

[0053] In this application, with the accelerated construction of new power systems, the proportion of new energy access has increased significantly. Its volatility, randomness and weak support have increased the uncertainty of the power grid, and the traditional model is difficult to achieve multi-objective-oriented dispatching and control. The problem is becoming more and more serious; in addition, the links between various departments in the power industry are becoming closer and closer, and the coupling of multiple structures of generation, transmission, transformation, distribution and use has formed a huge knowledge system. The traditional reliance on manpower to achieve the cognition of power knowledge can no longer meet the requirements of power grid dispatching. Therefore, this application proposes a method for constructing a power dispatching knowledge graph model, and establishes a new dispatching AI cognitive service platform, which is intended to solve the problem that the traditional reliance on manpower to achieve the cognition of power knowledge can no longer meet the requirements of power grid dispatching; it proposes to develop a power grid operation reinforcement learning environment based on OCS power system models and real-time data, power system analysis and simulation software, and openAI gym / Ray and other frameworks, which is intended to solve the problem that traditional decision-making models are difficult to achieve multi-objective-oriented dispatching and control.

[0054] For example, in a specific example of this application, the number of centralized new energy stations has reached 848 by the end of 2021, belonging to more than 400 different corporate entities; the number of distributed new energy stations has exceeded 80,000, affiliated with more than 3,000 investment institutions (excluding individual household owners), and the new power system generation, transmission, transformation, distribution, and use of multiple structures are coupled with each other, which will inevitably form a huge knowledge system, which directly leads to the difficulty of controlling and analyzing massive operating data. However, the traditional method of increasing the training of dispatching and operating personnel to achieve the cognition and application of the huge knowledge system has high time and economic costs, and the improvement of cognitive ability is limited, making it difficult to adapt to the development of the new situation. Therefore, it is necessary to explore new technical means to support operators to build new cognitive service capabilities.

[0055] It should be understood that knowledge graphs can quickly process various structured and unstructured texts, extract knowledge, and digest and organize it to form a reasonable knowledge structure, which greatly reduces the cost of knowledge induction and learning; based on knowledge graphs, functions such as power system control decision-making and automated duty intelligent question and answer can be further realized. Therefore, this application studies the application of knowledge graph technology in the field of power control, which is described as follows: (1) Research on the construction technology of dispatching business knowledge graphs: Research on the construction mode of knowledge graph models in the dispatching field, and construct a power dispatching knowledge graph model based on the power system model (IEC61970), real-time data and dispatching semantics (technical documents), and build a new dispatching AI cognitive service platform, including a knowledge base, an analysis AI algorithm library, etc. Figure 1 As shown in the figure, the problem of difficult automatic generation of knowledge of structured and unstructured data in dispatching business is solved. (2) Research on the fusion technology of power grid ontology graph and dispatching semantic graph: Research on the fusion technology of knowledge graph model in the dispatching field, realize the fusion of power system model graph, real-time data graph and dispatching semantic graph, and form a dispatching business graph. Figure 2As shown, the problem of knowledge existing independently and unable to establish connections for collaborative applications is solved. (3) Research on question-answering services based on the dispatch knowledge graph: Based on the dispatch AI cognitive service platform, research on the dispatch document automatic understanding, storage, training and question-answering technology based on the four-in-one algorithm DKEPLM (Dispatch Knowledge-Enhanced Pretrain Language Model), DMRC (Dispatch Machine Reading Comprehension), DPR (Dispatch Passage Ranking), and L2AD (Learn to Ask for Dispatch) to realize dispatch document machine reading and intelligent question-answering services. Figure 3 As shown, the problem of huge business data not being able to be easily applied is solved. (4) Research on reasoning services based on dispatching knowledge graphs: Based on the cognitive ability of the dispatching knowledge graph on power grid operation, combined with the business logic of power grid operation error prevention and power grid operation step generation, research on reasoning services based on dispatching knowledge graphs to support intelligent power grid operation error prevention and intelligent generation of operation tickets, such as Figure 4 shown.

[0056] It is worth mentioning that the randomness of renewable energy generation is high. The massive access to renewable energy will inevitably lead to problems such as the continuous decrease in system inertia, reduced ability to resist faults, more complex fault forms, and increased chain reaction risks, which greatly increases the uncertainty of power grid operation and makes the control problem more serious. Traditional single dispatching decisions can no longer achieve optimal control of the system, requiring joint analysis and decision-making of multiple systems, which cannot be achieved by simple human calculations alone. Therefore, it is necessary to study new technical means to assist operators in efficient and stable power grid control.

[0057] Application research in various industries has proven that reinforcement learning can optimize its own control strategy through autonomous learning through neural networks. It does not require precise objective functions; it can make real-time online decisions; it has stronger robustness and stable convergence results. It can quickly solve complex coupling problems. Based on reinforcement learning, it can further realize functions such as power system control decision-making and fault plan drills. The detailed description is as follows: (1) Use OCS system data and BPA tools to provide power system simulation data, simulate power system models under multiple scenarios, so that the intelligent agent can continuously learn and optimize its own control strategy, and establish a highly robust control intelligent agent, such as Figure 5(2) The control agent proposes appropriate control decisions according to the power grid conditions, assisting the dispatcher to complete the dispatch work more easily. When facing special scenarios such as typhoons, on the premise of maintaining the stability of the power grid, the optimal control plan suitable for different scenarios is obtained, and the automatic generation of agents in different scenarios is realized, such as Figure 6 .

[0058] In response to the above technical needs, the technical concept of this application is to construct a power grid physical model graph, a power grid operation data graph and a scheduling semantic graph, and integrate the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph. The knowledge graph can quickly process various structured and unstructured texts, extract knowledge, and digest and organize it to form a reasonable knowledge structure, which greatly reduces the cost of knowledge induction and learning. Furthermore, based on the power grid knowledge graph, a question-and-answer solution based on graph knowledge is constructed.

[0059] Accordingly, in the technical solution of the present application, a power grid physical model map, a power grid operation data map and a dispatching semantic map are first constructed. Specifically, the power grid physical model map is constructed through the power grid model knowledge document, the power grid operation data map is constructed through the graphed knowledge access of the power grid operation real-time data, and the dispatching semantic map is obtained through the graphing of the power grid operation procedure document.

[0060] The grid physical model diagram is the basis for describing the relationship between various physical elements in the power system, including generators, substations, distribution stations, etc. Understanding the characteristics and topological structures of these elements can help better understand the operation mechanism of the power system, thereby improving the understanding and grasp of power knowledge.

[0061] The grid operation data map is the key to describing the operating status of the power system, including parameters such as voltage, current, and power. Monitoring and analyzing these data can help to promptly discover problems and abnormal situations, and take appropriate measures to handle and repair them. The dispatch semantic map contains various dispatch rules and strategies, which is an important means to ensure the safe and stable operation of the power system. By in-depth research and learning of these rules and strategies, it can provide reference for system optimization and improvement, and improve the operating efficiency and quality of the power system.

[0062] Next, the grid physical model graph, the grid operation data graph and the dispatch semantic graph are integrated to obtain the grid knowledge graph. Here, by integrating the grid physical model graph, the grid operation data graph and the dispatch semantic graph, the problem of independent existence of knowledge and inability to establish connection and collaborative application can be solved. Specifically, the integration of the grid physical model graph, the grid operation data graph and the dispatch semantic graph is to obtain a more comprehensive and accurate grid knowledge graph, wherein the three sub-graphs respectively describe different aspects of knowledge information such as the physical structure, real-time operation status and dispatch procedures in the power system. By integrating them into a whole, the shortcomings of each in certain aspects can be made up, and the completeness and practicality of the grid knowledge graph can be further improved. Specifically, by integrating the grid physical model graph and the grid operation data graph, the connection between power equipment can be established, and it can be combined with real-time operation data to better understand the coupling relationship and influencing factors between each device, so as to achieve comprehensive monitoring and prediction of the operation of the power system. At the same time, by integrating the dispatch semantic graph, the power dispatch instruction can also be linked to the specific equipment operation to realize intelligent grid operation management.

[0063] In a specific example of the present application, the process of fusing the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph includes:

[0064] First, semantic modeling is performed on the power grid physical model graph, power grid operation data graph, and dispatch semantic graph, and the entities, attributes, and relationships therein are converted into a unified knowledge representation form. Then, a fusion strategy is designed to match and integrate the same entities, attributes, and relationships in the three sub-graphs. For example, the concept of a transmission line connecting two substations is involved in the power grid physical model graph, the power grid operation data graph, and the dispatch semantic graph. They can be matched into the same entity based on their commonalities. Then, based on the fusion strategy, the three sub-graphs are integrated into a comprehensive power grid knowledge graph. This process requires certain data cleaning and redundant information elimination processing to ensure that the structure of the obtained knowledge graph is reasonable, complete, and accurate. Finally, the knowledge graph is stored in a suitable graph database, and is indexed and optimized to support efficient query and retrieval of the intelligent question-answering system.

[0065] Next, obtain electricity knowledge questions. Here, obtaining electricity knowledge questions is to enable the intelligent question-answering system to understand and respond to user needs. In actual application scenarios, users may have different questions and needs about various aspects of the power system, such as equipment failures, operating status, dispatch instructions, etc. By obtaining these electricity knowledge questions, corresponding answers can be provided according to the specific needs of users, thereby meeting the user's information query needs. At the same time, the power system is a complex field that covers a large number of professional terms and knowledge points. If the user does not have relevant field knowledge, it is difficult to accurately describe his or her problem. Therefore, by obtaining electricity knowledge questions, users can better express their needs and obtain more accurate answers.

[0066] Next, after performing sentence recognition and entity recognition on the electricity knowledge question, the question is input into the semantic parsing model to obtain the semantic understanding feature vector of the electricity knowledge question. The purpose of performing sentence recognition and entity recognition on the electricity knowledge question is to convert the question expressed in natural language into a structured form that can be processed by the machine. In the electricity knowledge intelligent question-answering system, users may ask questions in various natural language forms, and these questions need to be converted into structured data that can be processed by computers before subsequent processing and analysis can be performed.

[0067] Specifically, sentence recognition and entity recognition for questions about power knowledge can split the question text into components such as concepts, relationships, and attributes, so as to better understand the intent and content of the question. For example, for a question involving power equipment failure, "Which equipment has failed?" can be identified as a query statement containing entities such as "equipment" and "failure", and further converted into a corresponding structured query. Through entity recognition and sentence recognition, the question can be deconstructed into basic semantic units, laying the foundation for subsequent semantic understanding and calculation. After obtaining the structured question after sentence recognition and entity recognition, it is necessary to input the semantic parsing model for deep semantic understanding to obtain the semantic understanding feature vector of the power knowledge question. This feature vector can represent the semantic meaning of the question text and is used for subsequent knowledge matching and classification. Through the processing of the semantic understanding model, the intelligent question-answering system can improve its ability to understand questions, so as to more accurately answer the power knowledge questions raised by users.

[0068] Then, the power grid knowledge graph is passed through the graph neural network model to obtain the power grid semantic knowledge graph feature matrix. Here, the power grid knowledge graph is passed through the graph neural network model to obtain the power grid semantic knowledge graph feature matrix in order to convert the information in the knowledge graph into a vector representation that can be processed and analyzed by a computer.

[0069] Traditional rule-based question-answering systems mainly rely on manually written rules and templates, which are difficult to handle large-scale and complex questions. Graph neural network technology can automatically learn the representation of nodes and edges from structured data, and perform graph-level calculations and reasoning on this basis. By inputting the power grid knowledge graph into the graph neural network model, the relationship between each entity in the graph and its embedded representation can be learned, so as to better understand the semantic relationship between these entities.

[0070] Specifically, by calculating the power grid knowledge graph through the graph neural network model, we can obtain the power grid semantic knowledge graph feature matrix, in which each entity is represented as a vector. This vector contains multi-dimensional features such as the entity's position, connection relationship, and semantic information in the power grid knowledge graph, which can be used for subsequent knowledge matching and classification tasks. At the same time, since the graph neural network has excellent processing effect on local subgraphs, it can also capture some global properties, such as important nodes and critical paths in the power system.

[0071] Next, the semantic understanding feature vector of the electric power knowledge question is used as the query feature vector, and the product between it and the feature matrix of the power grid semantic knowledge graph is calculated to obtain a classification feature vector. Here, the semantic understanding feature vector of the electric power knowledge question is used as the query feature vector, and the product between it and the feature matrix of the power grid semantic knowledge graph is calculated to obtain a classification feature vector, in order to establish a connection between the electric power knowledge question and the power grid knowledge graph, so as to realize question classification and answering.

[0072] In the intelligent question-answering system for electric power knowledge, questions raised by users need to be classified through a classification model to determine the type of question and the required answer information. Therefore, the query text needs to be converted into a computer-processable form and matched and compared with the power grid knowledge graph to find the best question classification and answer. Specifically, the semantic understanding feature vector of electric power knowledge questions is used as the query feature vector, and the question text can be represented as a multidimensional vector. This vector contains important information and semantic features of the question text, which can be used for subsequent question classification and answering. The feature matrix of the power grid semantic knowledge graph represents the vector representation of each entity in the knowledge graph, which can be used for knowledge matching and classification tasks.

[0073] By calculating the dot product or cosine similarity between the query feature vector and the feature matrix of the power grid semantic knowledge graph, a classification feature vector can be obtained. This vector can represent the relevance and matching degree between the current question and the corresponding entity in the power grid knowledge graph, thereby determining the best question classification and answer.

[0074] Finally, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to represent the sequence label of the candidate answer. That is, the classification feature vector is input into the classifier to determine the sequence label of the answer to the power knowledge question through the classifier. In this way, intelligent question and answer of power grid knowledge is realized based on the power grid knowledge graph.

[0075] In particular, in the technical solution of the present application, when the semantic understanding feature vector of the electric power knowledge question is used as the query feature vector and the product between it and the feature matrix of the power grid semantic knowledge graph is calculated to obtain the classification feature vector, the problem semantic information represented by the semantic understanding feature vector of the electric power knowledge question is essentially mapped to the high-dimensional semantic feature space of the power grid knowledge graph. Because the feature matrix of the power grid semantic knowledge graph is generated by a graph neural network model and the semantic understanding feature vector of the electric power knowledge question is generated by a semantic encoder, the feature matrix of the power grid semantic knowledge graph has a non-Euclidean topological structure, while the semantic understanding feature vector of the electric power knowledge question is still in the Euclidean space. Therefore, when the matrix product between the semantic understanding feature vector of the electric power knowledge question and the feature matrix of the power grid semantic knowledge graph is performed, the mismatch of the spatial structure paradigm will cause the local distribution of the classification feature vector to be mismatched, which will lead to the irregularity of the feature distribution of the classification feature vector, thereby affecting the accuracy of the classification results through the classifier.

[0076] Based on this, the applicant of the present application performs a second regularization of the Gaussian probability density parameters of the manifold surface on the classification feature vector V, which is specifically expressed as follows:

[0077] v′ i =(μσ)v i 2 +v i μ+(v i -σ)μ 2

[0078] where μ and σ are the eigenvalue sets v i ∈V, and v′ i is the eigenvalue of the i-th position of the optimized classification feature vector V′.

[0079] Specifically, in order to solve the problem of irregular distribution of high-dimensional feature distribution of the feature set of the classification feature vector V in the high-dimensional feature space of the query feature map, quadratic regularization is performed on each eigenvalue of the classification feature vector V through the likelihood of the Gaussian probability density parameter of the class probability distribution of the classifier, so as to smooth the eigenvalues ​​based on the equidistant distribution of the Gaussian probability density parameter in the parameter space of the target class probability, so as to obtain the regularized reorganization of the original probability density likelihood function expressed by the manifold surface of the high-dimensional feature in the parameter space, thereby improving the regularity of the feature distribution of the optimized classification feature vector V′, so as to improve the classification accuracy of the optimized classification feature vector V′ through the classifier.

[0080] Through the above-mentioned question-and-answer solution based on graph knowledge, the following technical effects can be achieved: 1) Power dispatching is a very complex field that requires processing of various data and information. The use of knowledge graphs and question-and-answer services can better integrate this information and provide users with faster and more accurate answers. 2) The power industry requires efficient solutions to meet customer needs and improve economic benefits. By providing algorithm-based technologies, enterprises can process information faster and more accurately, thereby improving production efficiency and business competitiveness. 3) With the development of society and the advancement of science and technology, people's requirements for the power industry are also constantly increasing. The use of technologies such as knowledge graphs and question-and-answer services can improve the intelligence level of enterprises, better meet customer needs, and improve the competitiveness of enterprises in the market. 4) As an important infrastructure, power dispatching requires professional skills and knowledge for its management and maintenance. The use of algorithm-based technologies can better manage and store this information, and can help cultivate more professional talents in the power industry.

[0081] Figure 7 FIG. 1 is a schematic diagram of a scenario of a method for semantic analysis of electric power knowledge according to an embodiment of the present application. Figure 7 As shown, in this application scenario, first, a grid physical model map is constructed (for example, Figure 7 C1 as shown in the figure), power grid operation data map (for example, Figure 7 C2) and the scheduling semantic graph (e.g., Figure 7 C3 shown in the figure), and questions to obtain electricity knowledge (e.g. Figure 7 Then, the obtained power grid physical model map, power grid operation data map, dispatch semantic map and power knowledge question are input into a server (for example, Figure 7In S) shown in , the server is capable of processing the power grid physical model map, the power grid operation data map, the scheduling semantic map and the power knowledge question value based on the power knowledge semantic analysis algorithm to generate a classification result of a sequence label for representing alternative answers.

[0082] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0083] In one embodiment of the present application, Figure 8 FIG. 1 is a flow chart of a method for semantic analysis of electric power knowledge according to an embodiment of the present application. Figure 8 As shown, according to the embodiment of the present application, the power knowledge semantic analysis method 100 includes: 110, constructing a power grid physical model map, a power grid operation data map and a scheduling semantic map; 120, integrating the power grid physical model map, the power grid operation data map and the scheduling semantic map to obtain a power grid knowledge map; 130, obtaining power knowledge questions; 140, after performing sentence recognition and entity recognition on the power knowledge questions, inputting the semantic parsing model to obtain a power knowledge question semantic understanding feature vector; 150, passing the power grid knowledge map through a graph neural network model to obtain a power grid semantic knowledge map feature matrix; 160, using the power knowledge question semantic understanding feature vector as a query feature vector, calculating the product between it and the power grid semantic knowledge map feature matrix to obtain a classification feature vector; 170, performing feature distribution optimization on the classification feature vector to obtain an optimized classification feature vector; and, 180, passing the optimized classification feature vector through a classifier to obtain a classification result, and the classification result is used to represent a sequence label of alternative answers.

[0084] Fig. 9 FIG. 1 is a schematic diagram of the architecture of the power knowledge semantic analysis method according to an embodiment of the present application. Fig. 9As shown, in the network architecture, first, a power grid physical model graph, a power grid operation data graph and a scheduling semantic graph are constructed; then, the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph are integrated to obtain a power grid knowledge graph; then, power knowledge questions are obtained; then, after sentence recognition and entity recognition are performed on the power knowledge questions, they are input into a semantic parsing model to obtain a power knowledge question semantic understanding feature vector; then, the power grid knowledge graph is passed through a graph neural network model to obtain a power grid semantic knowledge graph feature matrix; then, the power knowledge question semantic understanding feature vector is used as a query feature vector, and its product with the power grid semantic knowledge graph feature matrix is ​​calculated to obtain a classification feature vector; then, the classification feature vector is subjected to feature distribution optimization to obtain an optimized classification feature vector; and finally, the optimized classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to represent a sequence label of alternative answers.

[0085] Specifically, in step 110, a power grid physical model graph, a power grid operation data graph, and a scheduling semantic graph are constructed. In response to the above technical requirements, the technical concept of the present application is to construct a power grid physical model graph, a power grid operation data graph, and a scheduling semantic graph, and fuse the power grid physical model graph, the power grid operation data graph, and the scheduling semantic graph to obtain a power grid knowledge graph. The knowledge graph can quickly process various structured and unstructured texts, extract knowledge, and digest and organize it to form a reasonable knowledge structure, which greatly reduces the cost of knowledge induction and learning. Furthermore, a question-and-answer solution based on graph knowledge is constructed based on the power grid knowledge graph.

[0086] Accordingly, in the technical solution of the present application, a power grid physical model map, a power grid operation data map and a dispatching semantic map are first constructed. Specifically, the power grid physical model map is constructed through the power grid model knowledge document, the power grid operation data map is constructed through the graphed knowledge access of the power grid operation real-time data, and the dispatching semantic map is obtained through the graphing of the power grid operation procedure document.

[0087] The grid physical model diagram is the basis for describing the relationship between various physical elements in the power system, including generators, substations, distribution stations, etc. Understanding the characteristics and topological structures of these elements can help better understand the operation mechanism of the power system, thereby improving the understanding and grasp of power knowledge.

[0088] The grid operation data map is the key to describing the operating status of the power system, including parameters such as voltage, current, and power. Monitoring and analyzing these data can help to promptly discover problems and abnormal situations, and take appropriate measures to handle and repair them. The dispatch semantic map contains various dispatch rules and strategies, which is an important means to ensure the safe and stable operation of the power system. By in-depth research and learning of these rules and strategies, it can provide reference for system optimization and improvement, and improve the operating efficiency and quality of the power system.

[0089] Specifically, in step 120, the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph are fused to obtain a power grid knowledge graph. Next, the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph are fused to obtain a power grid knowledge graph. Here, by fusing the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph, the problem of independent existence of knowledge and inability to establish connection and collaborative application can be solved.

[0090] Specifically, the fusion of the power grid physical model graph, the power grid operation data graph and the dispatching semantic graph is to obtain a more comprehensive and accurate power grid knowledge graph, wherein the three sub-graphs respectively describe different aspects of knowledge information such as the physical structure, real-time operation status and dispatching procedures in the power system. Fusing them into a whole can make up for their respective shortcomings in certain aspects and further enhance the completeness and practicality of the power grid knowledge graph. Specifically, by fusing the power grid physical model graph and the power grid operation data graph, it is possible to establish a connection between power equipment and combine it with real-time operation data to better understand the coupling relationship and influencing factors between various devices, thereby achieving comprehensive monitoring and prediction of the operation of the power system. At the same time, by fusing the dispatching semantic graph, it is also possible to link power dispatching instructions with specific equipment operations to achieve intelligent power grid operation management.

[0091] In a specific example of the present application, the process of fusing the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph includes:

[0092] First, semantic modeling is performed on the power grid physical model graph, power grid operation data graph, and dispatch semantic graph, and the entities, attributes, and relationships therein are converted into a unified knowledge representation form. Then, a fusion strategy is designed to match and integrate the same entities, attributes, and relationships in the three sub-graphs. For example, the concept of a transmission line connecting two substations is involved in the power grid physical model graph, the power grid operation data graph, and the dispatch semantic graph. They can be matched into the same entity based on their commonalities. Then, based on the fusion strategy, the three sub-graphs are integrated into a comprehensive power grid knowledge graph. This process requires certain data cleaning and redundant information elimination processing to ensure that the structure of the obtained knowledge graph is reasonable, complete, and accurate. Finally, the knowledge graph is stored in a suitable graph database, and is indexed and optimized to support efficient query and retrieval of the intelligent question-answering system.

[0093] Specifically, in step 130, electricity knowledge questions are obtained. Next, electricity knowledge questions are obtained. Here, electricity knowledge questions are obtained so that the intelligent question-answering system can understand and respond to user needs. In actual application scenarios, users may have different questions and needs about various aspects of the power system, such as equipment failures, operating status, dispatch instructions, etc. By obtaining these electricity knowledge questions, corresponding answers can be provided according to the specific needs of users, thereby meeting the user's information query needs. At the same time, the power system is a complex field that covers a large number of professional terms and knowledge points. If the user does not have relevant field knowledge, it is difficult to accurately describe his or her own problem. Therefore, by obtaining electricity knowledge questions, users can better express their needs and obtain more accurate answers.

[0094] Specifically, in step 140, after sentence recognition and entity recognition are performed on the electricity knowledge question, the semantic parsing model is input to obtain a semantic understanding feature vector of the electricity knowledge question. Next, after sentence recognition and entity recognition are performed on the electricity knowledge question, the semantic parsing model is input to obtain a semantic understanding feature vector of the electricity knowledge question. Sentence recognition and entity recognition are performed on the electricity knowledge question in order to convert the questions expressed in natural language into a structured form that can be processed by a machine. In the electricity knowledge intelligent question-answering system, users may ask questions in various natural language forms, and these questions need to be converted into structured data that can be processed by computers before subsequent processing and analysis can be performed.

[0095] Specifically, sentence recognition and entity recognition for questions about power knowledge can split the question text into components such as concepts, relationships, and attributes, so as to better understand the intent and content of the question. For example, for a question involving power equipment failure, "Which equipment has failed?" can be identified as a query statement containing entities such as "equipment" and "failure", and further converted into a corresponding structured query. Through entity recognition and sentence recognition, the question can be deconstructed into basic semantic units, laying the foundation for subsequent semantic understanding and calculation. After obtaining the structured question after sentence recognition and entity recognition, it is necessary to input the semantic parsing model for deep semantic understanding to obtain the semantic understanding feature vector of the power knowledge question. This feature vector can represent the semantic meaning of the question text and is used for subsequent knowledge matching and classification. Through the processing of the semantic understanding model, the intelligent question-answering system can improve its ability to understand questions, so as to more accurately answer the power knowledge questions raised by users.

[0096] The semantic parsing model is a Bert model including a word embedding layer.

[0097] Fig.10 is a flowchart of the sub-steps of step 140 in the power knowledge semantic analysis method according to an embodiment of the present application, such as Fig.10 As shown, after sentence recognition and entity recognition are performed on the electricity knowledge question, the question is input into a semantic parsing model to obtain a semantic understanding feature vector of the electricity knowledge question, including: 141, sentence recognition and entity recognition are performed on the electricity knowledge question to obtain multiple question components, wherein the question components include concepts, relationships and attributes; 142, the multiple question components are word-embedded using the word embedding layer of the semantic parsing model to obtain multiple question component word embedding vectors; and, 143, the multiple question component word embedding vectors are contextually semantically encoded using the Bert model of the semantic parsing model to obtain the semantic understanding feature vector of the electricity knowledge question.

[0098] Fig.11 is a flowchart of the sub-steps of step 143 in the power knowledge semantic analysis method according to an embodiment of the present application, such as Fig.11As shown, the Bert model of the semantic parsing model is used to perform context semantic encoding on the multiple question component word embedding vectors to obtain the semantic understanding feature vector of the power knowledge question, including: 1431, one-dimensionally arranging the multiple question component word embedding vectors to obtain a global word embedding feature vector; 1432, calculating the product between the global word embedding feature vector and the transposed vector of each question component word embedding vector in the multiple question component word embedding vectors to obtain multiple self-attention association matrices; 1433, respectively normalizing each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; 1434, each standardized self-attention association matrix in the multiple standardized self-attention association matrices is passed through a Softmax classification function to obtain multiple probability values; and, 1435, each question component word embedding vector in the multiple question component word embedding vectors is weighted using each probability value in the multiple probability values ​​as a weight to obtain the semantic understanding feature vector of the power knowledge question.

[0099] The context encoder aims to mine the hidden patterns between the contexts in the word sequence. Optionally, the encoder includes: CNN (Convolutional Neural Network), Recursive NN (Recursive Neural Network), Language Model, etc. The CNN-based method has a relatively good extraction effect on local features, but it is not effective for the long-term dependency problem in sentences. Therefore, the encoder based on Bi-LSTM (Long Short-Term Memory) is widely used. Recursive NN treats sentences as tree structures rather than sequences. In theory, it has stronger representation capabilities, but it has weaknesses such as difficulty in sample annotation, easy gradient disappearance in deep layers, and difficulty in parallel computing. Therefore, it is rarely used in practical applications. Transformer is a widely used network structure. It has the characteristics of CNN and RNN, has a better extraction effect on global features, and has certain advantages in parallel computing compared to RNN (Recurrent Neural Network).

[0100] Specifically, in step 150, the power grid knowledge graph is passed through a graph neural network model to obtain a power grid semantic knowledge graph feature matrix. Then, the power grid knowledge graph is passed through a graph neural network model to obtain a power grid semantic knowledge graph feature matrix. Here, passing the power grid knowledge graph through a graph neural network model to obtain a power grid semantic knowledge graph feature matrix is ​​to convert the information in the knowledge graph into a vector representation that can be processed and analyzed by a computer.

[0101] Traditional rule-based question-answering systems mainly rely on manually written rules and templates, which are difficult to handle large-scale and complex questions. Graph neural network technology can automatically learn the representation of nodes and edges from structured data, and perform graph-level calculations and reasoning on this basis. By inputting the power grid knowledge graph into the graph neural network model, the relationship between each entity in the graph and its embedded representation can be learned, so as to better understand the semantic relationship between these entities.

[0102] Specifically, by calculating the power grid knowledge graph through the graph neural network model, we can obtain the power grid semantic knowledge graph feature matrix, in which each entity is represented as a vector. This vector contains multi-dimensional features such as the entity's position, connection relationship, and semantic information in the power grid knowledge graph, which can be used for subsequent knowledge matching and classification tasks. At the same time, since the graph neural network has excellent processing effect on local subgraphs, it can also capture some global properties, such as important nodes and critical paths in the power system.

[0103] The study of graph neural networks is closely related to graph embedding or network embedding, which is another topic of increasing interest in the data mining and machine learning communities. Graph embedding aims to represent the vertices in a graph as low-dimensional vectors by preserving the network topology and node content information of the graph so that they can be processed using simple machine learning algorithms (e.g., support vector machine classification). Many graph embedding algorithms are usually unsupervised algorithms, which can be roughly divided into three categories, namely matrix decomposition, random walks, and deep learning methods. At the same time, deep learning methods for graph embedding also belong to graph neural networks, including algorithms based on graph autoencoders (such as DNGR and SDNE) and unsupervised trained graph convolutional neural networks (such as GraphSage). Graph neural networks are divided into five major categories: Graph Convolution Networks (GCN), Graph Attention Networks, Graph Autoencoders, Graph Generative Networks, and Graph Spatial-temporal Networks.

[0104] There are two main methods for graph convolutional networks (GCN) to perform convolution operations: one is based on spectral decomposition, namely spectral decomposition graph convolution. The other is based on node space transformation, namely spatial graph convolution. Bruna et al. generalized convolutional neural networks to graph data for the first time and proposed two parallel graph convolution models - spectral decomposition graph convolution and spatial graph convolution; GNN based on autoencoder is called graph autoencoder (GAE), which can learn graph node information semi-supervised or unsupervised. In the field of deep learning, autoencoder (AE) is a type of artificial neural network that represents input information for learning; Graph Generative Network (GGN) is a type of GNN used to generate graph data, which uses certain rules to recombine nodes and edges, and finally generates a target graph with specific properties and requirements; Graph Recurrent Network (GRN) is the earliest GNN model. Compared with other GNN algorithms, GRN usually converts graph data into sequences, and the sequences will continue to evolve and change recursively during the training process; the attention mechanism allows a neural network to focus only on the information required for task learning, and it can select specific inputs. Introducing the attention mechanism in GNN allows the neural network to focus on nodes and edges that are more relevant to the task, improving the effectiveness of training and the accuracy of testing, thus forming a graph attention network (GAT).

[0105] Specifically, in step 160, the semantic understanding feature vector of the electric power knowledge question is used as the query feature vector, and the product between it and the feature matrix of the power grid semantic knowledge graph is calculated to obtain a classification feature vector. Next, the semantic understanding feature vector of the electric power knowledge question is used as the query feature vector, and the product between it and the feature matrix of the power grid semantic knowledge graph is calculated to obtain a classification feature vector. Here, the semantic understanding feature vector of the electric power knowledge question is used as the query feature vector, and the product between it and the feature matrix of the power grid semantic knowledge graph is calculated to obtain a classification feature vector, in order to establish a connection between the electric power knowledge question and the power grid knowledge graph, thereby realizing question classification and answering.

[0106] In the intelligent question-answering system for electric power knowledge, questions raised by users need to be classified through a classification model to determine the type of question and the required answer information. Therefore, the query text needs to be converted into a computer-processable form and matched and compared with the power grid knowledge graph to find the best question classification and answer. Specifically, the semantic understanding feature vector of electric power knowledge questions is used as the query feature vector, and the question text can be represented as a multidimensional vector. This vector contains important information and semantic features of the question text, which can be used for subsequent question classification and answering. The feature matrix of the power grid semantic knowledge graph represents the vector representation of each entity in the knowledge graph, which can be used for knowledge matching and classification tasks.

[0107] By calculating the dot product or cosine similarity between the query feature vector and the feature matrix of the power grid semantic knowledge graph, a classification feature vector can be obtained. This vector can represent the relevance and matching degree between the current question and the corresponding entity in the power grid knowledge graph, thereby determining the best question classification and answer.

[0108] Specifically, in step 170, the feature distribution of the classification feature vector is optimized to obtain an optimized classification feature vector. In particular, in the technical solution of the present application, when the power knowledge question semantic understanding feature vector is used as the query feature vector and the product between it and the power grid semantic knowledge graph feature matrix is ​​calculated to obtain the classification feature vector, the problem semantic information represented by the power knowledge question semantic understanding feature vector is essentially mapped to the high-dimensional semantic feature space of the power grid knowledge graph. Since the power grid semantic knowledge graph feature matrix is ​​generated by a graph neural network model and the power knowledge question semantic understanding feature vector is generated by a semantic encoder, the power grid semantic knowledge graph feature matrix has a non-Euclidean topological structure, while the power knowledge question semantic understanding feature vector is still in the Euclidean space. Therefore, when the matrix product between the power knowledge question semantic understanding feature vector and the power grid semantic knowledge graph feature matrix is ​​taken, the mismatch of the spatial structure paradigm will cause the local distribution of the classification feature vector to be mismatched, which will lead to the irregularity of the feature distribution of the classification feature vector, thereby affecting the accuracy of the classification result through the classifier.

[0109] Based on this, the applicant of the present application performs a second regularization of the Gaussian probability density parameter of the manifold surface on the classification feature vector V, which is specifically expressed as: optimizing the feature distribution of the classification feature vector to obtain an optimized classification feature vector, including: optimizing the feature distribution of the classification feature vector using the following optimization formula to obtain an optimized classification feature vector; wherein the optimization formula is:

[0110] v′ i =(μσ)v i2 +v i μ+(v i -σ)μ 2

[0111] Wherein, V represents the classification feature vector, μ and σ are the mean and standard deviation of the feature value set at all positions in the classification feature vector, and v i is the eigenvalue of the ith position of the classification feature vector, v′ i is the eigenvalue of the i-th position of the optimized classification feature vector.

[0112] Specifically, in order to solve the problem of irregular distribution of high-dimensional feature distribution of the feature set of the classification feature vector V in the high-dimensional feature space of the query feature map, quadratic regularization is performed on each eigenvalue of the classification feature vector V through the likelihood of the Gaussian probability density parameter of the class probability distribution of the classifier, so as to smooth the eigenvalues ​​based on the equidistant distribution of the Gaussian probability density parameter in the parameter space of the target class probability, so as to obtain the regularized reorganization of the original probability density likelihood function expressed by the manifold surface of the high-dimensional feature in the parameter space, thereby improving the regularity of the feature distribution of the optimized classification feature vector V′, so as to improve the classification accuracy of the optimized classification feature vector V′ through the classifier.

[0113] Specifically, in step 180, the optimized classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to represent the sequence label of the candidate answer. Finally, the classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to represent the sequence label of the candidate answer. That is, the classification feature vector is input into the classifier to determine the sequence label of the answer to the power knowledge question through the classifier. In this way, intelligent question and answer of power grid knowledge is realized based on the power grid knowledge graph.

[0114] Fig.12 is a flowchart of the sub-steps of step 180 in the power knowledge semantic analysis method according to an embodiment of the present application, such as Fig.12 As shown, the optimized classification feature vector is passed through a classifier to obtain a classification result, and the classification result is used to represent a sequence label of an alternative answer, including: 181, using multiple fully connected layers of the classifier to fully connect encode the optimized classification feature vector to obtain an encoded classification feature vector; and, 182, passing the encoded classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

[0115] In summary, the electric power knowledge semantic analysis method 100 based on the embodiment of the present application is explained, which constructs a power grid physical model graph, a power grid operation data graph and a scheduling semantic graph, and integrates the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph; through the knowledge graph, various structured and unstructured texts are quickly processed, knowledge is extracted, and it is digested and organized to form a reasonable knowledge structure; intelligent question and answer of power grid knowledge is realized based on the power grid knowledge graph, thereby reducing the cost of knowledge induction and learning.

[0116] In one embodiment of the present application, Fig.13 FIG. 1 is a block diagram of a power knowledge semantic analysis system according to an embodiment of the present application. Fig.13 As shown, according to the embodiment of the present application, the electric power knowledge semantic analysis system 200 includes: a graph component module 210, which is used to construct a power grid physical model graph, a power grid operation data graph and a scheduling semantic graph; a graph fusion module 220, which is used to fuse the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph; an electric power knowledge acquisition module 230, which is used to obtain electric power knowledge questions; a semantic parsing module 240, which is used to perform sentence recognition and entity recognition on the electric power knowledge questions, and then input the semantic parsing model to obtain a semantic understanding feature vector of the electric power knowledge questions; a graph neural network module 25 0, used to pass the power grid knowledge graph through a graph neural network model to obtain a power grid semantic knowledge graph feature matrix; a vector calculation module 260, used to use the power knowledge question semantic understanding feature vector as a query feature vector, and calculate the product between it and the power grid semantic knowledge graph feature matrix to obtain a classification feature vector; an optimization module 270, used to optimize the feature distribution of the classification feature vector to obtain an optimized classification feature vector; and a sequence label generation module 280 for alternative answers, used to pass the optimized classification feature vector through a classifier to obtain a classification result, and the classification result is used to represent the sequence label of the alternative answers.

[0117] In a specific example, in the above-mentioned electric power knowledge semantic analysis system, the semantic parsing model is a Bert model including a word embedding layer.

[0118] In a specific example, in the above-mentioned electric power knowledge semantic analysis system, the semantic parsing module includes: a recognition unit, used to perform sentence recognition and entity recognition on the electric power knowledge question to obtain multiple question components, wherein the question components include concepts, relations and attributes; a word embedding unit, used to use the word embedding layer of the semantic parsing model to word embed the multiple question components to obtain multiple question component word embedding vectors; and a context encoding unit, used to perform context semantic encoding on the multiple question component word embedding vectors using the Bert model of the semantic parsing model to obtain the electric power knowledge question semantic understanding feature vector.

[0119] In a specific example, in the above-mentioned electric power knowledge semantic analysis system, the context encoding unit includes: a one-dimensional arrangement subunit, which is used to arrange the multiple question component word embedding vectors in one dimension to obtain a global word embedding feature vector; a self-attention subunit, which is used to calculate the product between the global word embedding feature vector and the transposed vector of each question component word embedding vector in the multiple question component word embedding vectors to obtain multiple self-attention association matrices; a normalization subunit, which is used to standardize each self-attention association matrix in the multiple self-attention association matrices to obtain multiple standardized self-attention association matrices; a probability value calculation subunit, which is used to pass each standardized self-attention association matrix in the multiple standardized self-attention association matrices through a Softmax classification function to obtain multiple probability values; and a weighting subunit, which is used to weight each question component word embedding vector in the multiple question component word embedding vectors with each probability value in the multiple probability values ​​as a weight to obtain the electric power knowledge question semantic understanding feature vector.

[0120] In a specific example, in the above-mentioned power knowledge semantic analysis system, the optimization module is used to: optimize the feature distribution of the classification feature vector according to the following optimization formula to obtain an optimized classification feature vector; wherein the optimization formula is:

[0121] v′ i =(μσ)v i 2 +v i μ+(v i -σ)μ 2

[0122] Wherein, V represents the classification feature vector, μ and σ are the mean and standard deviation of the feature value set at all positions in the classification feature vector, and v i is the eigenvalue of the ith position of the classification feature vector, v′ i is the eigenvalue of the i-th position of the optimized classification feature vector.

[0123] In a specific example, in the above-mentioned electric power knowledge semantic analysis system, the sequence label generation module of the alternative answers includes: an encoding unit, used to use multiple fully connected layers of the classifier to fully connect the optimized classification feature vector to obtain an encoded classification feature vector; and a classification result unit, used to pass the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0124] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the above-mentioned power knowledge semantic analysis system have been referred to above. Figures 1 to 12 The description of the power knowledge semantic analysis method has been introduced in detail, and therefore, its repeated description will be omitted.

[0125] The present application also provides a computer program product, which includes instructions. When the instructions are executed, the device performs operations corresponding to the above method.

[0126] In one embodiment of the present application, a computer-readable storage medium is further provided, wherein the computer-readable storage medium stores a computer program for executing the above-mentioned method.

[0127] It should be understood that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware can be adopted. Moreover, the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code can be adopted.

[0128] The method, system, and computer program product of the embodiments of the present application are described by flowcharts and / or block diagrams. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 one or more processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 one or more processes and / or boxes Figure 1 A function specified in one or more boxes.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 one or more processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0131] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0132] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0133] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0134] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0135] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0136] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A method for semantic analysis of electric power knowledge, characterized in that: include: Construct the power grid physical model map, power grid operation data map and dispatch semantic map; The power grid physical model graph, the power grid operation data graph and the scheduling semantic graph are integrated to obtain a power grid knowledge graph; Get electricity knowledge questions; After performing sentence recognition and entity recognition on the power knowledge question, the question is input into a semantic parsing model to obtain a semantic understanding feature vector of the power knowledge question; Inputting the power grid knowledge graph into a graph neural network model for processing to obtain a power grid semantic knowledge graph feature matrix; Taking the power knowledge question semantic understanding feature vector as the query feature vector, calculating the product between the query feature vector and the power grid semantic knowledge graph feature matrix to obtain a classification feature vector; Performing feature distribution optimization on the classification feature vector to obtain an optimized classification feature vector; as well as Inputting the optimized classification feature vector into a classifier for processing to obtain a classification result, wherein the classification result is used to represent a sequence label of an alternative answer; The step of optimizing the feature distribution of the classification feature vector to obtain an optimized classification feature vector comprises: optimizing the feature distribution of the classification feature vector according to the following optimization formula to obtain an optimized classification feature vector; Wherein, the optimization formula is: ， in, and is the mean and standard deviation of the eigenvalue set at all positions in the classification feature vector, is the first The eigenvalues ​​at the positions, is the first The feature value at each position.

2. The power knowledge semantic analysis method according to claim 1 is characterized in that: The semantic parsing model includes a word embedding layer and a Bert model.

3. The power knowledge semantic analysis method according to claim 2 is characterized in that: After performing sentence recognition and entity recognition on the power knowledge question, the question is input into a semantic parsing model to obtain a semantic understanding feature vector of the power knowledge question, including: Performing sentence recognition and entity recognition on the power knowledge question to obtain a plurality of question components, wherein the question components include concepts, relations and attributes; Using the word embedding layer of the semantic parsing model to embed the multiple question components to obtain multiple question component word embedding vectors: and The Bert model of the semantic parsing model is used to perform context semantic encoding on the multiple question component word embedding vectors to obtain the semantic understanding feature vector of the power knowledge question.

4. The power knowledge semantic analysis method according to claim 3 is characterized in that: Using the Bert model of the semantic parsing model to perform context semantic encoding on the multiple question component word embedding vectors to obtain the power knowledge question semantic understanding feature vector, including: Arrange the multiple question component word embedding vectors in one dimension to obtain a global word embedding feature vector; Calculating the product of the global word embedding feature vector and the transposed vector of each question component word embedding vector in the multiple question component word embedding vectors to obtain multiple self-attention association matrices; Normalizing each of the plurality of self-attention association matrices to obtain a plurality of standardized self-attention association matrices; Inputting each of the plurality of standardized self-attention association matrices into a Softmax classification function for processing to obtain a plurality of probability values; and Each of the multiple probability values ​​is used as a weight to weight each of the multiple question component word embedding vectors to obtain the power knowledge question semantic understanding feature vector.

5. The method for semantic analysis of electric power knowledge according to claim 4, characterized in that: The optimized classification feature vector is input into a classifier for processing to obtain a classification result, and the classification result is used to represent a sequence label of the candidate answer, including: Using multiple fully connected layers of the classifier to perform fully connected encoding on the optimized classification feature vector to obtain an encoded classification feature vector; and The encoded classification feature vector is input into the Softmax classification function of the classifier for processing to obtain the classification result.

6. A power knowledge semantic analysis system, characterized in that: include: Graph component module, used to construct power grid physical model graph, power grid operation data graph and dispatch semantic graph; A graph fusion module, used for fusing the power grid physical model graph, the power grid operation data graph and the scheduling semantic graph to obtain a power grid knowledge graph; The power knowledge acquisition module is used to obtain power knowledge questions; A semantic parsing module, which is used to perform sentence recognition and entity recognition on the power knowledge question and then input the semantic parsing model to obtain a semantic understanding feature vector of the power knowledge question; A graph neural network module, used for inputting the power grid knowledge graph into a graph neural network model for processing to obtain a power grid semantic knowledge graph feature matrix; A vector calculation module, used for taking the power knowledge question semantic understanding feature vector as a query feature vector, and calculating the product between the query feature vector and the power grid semantic knowledge graph feature matrix to obtain a classification feature vector; An optimization module, used for optimizing the feature distribution of the classification feature vector to obtain an optimized classification feature vector; as well as A sequence label generation module for alternative answers, used for inputting the optimized classification feature vector into a classifier for processing to obtain a classification result, wherein the classification result is used to represent a sequence label of the alternative answer; The step of optimizing the feature distribution of the classification feature vector to obtain an optimized classification feature vector comprises: optimizing the feature distribution of the classification feature vector according to the following optimization formula to obtain an optimized classification feature vector; Wherein, the optimization formula is: ， in, and is the mean and standard deviation of the eigenvalue set at all positions in the classification feature vector, is the first The eigenvalues ​​at the positions, is the first The feature value at each position.

7. The electric power knowledge semantic analysis system according to claim 6, characterized in that: The semantic parsing model includes a word embedding layer and a Bert model.

8. The electric power knowledge semantic analysis system according to claim 7, characterized in that: The semantic parsing module comprises: A recognition unit, used for performing sentence recognition and entity recognition on the power knowledge question to obtain a plurality of question components, wherein the question components include concepts, relations and attributes; A word embedding unit, used to embed the multiple question components using the word embedding layer of the semantic parsing model to obtain multiple question component word embedding vectors; and A context encoding unit is used to perform context semantic encoding on the multiple question component word embedding vectors using the Bert model of the semantic parsing model to obtain the power knowledge question semantic understanding feature vector.

9. The electric power knowledge semantic analysis system according to claim 8, characterized in that: The context encoding unit comprises: A one-dimensional arrangement subunit, used for one-dimensionally arranging the multiple question component word embedding vectors to obtain a global word embedding feature vector; A self-attention subunit is used to calculate the product between the global word embedding feature vector and the transposed vector of each question component word embedding vector in the multiple question component word embedding vectors to obtain multiple self-attention association matrices; A normalization subunit, used for normalizing each of the plurality of self-attention association matrices to obtain a plurality of normalized self-attention association matrices; a probability value calculation subunit, configured to input each of the plurality of standardized self-attention association matrices into a Softmax classification function for processing to obtain a plurality of probability values; and The weighting subunit is used to weight each question component word embedding vector in the multiple question component word embedding vectors using each probability value in the multiple probability values ​​as a weight to obtain the power knowledge question semantic understanding feature vector.

Images