Design method and system of knowledge graph for risk warning information of power grid regulation

Abstract

A method and system for designing a knowledge graph of risk warning information for power grid regulation, the method comprising: collecting semi-structured data and unstructured data of power grid regulation to construct a dispatching knowledge base; using the dispatching knowledge base to design a hierarchical knowledge graph of risk warning for power grid regulation; using the knowledge graph of risk warning for power grid regulation in combination with expert experience knowledge to express the risk warning system of power grid regulation in a logical hierarchy; collecting power grid regulation risk warning notices, and extracting events from the notices of risk warning for power grid regulation to obtain the relationship between the facts and the elements of the event of the risk warning instance of power grid regulation, and constructing a graph of risk warning events for power grid regulation. The present invention obtains the relationship between the facts and the elements of the facts through event extraction, and establishes a graph of risk warning events for power grid regulation. The graph of risk warning events for power grid regulation integrates multiple types of events and entity knowledge, and is a specific form of the knowledge graph of risk warning information for power grid regulation.

Description

Technical Field

[0001] The present invention relates to the technical field of power system risk detection, and more specifically, to a method and system for designing a knowledge graph of power grid regulation risk warning information. Background Art

[0002] At present, fault information and risk warning information in the power grid control system are recorded in the form of "events" and stored in the system in files of different structures. In actual work, the existing dispatching system lacks the understanding of the correlation characterization of risk information, and lacks the convenience of systematic storage, query and retrieval of warning information. In recent years, driven by artificial intelligence as the core technology, logical reasoning graphs that describe the logical evolution relationship between events are receiving widespread attention. On the one hand, this type of graph focuses on the transfer of spatial semantics and temporal semantic states between events, and can express a stronger and more direct correlation than correlation semantics; on the other hand, it has a stronger knowledge carrying capacity. These two characteristics determine that reasoning graphs can support a variety of applications such as event prediction, logical question and answer, and logical verification.

[0003] Prior art 1 (CN111899089A) "Enterprise risk early warning method and system based on knowledge graph" obtains structured data and semi-structured data of enterprise information, processes them into standardized structure data and imports them into the data pool; obtains unstructured data of enterprise information, processes them into enterprise relationship data and imports them into the corpus; constructs the knowledge graph of the target enterprise based on the standardized structure data in the data pool and the enterprise relationship data in the corpus; uses the pre-trained risk identification model to mine the risk information of the target enterprise from the knowledge graph to achieve risk early warning. The power grid dispatching risk business is different from the generally understood fraud risk. The power grid dispatching risk business is mainly expressed in the form of an event and a specific situation. The hierarchical dispatching risk early warning graph schema should not only include ontological relationships, but also event hierarchies and related relationships. However, represented by prior art 1, when establishing the knowledge graph, only triple data in the risk information is extracted, lacking the relationship extraction and event hierarchy construction of the risk event system.

[0004] In recent years, there has been some research progress in domain knowledge graphs. However, due to the current level of natural language processing technology and the understanding of the risk warning business of power grid regulation, the existing technology of power grid regulation still has the following deficiencies in the construction and application of logical knowledge bases and graphs:

[0005] 1) The accuracy of automatic extraction is not high. At present, the data accuracy of events and relationships constructed and extracted based on the fully automatic method is not high. Among them, although the method of extracting by dependency syntax or regular rule template can obtain a high recall rate, the error rate is high. The extraction by deep learning sequence annotation method can ensure a high accuracy rate, but it requires a large-scale annotated corpus.

[0006] 2) The extraction of event elements in scheduling event extraction often requires the presetting of element categories in advance. The understanding of scheduling business requires the accumulation of relatively rich expert knowledge and the definition of relatively accurate knowledge and experience.

[0007] 3) The integration and linking of event knowledge and entity knowledge is still relatively blank. Linking entity knowledge to events in event knowledge can further enrich application functions. Summary of the invention

[0008] In order to address the deficiencies in the prior art, the purpose of the present invention is to provide a method and system for designing a knowledge graph for power grid regulation risk warning information, to sort out and study the power grid regulation risk warning events and the causal relationship structure of the events that are lacking in the prior art, to classify and structure the power grid regulation risk warning events, to obtain the causal relationship and causal elements through event extraction, and to establish a power grid regulation risk warning event graph, which integrates multiple types of events and entity knowledge, and is a specific form of the power grid regulation risk warning information knowledge graph.

[0009] The present invention adopts the following technical solution.

[0010] On the one hand, the present invention proposes a method for designing a knowledge graph of early warning information of power grid regulation risk, and the design method includes:

[0011] Step 1: Collect semi-structured data and unstructured data of power grid control; use the pre-processed semi-structured data and unstructured data to build a dispatching knowledge base;

[0012] Step 2: Design a hierarchical knowledge graph of power grid regulation risk warning using the dispatching knowledge base; Use the power grid regulation risk warning knowledge graph combined with expert experience knowledge to express the power grid regulation risk warning system in a logical hierarchy; The power grid regulation risk warning knowledge graph includes: a hierarchical system of abstract events of power grid regulation risk warning, and an argument structure of abstract events of power grid regulation risk warning;

[0013] Step 3, collecting power grid regulation risk warning notices, and extracting events from the power grid regulation risk warning notices to obtain the cause-and-effect relationship of power grid regulation risk warning instance events and the cause-and-effect elements of power grid regulation risk warning instance events;

[0014] Step 4, using the causal relationship of the power grid regulation risk warning instance event and the causal elements of the power grid regulation risk warning instance event, construct a power grid regulation risk warning event map; wherein the power grid regulation risk warning event map is a directed cyclic map, with the causal elements of the power grid regulation risk warning instance event as nodes, and the causal relationship of the power grid regulation risk warning instance event as directed edges.

[0015] In step 1, the semi-structured data of power grid control includes: dispatch system database data and text documents in a fixed format;

[0016] Unstructured data for power grid control includes: dispatching various documents without fixed formats, plans, logs, operation records, and pictures;

[0017] The preprocessing of semi-structured data includes: format conversion;

[0018] The preprocessing of unstructured data includes: punctuation processing, word segmentation, word frequency statistics and index library establishment.

[0019] In step 1, the dispatching knowledge base is used to store the hierarchical relationship between the concept knowledge of power grid regulation risk warning and the concept knowledge of power grid regulation risk warning; the dispatching knowledge base includes: a noun entity abstract knowledge base, a trait class abstract knowledge base and an action event abstract knowledge base;

[0020] The noun entity abstract knowledge base stores nouns with hyponymous semantic relationships; the trait abstract knowledge base stores adjectives or adverbs; the action event abstract knowledge base stores verbs;

[0021] Among them, the conceptual knowledge of power grid regulation risk warning is a fusion of the HowNet Chinese vocabulary, encyclopedia knowledge base and text materials and vocabulary in the field of power grid regulation.

[0022] In step 1, the construction of the scheduling knowledge base includes:

[0023] Step 1.1, obtain the abstract level from the HowNet Chinese word library and the encyclopedia knowledge base; wherein the Glossary file in the HowNet Chinese word library is used as the concept level system file;

[0024] Step 1.2, through manual annotation, obtain text data in the field of power grid regulation from the semi-structured data and unstructured data of power grid regulation; merge the text data with the concept hierarchy system file to form a hierarchical classification tree, in which the term with the highest word frequency is set as the root node, and the root node is taken as the entity;

[0025] Step 1.3, using the word vector tool to train text data in the field of power grid regulation, to obtain word vector text in the field of power grid regulation containing entities and a word list in the field of power grid regulation;

[0026] Step 1.4, set all words in the word vector text in the field of power grid regulation as seed words, calculate the similarity between the remaining text materials in the field of power grid regulation and the seed words except the seed words, and take the words with similarity greater than the similarity threshold as synonyms and store them;

[0027] Step 1.5, summarizing the synonyms obtained in step 1.4 and the synonyms obtained in the synonym dictionary to obtain the concept knowledge of power grid regulation risk warning;

[0028] Step 1.6, construct a hierarchical directed graph of the entities and superordinate concepts of the words obtained in step 1.2. Based on the shortest path algorithm, the maximum value of the shortest path of each entity is used as the hierarchical abstract path of the entity, and the hierarchical relationship between the conceptual knowledge of power grid regulation risk warning is represented by the hierarchical abstract path.

[0029] In step 2, the hierarchical system of scheduling risk warning events is constructed based on a rule combination method, which is a combination of a top-down method and a bottom-up method. The construction process of the hierarchical system includes:

[0030] Step 2.1, construct a hierarchical system of dispatching risk warning events from top to bottom, that is, based on expert experience and knowledge, summarize, classify and define dispatching risk warning events according to their characteristics, and design a hierarchical system of dispatching risk warning events according to the business scenarios of power grid regulation risk warning, specifically:

[0031] Step 2.1.1, obtain core actions from the scheduling knowledge base as first-level events; core actions include verbs, nouns, and noun-verb phrases;

[0032] Step 2.1.2, based on the primary event, obtain the secondary event through the method of behavior subject + core action; the behavior subject includes nouns and noun phrases;

[0033] Step 2.1.3, taking the attribute event describing the second-level event as the third-level event;

[0034] Step 2.1.4, repeat step 2.1.3 for all events of level 3 and above, and gradually iterate to obtain all events of level 4 and above;

[0035] Step 2.2, construct a hierarchical system of dispatch risk warning events from the bottom up, that is, classify the secondary events obtained in step 2.1.2 according to the semi-structured data and unstructured data of power grid regulation, and classify the secondary events of the same type into the first-level events of the same type, specifically:

[0036] Step 2.2.1, identifying various event names from multiple secondary events by extracting event phrases and event boundaries;

[0037] Step 2.2.2, identify the entity types of words for multiple secondary events by means of entity recognition, use the entity types of words to replace words, and obtain the upper-level events of the secondary events;

[0038] Step 2.2.3, abstract the core actions in the upper-level events as a supplement to the first-level events.

[0039] In step 2, the argument structure of the scheduling risk warning event includes: event trigger words and elements describing the event structure; the construction of the argument structure of the scheduling risk warning event is automatically completed by the machine, including:

[0040] Step 2.3.1, initialize the candidate event trigger words to obtain the event trigger word feature vector;

[0041] Step 2.3.2, automatically updating the learning event trigger word feature vector and the combined event trigger word feature vector based on the neural network model;

[0042] Step 2.3.3, based on the combination result of the event trigger word feature vector obtained in step 2.3.2, output the argument structure of different classifications based on the Softmax model;

[0043] Step 2.3.4, for the dispatch risk warning event, add the elements describing the event structure, specifically:

[0044] 1) Use common arguments as elements to describe event structure, including time, place, person, and organization;

[0045] 2) Based on the type of scheduling risk warning events, the core actions, subject and object of the scheduling risk warning events are extracted as elements to describe the event structure.

[0046] In step 3, the method for extracting the relationship between the event and the event of the power grid regulation risk warning instance includes:

[0047] 1) Based on relationship qualifiers, part-of-speech filtering and event validity, the relationship between power grid regulation risk warning notice is extracted according to the relationship type;

[0048] 2) Based on the relational qualifiers and dependency syntactic rules, the relationship between the power grid regulation risk warning notice is extracted according to the relationship type;

[0049] 3) Based on sequence annotation, the relationship between power grid regulation risk warning notice is extracted according to the relationship type;

[0050] The method for extracting the causal elements of the power grid regulation risk warning instance event includes: extracting the general arguments related to the power grid regulation risk warning instance event from the power grid regulation risk warning instance event; wherein the general arguments include the core action of the event, the subject of the event, the object of the event, the time of the event, the place of the event, the event-related entities, the emotion of the event, the tense of the event, and the certainty of the event; the event-related entities include related persons, related companies, and related institutions.

[0051] In step 3, event extraction of the power grid regulation risk warning notice includes:

[0052] Step 3.1, event detection, i.e. identifying the event trigger words in the text of the power grid regulation risk warning notice, and judging the event type according to the type of the event trigger words; each event type corresponds to a unique event representation framework; and the event representation framework is used to determine the relationship between the power grid regulation risk warning instance event;

[0053] Step 3.2, element identification, is to determine whether the entity in the text of the power grid regulation risk warning notice is an event element based on the event representation framework, and determine the role of the event element; use the role of the event element to determine the event element of the power grid regulation risk warning instance event.

[0054] In step 4, the directed edges represent the logical relationships between events, which include: causal relationships, conditional relationships, mutually exclusive relationships, composition relationships, transition relationships, concurrent relationships, and hierarchical relationships.

[0055] On the other hand, the present invention also proposes a power grid regulation risk warning information knowledge graph design system, which implements a power grid regulation risk warning information knowledge graph design method.

[0056] The beneficial effects of the present invention are as follows:

[0057] 1) The present invention proposes a method for designing a power grid regulation risk warning event map that integrates multiple types of events and entity knowledge, and organically realizes the construction of an event map in the field from three perspectives: abstract events, instance events, and entity knowledge. The resulting power grid regulation risk warning event map has a clear hierarchy and a complete system, and is a specific form of the power grid regulation risk warning information knowledge map;

[0058] 2) The present invention proposes an event extraction method for power grid regulation risk warning notices, which has high extraction accuracy and does not require large-scale corpus annotation;

[0059] 3) The present invention proposes a method for constructing a scheduling knowledge base, which realizes external and internal hierarchical fusion to obtain scheduling concept knowledge. The scheduling knowledge base is also a multi-level abstract knowledge base; the knowledge base is suitable for providing accurate knowledge and experience for understanding scheduling business, thereby realizing operations without the need for expert experience and knowledge, and without the need to pre-set element categories in advance;

[0060] 4) The present invention proposes a graph structure in which instance event knowledge and entity knowledge of power grid regulation risk warning coexist. By integrating instance event knowledge and entity knowledge, entity knowledge is linked to instance events of instance event knowledge, thereby further enriching the semantic information of the knowledge graph of power grid regulation risk warning information;

[0061] 5) The power grid regulation risk warning event map constructed by the method of the present invention can integrate the advantages of both static entity knowledge and event knowledge, further broaden the application scope of dispatch fault query, and provide warning and reasoning services based on event knowledge. It can also be used as a knowledge base to expand field experience resources and improve the safety and security of power grid dispatch. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flowchart of the steps of the method for designing the knowledge graph of the risk warning information of power grid regulation of the present invention;

[0063] Figure 2 is a schematic diagram of the structure of a scheduling knowledge base in an embodiment of the present invention;

[0064] Figure 3 is a schematic diagram of a hierarchical system of events in an embodiment of the present invention;

[0065] Figure 4 It is a flow chart for constructing a power grid regulation risk warning event map in an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The present application is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present application.

[0067] like Figure 1 On the one hand, the present invention proposes a method for designing a knowledge graph of power grid regulation risk warning information, and the design method includes steps 1 to 4.

[0068] Step 1: Collect semi-structured data and unstructured data of power grid control; use the pre-processed semi-structured data and unstructured data to build a dispatching knowledge base.

[0069] In step 1, the semi-structured data of power grid control includes: dispatch system database data and text documents in a fixed format;

[0070] Unstructured data for power grid control includes: dispatching various documents without fixed formats, plans, logs, operation records, and pictures;

[0071] The preprocessing of semi-structured data includes: format conversion;

[0072] The preprocessing of unstructured data includes: punctuation processing, word segmentation, word frequency statistics and index library establishment.

[0073] In step 1, the dispatching knowledge base is used to store the hierarchical relationship between the concept knowledge of power grid regulation risk warning and the concept knowledge of power grid regulation risk warning; the dispatching knowledge base includes: a noun entity abstract knowledge base, a trait class abstract knowledge base and an action event abstract knowledge base;

[0074] The noun entity abstraction knowledge base stores nouns with a hierarchical semantic relationship. In this preferred embodiment, for example, "Hua Ke Transformer" and "Substation", "Substation" and "Grid Equipment" are a pair of nouns describing a hierarchical semantic relationship. "Hua Ke Transformer" can be abstracted upwards into "Substation", and "Substation" can be further abstracted upwards into "Grid Equipment". A large number of entity pairs with such a hierarchical semantic relationship together form a huge knowledge network.

[0075] The trait class abstract knowledge base stores adjectives or adverbs. In the preferred embodiment, "strengthen" and "ensure" belong to the trait class abstract knowledge base.

[0076] The action event abstract knowledge base stores verbs, such as "maintenance" and "operation" in the preferred embodiment, which belong to the trait class abstract knowledge base.

[0077] Among them, the conceptual knowledge of power grid regulation risk warning is a fusion of the HowNet Chinese vocabulary, encyclopedia knowledge base and text materials and vocabulary in the field of power grid regulation.

[0078] In this preferred embodiment, the structure of the scheduling knowledge base is as follows: Figure 2 shown.

[0079] In step 1, the construction of the scheduling knowledge base includes:

[0080] Step 1.1, obtain the abstract level from the HowNet Chinese word library and the encyclopedia knowledge base; wherein the Glossary file in the HowNet Chinese word library is used as the concept level system file;

[0081] Step 1.2, through manual annotation, obtain text data in the field of power grid regulation from the semi-structured data and unstructured data of power grid regulation; merge the text data with the concept hierarchy system file to form a hierarchical classification tree, in which the term with the highest word frequency is set as the root node, and the root node is taken as the entity;

[0082] Step 1.3, using the word vector tool to train text data in the field of power grid regulation, to obtain word vector text in the field of power grid regulation containing entities and a word list in the field of power grid regulation;

[0083] Step 1.4, set all words in the word vector text in the field of power grid regulation as seed words, calculate the similarity between the remaining text materials in the field of power grid regulation and the seed words except the seed words, and take the words with similarity greater than the similarity threshold as synonyms and store them;

[0084] Step 1.5, summarizing the synonyms obtained in step 1.4 and the synonyms obtained in the synonym dictionary to obtain the concept knowledge of power grid regulation risk warning;

[0085] Step 1.6, construct a hierarchical directed graph of the entities and superordinate concepts of the words obtained in step 1.2. Based on the shortest path algorithm, the maximum value of the shortest path of each entity is used as the hierarchical abstract path of the entity, and the hierarchical relationship between the conceptual knowledge of power grid regulation risk warning is represented by the hierarchical abstract path.

[0086] Step 2: Design a hierarchical knowledge graph for power grid regulation risk warning using the dispatching knowledge base; use the power grid regulation risk warning knowledge graph combined with expert experience knowledge to express the power grid regulation risk warning system in a logical hierarchy; the power grid regulation risk warning knowledge graph includes: a hierarchical system of abstract events for power grid regulation risk warning, and an argument structure for abstract events for power grid regulation risk warning.

[0087] The event tuple includes two levels of meaning, one is the event hierarchy system, and the other is the event argument structure. Among them, the event system focuses more on the event relational knowledge and the event composition relational knowledge. The event argument structure needs to be constrained according to a framework system similar to the ACE framework for the corresponding argument structure under a certain type of event. Among them, the event hierarchy system is as follows: Figure 3 shown.

[0088] In step 2, the hierarchical system of scheduling risk warning events is constructed based on a rule combination method, which is a combination of a top-down method and a bottom-up method. The construction process of the hierarchical system includes:

[0089] Step 2.1, construct a hierarchical system of dispatching risk warning events from top to bottom, that is, based on expert experience and knowledge, summarize, classify and define dispatching risk warning events according to their characteristics, and design a hierarchical system of dispatching risk warning events according to the business scenarios of power grid regulation risk warning, specifically:

[0090] Step 2.1.1, obtain the core action as the first-level event from the scheduling knowledge base; the core action includes verbs, nouns, and noun-verb phrases; in this preferred embodiment, "fault information warning notification" is a first-level event;

[0091] Step 2.1.2, based on the primary event, obtain the secondary event through the method of behavior subject + core action; the behavior subject includes nouns and noun phrases; in this preferred embodiment, for the primary event of "fault information warning notification", secondary events such as "busbar fault tripping" and "loss of load" are obtained;

[0092] Step 2.1.3, taking the attribute event describing the second-level event as the third-level event; in this preferred embodiment, from the second-level event "loss of load", "loss of load 110,000 kilowatts" can be further obtained as the third-level event;

[0093] Step 2.1.4, repeat step 2.1.3 for events at level three and above, and gradually iterate to obtain events at level four and above; Step 2.1.3 describes a set of new attribute rules, according to which more levels of events can be gradually iterated, thus forming a hierarchical system of events.

[0094] The core of the top-down event hierarchy system is the design of combination rules and the acquisition of corresponding subjects, attributes and other components. For example, how to obtain enough first-level core actions can quickly expand a large number of event types, but it is obvious that there are obvious language rules between event types. In contrast to the bottom-up event hierarchy system, the bottom-up system starts from specific instance events, gradually clusters instance events through event abstraction, event clustering, etc., and finally obtains first-level events. It is an abstract generalization process from the final event to the first-level event.

[0095] Step 2.2, construct a hierarchical system of dispatch risk warning events from the bottom up, that is, classify the secondary events obtained in step 2.1.2 according to the semi-structured data and unstructured data of power grid regulation, and classify the secondary events of the same type into the first-level events of the same type, specifically:

[0096] Step 2.2.1, identifying various event names from multiple secondary events by extracting event phrases and event boundaries;

[0097] Step 2.2.2, identify the entity types of words for multiple secondary events by means of entity recognition, use the entity types of words to replace words, and obtain the upper-level events of the secondary events;

[0098] Step 2.2.3, abstract the core actions in the upper-level events as a supplement to the first-level events.

[0099] In this preferred embodiment, for example, "The transmission operation and inspection room and the cable operation and inspection room strengthen the special patrol of the 220 kV Huili 2Y05 / 2Y06 line" becomes "[role][line][action]", and "Partial power outage of 220 kV and 10 kV at Huake substation" becomes "[substation] power outage".

[0100] If we get more abstract information about the entity, such as the voltage level attributes of Huake transformer, such as "220 kV" and "10 kV", we can get more forms of event hierarchy. In this way, the core actions in the event can be further abstracted, and finally a complete event system can be obtained. For example, the event system of "partial power loss of Huake transformer 220 kV and 10 kV" and the transformation are: "[Substation] power loss" → "[Substation] [fault phenomenon]" → "[Grid equipment] [fault phenomenon]" → "fault phenomenon".

[0101] The argument structures of some events in the existing technology are all based on manual construction and cannot contain knowledge in the specific field of power grid. Therefore, how to use machines to automatically complete the construction of event argument structures is also a big problem.

[0102] In step 2, the argument structure of the scheduling risk warning event includes: event trigger words and elements describing the event structure; the construction of the argument structure of the scheduling risk warning event is automatically completed by the machine, including:

[0103] Step 2.3.1, initialize the candidate event trigger words to obtain the event trigger word feature vector;

[0104] Step 2.3.2, automatically updating the learning event trigger word feature vector and the combined event trigger word feature vector based on the neural network model;

[0105] Step 2.3.3, based on the combination result of the event trigger word feature vector obtained in step 2.3.2, output the argument structure of different classifications based on the Softmax model;

[0106] Step 2.3.4, for the dispatch risk warning event, add the elements describing the event structure, specifically:

[0107] 1) Use common arguments as elements to describe event structure, including time, place, person, and organization;

[0108] 2) Based on the type of scheduling risk warning events, the core actions, subject and object of the scheduling risk warning events are extracted as elements to describe the event structure.

[0109] Step 3, collect power grid regulation risk warning notices, and extract events from the power grid regulation risk warning notices to obtain the causal relationship of the power grid regulation risk warning instance events and the causal elements of the power grid regulation risk warning instance events.

[0110] In step 3, the method for extracting the relationship between the event and the event of the power grid regulation risk warning instance includes:

[0111] 1) Based on relationship qualifiers, part-of-speech filtering and event validity, the relationship between power grid regulation risk warning notice is extracted according to the relationship type;

[0112] In this preferred embodiment, the relational qualifier refers to the associated indicator that can clearly show the components of the matter, such as "lead to", "because-so" and other causal association words, and the same applies to the other seven kinds of matter-of-fact association words. Through the limitation of the relational qualifier, the event description field is extracted; part-of-speech filtering refers to setting the part-of-speech category to be retained after word segmentation and part-of-speech tagging by using the Chinese word segmenter; event validity refers to the event must contain specific event components, such as event action, event subject, object, etc.

[0113] 2) Based on the relational qualifiers and dependency syntactic rules, the relationship between the power grid regulation risk warning notice is extracted according to the relationship type;

[0114] In this preferred embodiment, it specifically includes determining the candidate events through relational qualifiers, determining the event subject and event object corresponding to the conjunctions by means of dependency syntactic analysis, and expanding the events through the semantic relationship between the modifiers.

[0115] 3) Based on sequence labeling, the relationship between the power grid control risk warning notice is extracted according to the relationship type; in this preferred embodiment, it includes but is not limited to labeling the start event and end event descriptions using tag sets such as BIO or BIOS, and finally training the sequence labeling model for labeling.

[0116] The method for extracting the causal elements of the power grid regulation risk warning instance event includes: extracting the general arguments related to the power grid regulation risk warning instance event from the power grid regulation risk warning instance event; wherein the general arguments include the core action of the event, the subject of the event, the object of the event, the time of the event, the place of the event, the event-related entities, the emotion of the event, the tense of the event, and the certainty of the event; the event-related entities include related persons, related companies, and related institutions.

[0117] In step 3, event extraction of the power grid regulation risk warning notice includes:

[0118] Step 3.1, event detection, i.e. identifying the event trigger words in the text of the power grid regulation risk warning notice, and judging the event type according to the type of the event trigger words; each event type corresponds to a unique event representation framework; and the event representation framework is used to determine the relationship between the power grid regulation risk warning instance event;

[0119] An instance event consists of an event trigger and event elements that describe the event structure. Event elements include event participants, event time, event location, etc. In addition, an event may also include attribute information, including but not limited to: type, subcategory, mode, tendency, universality, and tense.

[0120] Step 3.2, element identification, is to determine whether the entity in the text of the power grid regulation risk warning notice is an event element based on the event representation framework, and determine the role of the event element; use the role of the event element to determine the event element of the power grid regulation risk warning instance event.

[0121] The input of step 3.2 depends on the output of step 3.1, and can also be processed in a more detailed manner, turning event extraction into 3 or 4 subtasks, such as: trigger word recognition, event type classification, element recognition, and role classification tasks.

[0122] In this preferred embodiment, the extraction of instance event elements is a structured representation of events. The result of the structured representation is the extraction of common arguments related to the event, namely event elements. The common arguments include: core actions, which refer to the core verbs in the power grid dispatching faults and control instructions; action subjects, which refer to the agents around the core verbs; action objects, which refer to the recipients around the core verbs; event time, which refers to the background time when the event occurs; event location, which refers to the background location when the event occurs; event-related persons, which refer to the related persons of the event; event degree, which refers to the level of the warning event; event category, which refers to the classification of the event; event status, which refers to the tense of the event, including the two states of executed and unexecuted; event source, which refers to the text source where the event is extracted; event compilation time; event compiler; event review event; event reviewer.

[0123] Step 4, using the causal relationship of the power grid regulation risk warning instance event and the causal elements of the power grid regulation risk warning instance event, construct a power grid regulation risk warning event map; wherein the power grid regulation risk warning event map is a directed cyclic map, with the causal elements of the power grid regulation risk warning instance event as nodes, and the causal relationship of the power grid regulation risk warning instance event as directed edges.

[0124] In this preferred embodiment, the power grid regulation risk warning event map is a domain event map. The domain event map is a directed cyclic graph, where nodes represent events and directed edges represent logical relationships between events such as cause and effect, conditions, mutual exclusion, composition, transition, and superior and subordinate. The domain event map focuses on both the evolutionary logical relationship between external events and the rich attribute information of the events themselves. It is a unified combination of abstract events and event knowledge bases.

[0125] The types of relationship between things in the domain event map include: causal things, conditional things, superior and subordinate things, parallel things, mutually exclusive things, sequential things, transitional things, and composition things.

[0126] In the present invention, the power grid regulation risk warning event does not involve mutually exclusive events, sequential events, and transition events, but only includes the other five types of events. In step 4, the directed edge represents the logical relationship between events, and the logical relationship between events includes: causal relationship, conditional relationship, mutually exclusive relationship, composition relationship, transition relationship, concurrent relationship, and superior-subordinate relationship.

[0127] In this preferred embodiment, the cause and effect events include: Event A: Huake transformer 220 kV operating busbar fault tripping; Event B: Huake transformer 220 kV, 10 kV partial power outage.

[0128] In this preferred embodiment, the conditional events include: Event A: risk cause; Event B: control measures.

[0129] In this preferred embodiment, the upper and lower level events include: Event A: management and control measures; Event B: the power dispatching and control center executes according to the requirements of the Nanjing local dispatching documents.

[0130] In this preferred embodiment, the parallel events include: Event A: Operation and Maintenance Department (Maintenance Branch) Substation Operation and Maintenance Room: Strengthen the operation and maintenance of related equipment of 220 kV Xiaozhuang Substation (especially Xiaozhong 2578 interval), Central Gate Substation (especially 220 kV busbar and interval), Xiaguan Substation (especially Xiazhong 2579 interval) to ensure safe operation; Event B: Operation and Maintenance Department (Maintenance Branch) Transmission Operation and Inspection Room, Operation and Maintenance Department (Maintenance Branch) Cable Operation and Inspection Room: Strengthen special patrols of 220 kV Xiaozhong 2578 and Xiazhong 2579 lines to ensure safe operation.

[0131] In this preferred embodiment, the components include: Event A: The power dispatching and control center and the power supply service command center strengthen the monitoring and control of the Huake transformer main transformer, the Longshan transformer main transformer, and the 110 kV Hualong Line 774 line load; Event B: The power supply service command center strengthens the monitoring of the Huake transformer main transformer.

[0132] On the other hand, the present invention also proposes a power grid regulation risk warning information knowledge graph design system, which implements a power grid regulation risk warning information knowledge graph design method.

[0133] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

Claims

1. A knowledge graph design method for early warning information of power grid regulation risk, characterized in that: The design method comprises: Step 1, collect semi-structured data and unstructured data of power grid control; use the pre-processed semi-structured data and unstructured data to build a dispatching knowledge base; wherein, the dispatching knowledge base is used to store the conceptual knowledge of power grid control risk warning and the hierarchical relationship between the conceptual knowledge of power grid control risk warning; the dispatching knowledge base includes: a noun entity abstract knowledge base, a trait class abstract knowledge base and an action event abstract knowledge base; the noun entity abstract knowledge base stores nouns with a hierarchical relationship; the trait class abstract knowledge base stores adjectives or adverbs; the action event abstract knowledge base stores verbs; the conceptual knowledge of power grid control risk warning is a fusion of the Chinese word base of HowNet, the encyclopedia knowledge base and the text materials and thesaurus in the field of power grid control; in step 1, the construction of the dispatching knowledge base includes: Step 1.1, obtain the abstract level from the HowNet Chinese word library and the encyclopedia knowledge base; wherein the Glossary file in the HowNet Chinese word library is used as the concept level system file; Step 1.2, through manual annotation, obtain text data in the field of power grid regulation from the semi-structured data and unstructured data of power grid regulation; merge the text data with the concept hierarchy system file to form a hierarchical classification tree, in which the term with the highest word frequency is set as the root node, and the root node is taken as the entity; Step 1.3, using the word vector tool to train text data in the field of power grid regulation, to obtain word vector text in the field of power grid regulation containing entities and a word list in the field of power grid regulation; Step 1.4, set all words in the word vector text in the field of power grid regulation as seed words, calculate the similarity between the remaining text materials in the field of power grid regulation and the seed words except the seed words, and take the words with similarity greater than the similarity threshold as synonyms and store them; Step 1.5, summarizing the synonyms obtained in step 1.4 and the synonyms obtained in the synonym dictionary to obtain the concept knowledge of power grid regulation risk warning; Step 1.6, construct a directed graph of the entity and the superordinate concept of the word obtained in step 1.2, and based on the shortest path algorithm, take the maximum value of the shortest path of each entity as the superordinate and subordinate abstract path of the entity, and use the superordinate and subordinate abstract path to represent the superordinate and subordinate hierarchical relationship between the concept knowledge of power grid regulation risk warning; Step 2: Design a hierarchical knowledge graph of power grid regulation risk warning using the dispatching knowledge base; Use the power grid regulation risk warning knowledge graph combined with expert experience knowledge to express the power grid regulation risk warning system in a logical hierarchy; The power grid regulation risk warning knowledge graph includes: a hierarchical system of abstract events of power grid regulation risk warning, and an argument structure of abstract events of power grid regulation risk warning; In step 2, the hierarchical system of scheduling risk warning events is constructed based on a rule combination method, which is a combination of a top-down method and a bottom-up method. The construction process of the hierarchical system includes: Step 2.1, construct a hierarchical system of dispatching risk warning events from top to bottom, that is, based on expert experience and knowledge, summarize, classify and define dispatching risk warning events according to their characteristics, and design a hierarchical system of dispatching risk warning events according to the business scenarios of power grid regulation risk warning, specifically: Step 2.1.1, obtain core actions from the scheduling knowledge base as first-level events; core actions include verbs, nouns, and noun-verb phrases; Step 2.1.2, based on the primary event, obtain the secondary event through the method of behavior subject + core action; the behavior subject includes nouns and noun phrases; Step 2.1.3, taking the attribute event describing the second-level event as the third-level event; Step 2.1.4, repeat step 2.1.3 for all events of level 3 and above, and gradually iterate to obtain all events of level 4 and above; Step 2.2, construct a hierarchical system of dispatch risk warning events from the bottom up, that is, classify the secondary events obtained in step 2.1.2 according to the semi-structured data and unstructured data of power grid regulation, and classify the secondary events of the same type into the first-level events of the same type, specifically: Step 2.2.1, identifying various event names from multiple secondary events by extracting event phrases and event boundaries; Step 2.2.2, identify the entity types of words for multiple secondary events by means of entity recognition, use the entity types of words to replace words, and obtain the upper-level events of the secondary events; Step 2.2.3, abstract the core actions in the upper-level events as a supplement to the first-level events; In step 2, the argument structure of the scheduling risk warning event includes: event trigger words and elements describing the event structure; the construction of the argument structure of the scheduling risk warning event is automatically completed by the machine, including: Step 2.3.1, initialize the candidate event trigger words to obtain the event trigger word feature vector; Step 2.3.2, automatically updating the learning event trigger word feature vector and the combined event trigger word feature vector based on the neural network model; Step 2.3.3, based on the combination result of the event trigger word feature vector obtained in step 2.3.2, output the argument structure of different classifications based on the Softmax model; Step 2.3.4, for the dispatch risk warning event, add the elements describing the event structure, specifically: 1) Use common arguments as elements to describe the structure of events, including time, place, person, and organization; 2) Based on the type of scheduling risk warning events, the core actions, subject and object of the scheduling risk warning events are extracted as elements to describe the event structure; Step 3, collecting power grid regulation risk warning notices, and extracting events from the power grid regulation risk warning notices to obtain the cause-and-effect relationship of power grid regulation risk warning instance events and the cause-and-effect elements of power grid regulation risk warning instance events; Step 4, using the causal relationship of the power grid regulation risk warning instance event and the causal elements of the power grid regulation risk warning instance event, construct a power grid regulation risk warning event map; wherein the power grid regulation risk warning event map is a directed cyclic map, with the causal elements of the power grid regulation risk warning instance event as nodes, and the causal relationship of the power grid regulation risk warning instance event as directed edges.

2. The method for designing a knowledge graph of power grid regulation risk warning information according to claim 1, characterized in that: In step 1, the semi-structured data of power grid control includes: dispatch system database data and text documents in a fixed format; Unstructured data for power grid control includes: dispatching various documents without fixed formats, plans, logs, operation records, and pictures; The preprocessing of semi-structured data includes: format conversion; The preprocessing of unstructured data includes: punctuation processing, word segmentation, word frequency statistics and index library establishment.

3. The method for designing a knowledge graph of power grid regulation risk warning information according to claim 1, characterized in that: In step 3, the method for extracting the relationship between the event and the event of the power grid regulation risk warning instance includes: 1) Based on relationship qualifiers, part-of-speech filtering and event validity, the relationship between power grid regulation risk warning notice is extracted according to the relationship type; 2) Based on the relational qualifiers and dependency syntax rules, the relationship between the power grid regulation risk warning notice is extracted according to the relationship type; 3) Based on sequence annotation, the relationship between power grid regulation risk warning notice is extracted according to the relationship type; The method for extracting the causal elements of the power grid regulation risk warning instance event includes: extracting the general arguments related to the power grid regulation risk warning instance event from the power grid regulation risk warning instance event; wherein the general arguments include the core action of the event, the subject of the event, the object of the event, the time of the event, the place of the event, the event-related entities, the emotion of the event, the tense of the event, and the certainty of the event; the event-related entities include related persons, related companies, and related institutions.

4. The method for designing a knowledge graph of power grid regulation risk warning information according to claim 3 is characterized in that: In step 3, event extraction of the power grid regulation risk warning notice includes: Step 3.1, event detection, i.e. identifying the event trigger words in the text of the power grid regulation risk warning notice, and judging the event type according to the type of the event trigger words; each event type corresponds to a unique event representation framework; and the event representation framework is used to determine the relationship between the power grid regulation risk warning instance event; Step 3.2, element identification, is to determine whether the entity in the text of the power grid regulation risk warning notice is an event element based on the event representation framework, and determine the role of the event element; use the role of the event element to determine the event element of the power grid regulation risk warning instance event.

5. The method for designing a knowledge graph of power grid regulation risk warning information according to claim 4 is characterized in that: In step 4, the directed edges represent the logical relationships between events, which include: causal relationships, conditional relationships, mutually exclusive relationships, composition relationships, transition relationships, concurrent relationships, and hierarchical relationships.

6. A knowledge graph design system for early warning information of power grid regulation risk, characterized in that: The system implements the power grid regulation risk warning information knowledge graph design method as described in any one of claims 1-5.

Images