CN-121981121-A - Entity identification method and device of power dispatching text and electronic equipment

CN121981121ACN 121981121 ACN121981121 ACN 121981121ACN-121981121-A

Abstract

The application discloses an entity identification method and device of a power dispatching text and electronic equipment. The method comprises the steps of determining initial feature vectors corresponding to a plurality of segmentation units respectively based on a target power dispatching text, correcting the initial feature vectors corresponding to the segmentation units respectively to obtain target feature vectors corresponding to the segmentation units respectively, determining entity tag score vectors corresponding to the segmentation units respectively based on the target feature vectors corresponding to the segmentation units, determining target entity tag vectors of the target power dispatching text based on the entity tag score vectors corresponding to the segmentation units respectively, and determining target entity identification results of the target power dispatching text based on the target entity tag vectors. The application solves the technical problem of inaccurate entity identification result of the power dispatching text in the related technology.

Inventors

SHA LICHENG
CHEN QIAN
SUN YIFEI
YAO YIDI
WU YAN
XI SHAOQING
WANG CHUNLING
WEI TIANYING
WANG HAIYUN
Xing Guangzheng
YANG LIPING
Wu Zaichi

Assignees

国网北京市电力公司

Dates

Publication Date: 20260505
Application Date: 20260130

Claims (10)

1. An entity identification method of a power dispatching text is characterized by comprising the following steps: determining initial feature vectors respectively corresponding to a plurality of segmentation units based on a target power dispatching text, wherein the segmentation units are obtained by segmenting the target power dispatching text; Correcting the initial feature vectors respectively corresponding to the plurality of segmentation units to obtain target feature vectors respectively corresponding to the plurality of segmentation units; Determining entity tag score vectors corresponding to the plurality of segmentation units respectively based on target feature vectors corresponding to the plurality of segmentation units respectively, wherein the entity tag score vectors are used for indicating the matching degree of the corresponding segmentation units and preset entity tags, and the entities are words in the target power dispatching text, and each entity at least corresponds to one segmentation unit; Determining a target entity tag vector of the target power dispatching text based on the entity tag score vectors respectively corresponding to the plurality of segmentation units, wherein the target entity tag vector is formed by target entity tags respectively corresponding to the plurality of segmentation units; And determining a target entity identification result of the target power dispatching text based on the target entity label vector, wherein the target entity identification result is used for describing entity categories corresponding to a plurality of entities included in the target power dispatching text.
2. The method according to claim 1, wherein determining initial feature vectors respectively corresponding to the plurality of segmentation units based on the target power schedule text comprises: Encoding the plurality of segmentation units respectively to obtain feature vector sets corresponding to the plurality of segmentation units respectively, wherein the feature vector sets comprise semantic feature vectors, fragment feature vectors and position feature vectors of the corresponding segmentation units; for any segmentation unit in the plurality of segmentation units, carrying out feature fusion on the semantic feature vector, the segment feature vector and the position feature vector of the any segmentation unit to obtain a fusion feature vector of the any segmentation unit; Correcting the spliced feature vector of any segmentation unit to obtain an initial feature vector of any segmentation unit; And determining initial feature vectors corresponding to the plurality of segmentation units respectively by adopting a mode of determining the initial feature vector of any segmentation unit.
3. The method according to claim 1, wherein the correcting the initial feature vectors respectively corresponding to the plurality of segmentation units to obtain the target feature vectors respectively corresponding to the plurality of segmentation units includes: based on the initial feature vectors respectively corresponding to the plurality of segmentation units, a two-way long-short-term memory network is adopted to obtain first feature vectors respectively corresponding to the plurality of segmentation units; and obtaining target feature vectors respectively corresponding to the plurality of segmentation units by adopting a hierarchical attention mechanism based on the first feature vectors respectively corresponding to the plurality of segmentation units.
4. The method of claim 3, wherein the obtaining, based on the first feature vectors respectively corresponding to the plurality of segmentation units, the target feature vectors respectively corresponding to the plurality of segmentation units by using a hierarchical attention mechanism includes: Determining position weights corresponding to the plurality of segmentation units respectively based on first feature vectors corresponding to the plurality of segmentation units respectively, wherein the position weights are used for quantitatively describing the importance of the positions of the corresponding segmentation units in the target power dispatching text; determining a global semantic vector of the target power scheduling text based on first feature vectors respectively corresponding to the plurality of segmentation units and position weights respectively corresponding to the plurality of segmentation units, wherein the global semantic vector is used for describing context dependency relationships of the plurality of entities included in the target power scheduling text; Based on the first feature vectors respectively corresponding to the plurality of segmentation units and the preset embedded vectors respectively corresponding to the plurality of segmentation units, obtaining entity perception weights respectively corresponding to the plurality of segmentation units, wherein the entity perception weights are used for quantitatively describing the importance of the corresponding segmentation units in the target power dispatching text; And determining target feature vectors respectively corresponding to the plurality of segmentation units based on the first feature vectors respectively corresponding to the plurality of segmentation units, the position weights respectively corresponding to the plurality of segmentation units, the global semantic vector and the entity perception weights respectively corresponding to the plurality of segmentation units.
5. The method according to any one of claims 1 to 4, wherein determining the target entity tag vector of the target power dispatch text based on the entity tag score vectors respectively corresponding to the plurality of segmentation units includes: determining a plurality of initial entity tag vectors based on the preset entity tags; Screening the plurality of initial entity tag vectors based on constraint conditions to obtain a plurality of screened entity tag vectors; And determining the target entity label vector based on the plurality of screening entity label vectors and the entity label score vectors respectively corresponding to the plurality of segmentation units.
6. The method of claim 5, wherein determining the target entity tag vector based on the plurality of filtered entity tag vectors and the entity tag score vectors respectively corresponding to the plurality of segmentation units comprises: Determining label scores corresponding to the screening entity label vectors respectively and transfer scores corresponding to the screening entity label vectors respectively based on the screening entity label vectors and the entity label score vectors corresponding to the splitting units respectively, wherein the label scores are used for indicating the matching degree of the corresponding screening entity label vectors and the target power dispatching text, and the transfer scores are used for indicating the rationality degree of adjacent entity label conversion in the corresponding screening entity label vectors; Based on the label scores respectively corresponding to the screening entity label vectors and the transfer scores respectively corresponding to the screening entity label vectors, obtaining total scores respectively corresponding to the screening entity label vectors; And determining the screening entity tag vector corresponding to the maximum total score in the total scores respectively corresponding to the screening entity tag vectors as the target entity tag vector.
7. The method of claim 1, wherein the determining the target entity identification result of the target power schedule text based on the target entity tag vector comprises: identifying a first entity label representing an entity starting position from target entity labels respectively corresponding to the plurality of segmentation units; Based on the first entity tag, the target entity tag vector is segmented to obtain a plurality of target entity tag sub-vectors, wherein the target entity tag sub-vectors are in one-to-one correspondence with the plurality of entities; and determining the target entity identification result based on the plurality of target entity tag sub-vectors.
8. An entity recognition device of a power dispatching text, comprising: The first determining module is used for determining initial feature vectors corresponding to a plurality of segmentation units respectively based on a target power dispatching text, wherein the plurality of segmentation units are obtained by segmenting the target power dispatching text; the target feature vector determining module is used for respectively correcting the initial feature vectors respectively corresponding to the plurality of segmentation units to obtain target feature vectors respectively corresponding to the plurality of segmentation units; The second determining module is used for determining entity tag score vectors respectively corresponding to the plurality of segmentation units based on target feature vectors respectively corresponding to the plurality of segmentation units, wherein the entity tag score vectors are used for indicating the matching degree of the corresponding segmentation units and preset entity tags, and the entities are words in the target power dispatching text, and each entity at least corresponds to one segmentation unit; a third determining module, configured to determine a target entity tag vector of the target power scheduling text based on entity tag score vectors corresponding to the plurality of segmentation units, where the target entity tag vector is formed by target entity tags corresponding to the plurality of segmentation units; and the fourth determining module is used for determining a target entity identification result of the target power dispatching text based on the target entity tag vector, wherein the target entity identification result is used for describing entity categories corresponding to a plurality of entities included in the target power dispatching text.
9. A non-volatile storage medium, characterized in that it stores a plurality of instructions adapted to be loaded by a processor and to perform the entity identification method of power dispatch text of any one of claims 1 to 7.
10. An electronic device comprising one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the entity identification method of power dispatch text of any one of claims 1 to 7.

Description

Entity identification method and device of power dispatching text and electronic equipment Technical Field The application relates to the field of power systems, in particular to an entity identification method and device of a power dispatching text and electronic equipment. Background Along with the continuous innovation of the construction of a novel power system, the industrial scale is driven to be continuously expanded, so that the novel power system is gradually developed into a modern industry integrating high technical threshold, wide coverage and multidisciplinary fusion. Meanwhile, with the continuous increase of the scale and complexity of the power system, especially in the field of power dispatching, massive professional text information such as regulations, reports, logs and the like is generated and relied on every day. These texts are not only the direct basis for scheduling decisions, but also an important carrier of industry knowledge. In conventional modes, the processing of such information is often dependent upon manual reading and understanding, is inefficient, and is prone to error. The automated entity identification technology can rapidly and accurately extract key entity information such as equipment, operation instructions, fault states and the like from the texts, so that the efficiency and accuracy of scheduling decision are remarkably improved, and the method is beneficial to real-time monitoring and rapid response of various conditions in a power system. In the related art, a method based on a preset rule is adopted to identify the entity in the power dispatching text, so that complex semantic changes and diversity of professional terms in the power dispatching text cannot be dealt with, the rule design becomes very complicated along with the increase of entity types and the complexity of relationships among the entities, and all possible entity combinations are difficult to cover, so that the problems of poor flexibility, limited coverage and insufficient accuracy exist, and further the error of the obtained entity identification result is larger. Therefore, the technical problem that the entity identification result of the power dispatching text in the related technology is inaccurate. In view of the above problems, no effective solution has been proposed at present. Disclosure of Invention The embodiment of the application provides a method and a device for identifying an entity of a power dispatching text and electronic equipment, which are used for at least solving the technical problem that the entity identification result of the power dispatching text is inaccurate in the related technology. According to one aspect of the embodiment of the application, an entity identification method of a power dispatching text is provided, which comprises the steps of determining initial feature vectors corresponding to a plurality of segmentation units respectively based on the target power dispatching text, correcting the initial feature vectors corresponding to the plurality of segmentation units respectively to obtain target feature vectors corresponding to the plurality of segmentation units respectively, determining entity tag score vectors corresponding to the plurality of segmentation units respectively based on the target feature vectors corresponding to the plurality of segmentation units respectively, wherein the entity tag score vectors are used for indicating the matching degree of the corresponding segmentation units and preset entity tags, the entity is a word in the target power dispatching text, each entity corresponds to at least one segmentation unit, determining a target entity tag vector of the target power dispatching text based on the entity tag score vectors corresponding to the plurality of segmentation units respectively, determining a target entity recognition result of the target power dispatching text based on the target entity tag vectors corresponding to the plurality of segmentation units, and describing the category of the target entity corresponding to the target power dispatching text. According to another aspect of the embodiment of the application, an entity identification device of a power dispatching text is provided, which comprises a first determining module, a target feature vector determining module and a fourth determining module, wherein the first determining module is used for determining initial feature vectors corresponding to a plurality of segmentation units respectively based on the target power dispatching text, the plurality of segmentation units are obtained by segmenting the target power dispatching text, the target feature vector determining module is used for respectively correcting the initial feature vectors corresponding to the plurality of segmentation units to obtain target feature vectors corresponding to the plurality of segmentation units respectively, the second determining module is used for determining entity label sco