CN-121980029-A - Clean energy field label mapping method, device, equipment and medium

CN121980029ACN 121980029 ACN121980029 ACN 121980029ACN-121980029-A

Abstract

The application provides a method, a device, equipment and a medium for mapping a label in the field of clean energy, and relates to the technical field of data processing. The method comprises the steps of firstly carrying out global semantic mining on clean energy data to be analyzed to form clean energy global features, secondly carrying out data extraction on the clean energy data to be analyzed to obtain clean energy field data, carrying out semantic mining on the clean energy field data to form clean energy field features, wherein the clean energy field data comprises at least one text unit which has a correlation with the clean energy field in the clean energy data to be analyzed, then carrying out semantic constraint on the clean energy global features based on the clean energy field features to form clean energy constraint features, and finally carrying out semantic reduction on the clean energy constraint features to form target tag data. Based on the above, the problem of low accuracy of label mapping in the prior art can be improved.

Inventors

ZHANG BAOPING
LI DONGFANG
YU ZHANGTAO
QIAN DESONG
SHENG HUAYING
WANG WENYONG
CHEN MINGXUAN

Assignees

三峡科技有限责任公司

Dates

Publication Date: 20260505
Application Date: 20260108

Claims (10)

1. The clean energy field label mapping method is characterized by comprising the following steps of: performing global semantic mining on clean energy data to be analyzed to form clean energy global features; Extracting data from the clean energy data to be analyzed to obtain clean energy field data, and performing semantic mining on the clean energy field data to form clean energy field features, wherein the clean energy field data comprises at least one text unit with a correlation relationship with the clean energy field in the clean energy data to be analyzed; based on the clean energy field characteristics, carrying out semantic constraint on the clean energy global characteristics to form clean energy constraint characteristics; And carrying out semantic reduction on the clean energy constraint characteristics to form target tag data.
2. The clean energy domain label mapping method according to claim 1, wherein the step of performing global semantic mining on the clean energy data to be analyzed to form a clean energy global feature comprises: Word embedding processing is carried out on clean energy data to be analyzed to obtain clean energy embedded features, and segmentation processing is carried out on the clean energy embedded features to form a plurality of clean energy local features, wherein the granularity of the segmentation processing is greater than or equal to that of a text unit; sequentially carrying out association fusion on two adjacent clean energy local features in the plurality of clean energy local features along a first direction to obtain a first energy fusion feature; Sequentially performing association fusion on two adjacent cleaning energy local features in the plurality of cleaning energy local features along a second direction to obtain a second energy fusion feature, wherein the first direction and the second direction refer to a direction from a first cleaning energy local feature to a last cleaning energy local feature and a direction from the last cleaning energy local feature to the first cleaning energy local feature respectively; And polymerizing the first energy fusion feature and the second energy fusion feature to obtain a clean energy global feature.
3. The clean energy domain label mapping method according to claim 1, wherein the step of extracting the data of the clean energy to be analyzed to obtain clean energy domain data, and performing semantic mining on the clean energy domain data to form clean energy domain features comprises the steps of: Extracting each text unit belonging to the clean energy field from the clean energy data to be analyzed to obtain the clean energy field data; word embedding processing is carried out on the clean energy field data to obtain energy field embedding characteristics; And carrying out deep semantic mining on the embedded features of the energy field to form the features of the clean energy field.
4. The clean energy domain label mapping method of claim 3, wherein the step of deep semantic mining the energy domain embedded features to form clean energy domain features comprises: Clustering the text unit embedded features corresponding to each text unit included in the energy field embedded features to form at least one embedded feature cluster; Aiming at each embedded feature cluster, respectively carrying out association fusion on each other text unit embedded feature in the embedded feature cluster and the central text unit embedded feature in the embedded feature cluster to form at least one cluster internal fusion feature, and carrying out aggregation on the at least one cluster internal fusion feature to form a target fusion feature; determining a central target fusion feature in target fusion features corresponding to each embedded feature cluster, and respectively carrying out association fusion on each other target fusion feature and the central target fusion feature to form at least one cluster external fusion feature; And polymerizing the at least one cluster external fusion feature to form a clean energy domain feature.
5. The clean energy domain label mapping method of claim 1, wherein the step of semantically constraining the clean energy global feature based on the clean energy domain feature to form a clean energy constrained feature comprises: Performing multiple semantic diffusion on the clean energy field features to form multiple clean energy diffusion features, wherein one semantic diffusion corresponds to one clean energy diffusion feature; and respectively carrying out semantic constraint on the clean energy global features based on each clean energy diffusion feature to form clean energy constraint features.
6. The clean energy domain label mapping method of claim 5, wherein the step of performing a plurality of semantic diffusions on the clean energy domain features to form a plurality of clean energy diffusion features comprises: at least one pooling treatment is carried out on the clean energy domain features to form at least one energy domain pooled feature, and self-attention treatment is carried out on each energy domain pooled feature to form at least one clean energy diffusion feature; And performing at least one convolution treatment on the clean energy domain features to form at least one energy domain convolution feature, and performing self-attention treatment on each energy domain convolution feature to form at least one clean energy diffusion feature.
7. The clean energy domain label mapping method of claim 5, wherein the step of semantically constraining the clean energy global features based on each of the clean energy diffusion features, respectively, to form clean energy constrained features comprises: Aiming at each clean energy diffusion feature, carrying out association fusion on the clean energy diffusion feature and the clean energy global feature so as to realize preliminary semantic constraint and form an energy preliminary constraint feature corresponding to the clean energy diffusion feature; Performing gating mapping on the energy preliminary constraint features aiming at each energy preliminary constraint feature to form gating mapping parameter distribution corresponding to the energy preliminary constraint features, and performing gating adjustment on the clean energy global features based on the gating mapping parameter distribution to realize semantic constraint and form energy gating constraint features corresponding to the energy preliminary constraint features; And respectively carrying out weighted aggregation on each energy gating constraint characteristic based on gating focusing indexes of each gating mapping parameter distribution to form clean energy constraint characteristics, wherein the gating focusing indexes are used for reflecting the distribution characteristics of all parameters in the gating mapping parameter distribution.
8. A clean energy field tag mapping apparatus, comprising: the first semantic mining module is used for carrying out global semantic mining on clean energy data to be analyzed to form clean energy global features; the second semantic mining module is used for extracting data of the clean energy data to be analyzed to obtain clean energy field data, and carrying out semantic mining on the clean energy field data to form clean energy field features, wherein the clean energy field data comprises at least one text unit which has a correlation with the clean energy field in the clean energy data to be analyzed; the semantic constraint module is used for carrying out semantic constraint on the clean energy global features based on the clean energy field features to form clean energy constraint features; the semantic restoration module is used for carrying out semantic restoration on the clean energy constraint characteristics to form target tag data.
9. An electronic device, comprising: A memory for storing a computer program; a processor coupled to the memory for executing the computer program stored in the memory to implement the clean energy field tag mapping method of any one of claims 1-7.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program which, when run, performs the clean energy field label mapping method according to any one of claims 1-7.

Description

Clean energy field label mapping method, device, equipment and medium Technical Field The application relates to the technical field of data processing, in particular to a method, a device, equipment and a medium for mapping labels in the field of clean energy. Background The automatic label mapping technology in the current clean energy field mainly adapts to industry scenes around a general text processing framework, and realizes automatic or semi-automatic label mapping through methods such as a manual rule mapping technology, a general NLP model migration technology, a simple semantic similarity matching technology and the like. Based on the technical schemes, some depending field specialists manually formulate tag matching rules, associate the content in the clean energy texts with preset tags in the modes of keyword accurate matching, regular expressions and the like, some directly adopt general text classification models such as BERT, TF-ID and the like, fine-tune through a small amount of clean energy labeling data to realize automatic tag mapping, and other ones select tags with highest similarity as mapping results by calculating semantic similarity between text content and the tags. The general NLP model is not optimized for the technical terms (such as 'light storage and charge integration', 'wind disposal rate', 'green card transaction') in the clean energy field, and label mapping deviation is easy to occur, for example, the 'energy storage system charge and discharge efficiency' is mapped into a 'photovoltaic-performance index'. The manual rule mapping requires an expert to customize rules one by one, and is difficult to ensure to form comprehensive mapping rules in face of newly added terms and labels of massive subdivision scenes (such as photovoltaics, wind power, energy storage, hydrogen energy and the like) in the clean energy field. Therefore, in the prior art, there is a problem that the accuracy of the clean energy field tag mapping is not high. Disclosure of Invention Accordingly, the present application is directed to a method, apparatus, device and medium for mapping labels in clean energy field, so as to solve the problem of low accuracy of label mapping in clean energy field in the prior art. In order to achieve the above purpose, the application adopts the following technical scheme: A clean energy domain label mapping method comprising: performing global semantic mining on clean energy data to be analyzed to form clean energy global features; Extracting data from the clean energy data to be analyzed to obtain clean energy field data, and performing semantic mining on the clean energy field data to form clean energy field features, wherein the clean energy field data comprises at least one text unit with a correlation relationship with the clean energy field in the clean energy data to be analyzed; based on the clean energy field characteristics, carrying out semantic constraint on the clean energy global characteristics to form clean energy constraint characteristics; And carrying out semantic reduction on the clean energy constraint characteristics to form target tag data. In a preferred option of the present application, in the above method for mapping a clean energy field tag, the step of performing global semantic mining on the clean energy data to be analyzed to form a clean energy global feature includes: Word embedding processing is carried out on clean energy data to be analyzed to obtain clean energy embedded features, and segmentation processing is carried out on the clean energy embedded features to form a plurality of clean energy local features, wherein the granularity of the segmentation processing is greater than or equal to that of a text unit; sequentially carrying out association fusion on two adjacent clean energy local features in the plurality of clean energy local features along a first direction to obtain a first energy fusion feature; Sequentially performing association fusion on two adjacent cleaning energy local features in the plurality of cleaning energy local features along a second direction to obtain a second energy fusion feature, wherein the first direction and the second direction refer to a direction from a first cleaning energy local feature to a last cleaning energy local feature and a direction from the last cleaning energy local feature to the first cleaning energy local feature respectively; And polymerizing the first energy fusion feature and the second energy fusion feature to obtain a clean energy global feature. In a preferred option of the present application, in the above method for mapping a clean energy domain label, the step of extracting data from the clean energy data to be analyzed to obtain the clean energy domain data, and performing semantic mining on the clean energy domain data to form the clean energy domain feature includes: Extracting each text unit belonging to the clean energy field from the clean energy data to be a