CN-121996967-A - Industrial electric response information intelligent analysis method and system based on large language model

Abstract

The application relates to an intelligent analysis method and system for industrial movie response information based on a large language model, which belong to the technical field of artificial intelligence and energy power intersection, and aim at the problem that unstructured information cannot be automatically identified to cause low efficiency on industrial movie response in the prior art; the method comprises the steps of constructing a semantic understanding and influence recognition model based on field knowledge enhancement, extracting key entities, attributes, relations and numerical values by inputting preprocessing data, constructing a power influence path, integrating standardized power influence event vectors and outputting the standardized power influence event vectors. The application can capture signals from network information, realize early warning and analysis of power utilization impact, break through the limitation of traditional numerical data, enrich the information dimension of decision, and can be docked into the existing power service system, thereby having extremely high engineering application value and practicability.

Inventors

• FANG ZHICHUN
• CHEN LINHONG
• WU HAOTIAN
• YIN HONGYUAN
• CHENG YUAN
• WU HAO
• RUAN JIAN
• WANG JUN
• WANG HUIDONG
• YE HONGDOU
• LI LEI
• CAO RUIFENG
• ZHANG YUNLEI
• LU PENGFEI
• XIAO JIDONG
• JIANG CHI

Assignees

• 国网浙江省电力有限公司营销服务中心

Dates

Publication Date

20260508

Application Date

20260128

Claims (9)

1. 1. An intelligent analysis method for industrial electric response information based on a large language model is characterized by comprising the following steps: s1, collecting power consumption information data, and preprocessing the collected data; s2, constructing a semantic understanding and influence recognition model based on field knowledge enhancement, and inputting the result of the step S1 into the constructed semantic understanding and influence recognition model to extract key entities, attributes, relations and numerical values; s3, constructing a power influence path based on the result of the step S2; and S4, integrating the power influence paths constructed in the step S3, and outputting the normalized power influence event vectors.
2. 2. The intelligent analysis method for industrial electrical response information based on large language model according to claim 1, wherein the step S1 comprises: S1.1, defining information sources related to power consumption, including but not limited to government portals, mainstream news media, social media platforms, industry forums, marketing company notices, weather information platforms; s1.2, periodically collecting multi-mode heterogeneous information data in an information source related to power consumption; s1.3, preprocessing the acquired data; S1.4, outputting format standardized information, wherein the format standardized information comprises an information ID, a source, release time, an original text and a credibility weight.
3. 3. The intelligent analysis method for industrial electrical response information based on large language model according to claim 1, wherein the step S2 comprises: S2.1, constructing a domain knowledge enhancement model based on a pre-training language model, and performing instruction fine adjustment on the general large model by using a power industry knowledge graph, a power industry standard file and industry expert experience to identify key entities in the result of the step S1; S2.2, inputting the result of the step S1 into the domain knowledge enhancement model, identifying key information events and the type of the power utilization effect influencing electricity utilization, and outputting the power utilization effect and key entities; S2.3, carrying out structured packaging on the information obtained by identification.
4. 4. The intelligent analysis method for industrial electrical response information based on large language model according to claim 1, wherein the step S3 comprises: s3.1, forming a thinking link template by adopting a predefined thinking link as a prompt word based on the result of the step S2; S3.2, analyzing the generated multiple thinking chains, extracting each step in the chains as a causal node, wherein the node content is a simplified causal fact description, the relation between adjacent nodes is defined as directed edges, the direction of the edges represents the causal transmission direction, and causal element extraction and node relation pairs are constructed; And S3.3, fusing all nodes and edges deduced from different events to form a directed acyclic graph, and constructing a conduction path from information to electricity.
5. 5. The intelligent analysis method for industrial electrical response information based on large language model according to claim 1, wherein the power impact event vector comprises an event unique identifier, an event title, an event source, an event release time, geographical location information, industry classification information, load trend information, time range information and label information.
6. 6. The intelligent analysis method for industrial electric response information based on the large language model according to claim 5, wherein the geographic position information comprises event positioning area information and event positioning province information, the industry classification information comprises industry names, the load trend information comprises load index types, load change directions and influence intensities, the time range information comprises time influence ranges, and the tag information comprises event influence types, specific influence matters, influence logic chain output and supportive original text summaries.
7. 7. The intelligent analysis method for industrial electrical response information based on large language model according to claim 1, wherein the result of S4 is output in JSON/XML format through RESTful API.
8. 8. An industrial electric response information intelligent analysis system based on a large language model is characterized by comprising: the collecting module is used for collecting power consumption information data and preprocessing the collected data; The processing module is used for constructing a semantic understanding and influence recognition model based on field knowledge enhancement, and extracting key entities, attributes, relationships and numerical values by inputting the result of the collecting module into the constructed semantic understanding and influence recognition model; the construction module is used for constructing an electric power influence path based on the result of the processing module; And the integration module integrates the power influence paths constructed in the construction module, integrates the standardized power influence event vectors and outputs the normalized power influence event vectors.
9. 9. An industrial electric response information intelligent analysis system based on a large language model, which is characterized by comprising a memory and one or more processors, wherein executable codes are stored in the memory, and the one or more processors are used for realizing the industrial electric response information intelligent analysis method based on the large language model according to any one of claims 1 to 7 when the executable codes are executed.

Description

Industrial electric response information intelligent analysis method and system based on large language model Technical Field The invention belongs to the technical field of artificial intelligence and energy power intersection, and particularly relates to an industrial electronic information intelligent analysis method and system based on a large language model. Background Along with the promotion of 'double carbon' targets and the construction of novel power systems, the refined prediction and regulation of power supply and demand balance are increasingly important. Industrial electricity accounts for more than 80% of the electricity consumption of the whole society, and fluctuation of the industrial electricity is driven by multiple factors such as macroscopic economy, industrial policy, emergency, technical upgrading, international trade and the like. However, existing power load prediction systems rely on historical power usage data, meteorological data, and preset macro economic indicators, and lack sensing and evaluation capabilities for the real-time impact of "unstructured text information". The current public opinion monitoring system or NLP information extraction tool is mainly focused on applications such as emotion analysis, topic classification, keyword extraction and the like, and a causal reasoning chain of text event, industrial behavior and electricity utilization change cannot be established. For example, in a news of "release high energy consumption industry limit notification" in a certain province, the existing model can only identify keywords such as "limit production", "high energy consumption", but cannot infer that the keywords will cause specific electricity consumption effects such as "electricity consumption reduction in steel industry", "regional power grid load reduction", and the like. In addition, existing models are typically analyzed after an event occurs and is reflected on electricity usage data, and lack predictive capabilities for sudden, aperiodic electricity usage shocks (e.g., sudden load management measures, large cultural activities, etc.). Finally, the existing method lacks a structured and machine-readable output format, so that the analysis result is difficult to be automatically invoked and integrated by a power grid dispatching system, an energy trading platform and other downstream systems. Therefore, an intelligent system capable of automatically understanding text semantics, reasoning influence paths and outputting standardized labels is needed, and end-to-end automatic analysis of 'information and electricity influence' is realized. Disclosure of Invention The invention aims to provide an industrial electronic information intelligent analysis method and system based on a large language model, which solve the problems that unstructured information cannot be automatically identified in the prior art, influence on industrial electricity is not achieved, influence path reasoning is lacked, an output result is unstructured and the like. In order to achieve the above purpose, the invention provides an industrial electric response information intelligent analysis method based on a large language model, which comprises the following steps: s1, collecting power consumption information data, and preprocessing the collected data; s2, constructing a semantic understanding and influence recognition model based on field knowledge enhancement, and inputting the result of the step S1 into the constructed semantic understanding and influence recognition model to extract key entities, attributes, relations and numerical values; s3, constructing a power influence path based on the result of the step S2; and S4, integrating the power influence paths constructed in the step S3, and outputting the normalized power influence event vectors. The application can realize at least the following effects: (1) Automatically identifying whether the information affects industrial electricity utilization; (2) Automatically reasoning the conductive path of the mobile phone; (3) Automatically generating a structured label containing industry, region, time, direction/amplitude influence and other dimensions; (4) And supporting the machine calling and the visual early warning of the structured output. Further, the step S1 includes: s1.1, defining information sources related to power consumption, including but not limited to government web portals (policy release), mainstream news media (economic and social news), social media platforms (public moods, hot topics, sudden public opinion), industry forums (industry dynamics), marketing company notices (important enterprise wind directions), weather information platforms and the like; s1.2, periodically collecting multi-mode heterogeneous information data in an information source related to power consumption by utilizing a plurality of input modes such as regular crawling of a web crawler, subscription of an API (application program interface), ma