Industrial electricity consumption influence information intelligent analysis method and system based on large language model

Abstract

The invention discloses an industrial power consumption influence information intelligent analysis method and system based on a large language model, belongs to the technical field of artificial intelligence and energy power crossing, and aims to solve the problem of low efficiency caused by the fact that the influence of unstructured information on industrial power consumption cannot be automatically identified in the prior art. The invention provides an industrial electricity consumption influence information intelligent analysis method based on a large language model. The method comprises the steps that power consumption information data are collected and preprocessed; constructing a semantic understanding and influence recognition model based on domain knowledge enhancement, and extracting key entities, attributes, relationships and numerical values by inputting preprocessed data; and constructing a power influence path, and integrating and outputting the standardized power influence event vectors. According to the invention, signals can be captured from network information, and early warning and analysis of electricity impact can be realized; the limitation of traditional numerical data is broken through, and the decision information dimension is enriched; the method can be connected to an existing power business system, and has extremely high engineering application value and practicability.

Description

Technical Field

[0001] This invention belongs to the field of interdisciplinary technology of artificial intelligence and energy and power, specifically involving an intelligent analysis method and system for industrial electricity consumption impact information based on a large language model. Background Technology

[0002] With the advancement of "dual-carbon" goals and the construction of new power systems, the precise forecasting and regulation of electricity supply and demand balance is becoming increasingly important. Industrial electricity consumption accounts for more than 80% of total social electricity consumption, and its fluctuations are driven by multiple factors such as macroeconomic conditions, industrial policies, unforeseen events, technological upgrades, and international trade. However, existing power load forecasting systems largely rely on historical electricity consumption data, meteorological data, and preset macroeconomic indicators, lacking the ability to perceive and assess the real-time impact of "unstructured text information."

[0003] Current public opinion monitoring systems or NLP information extraction tools mainly focus on applications such as sentiment analysis, topic classification, and keyword extraction, and cannot establish a causal reasoning chain of "text event → industry behavior → electricity consumption change". For example, for a news report that "a province issued a notice to restrict production in high-energy-consuming industries", existing models can only identify keywords such as "production restriction" and "high energy consumption", but cannot infer the specific electricity consumption impacts such as "decrease in electricity consumption in the steel industry" and "reduction in regional power grid load".

[0004] Furthermore, existing models typically only perform analysis after an event occurs and is reflected in electricity consumption data, lacking the ability to predict sudden, non-periodic electricity consumption shocks (such as sudden load management measures, large-scale cultural events, etc.).

[0005] Finally, the lack of a structured, machine-readable output format in existing methods makes it difficult for the analysis results to be automatically called and integrated by downstream systems such as power grid dispatching systems and energy trading platforms.

[0006] Therefore, there is an urgent need for an intelligent system that can automatically understand text semantics, infer the impact path, and output standardized labels to achieve end-to-end automated analysis of "intelligence information → electricity consumption impact". Summary of the Invention

[0007] This invention aims to provide an intelligent analysis method and system for industrial electricity consumption impact information based on a large language model, solving problems in existing technologies such as the inability to automatically identify the impact of unstructured information on industrial electricity consumption, lack of impact path reasoning, and unstructured output results.

[0008] To achieve the above objectives, this invention proposes an intelligent analysis method for industrial electricity consumption impact information based on a large language model, comprising the following steps: S1, collect electricity consumption information data and preprocess the collected data; S2, construct a semantic understanding and influence recognition model based on domain knowledge enhancement, and extract key entities, attributes, relationships and values ​​by inputting the results of step S1 into the constructed semantic understanding and influence recognition model; S3, Based on the results of step S2, construct the power impact path; S4 integrates the power impact paths constructed in S3 with standardized power impact event vectors and outputs them.

[0009] This application can achieve at least the following effects: (1) Automatically identify whether intelligence information affects industrial electricity consumption; (2) The transmission path of the impact of automatic reasoning information on electricity consumption; (3) Automatically generate structured tags containing dimensions such as industry, region, time, and direction / amplitude of influence; (4) Supports structured output for machine calls and visual early warnings.

[0010] Further, step S1 includes: S1.1 defines information sources related to electricity consumption as including but not limited to: government portals (policy releases), mainstream news media (economic and social news), social media platforms (public sentiment, hot topics, and public opinion emergencies), industry forums (industry dynamics), listed company announcements (key enterprise trends), and meteorological information platforms, etc. S1.2, using multiple input methods such as web crawler timed crawling, API interface subscription, and manual uploading, periodically collect multimodal heterogeneous information data from information sources related to power consumption; S1.3 Preprocess the collected data, such as: text deduplication, HTML tag cleaning, language recognition (generally Chinese and English), word segmentation and entity extraction, timestamp extraction, and source credibility scoring (e.g., .gov>.org>.com>others); S1.4 Output standardized information, which includes information ID, source, publication time, original text, and credibility weight.

[0011] Further, step S2 includes: S2.1 Construct a domain knowledge enhancement model based on a pre-trained language model. Use high-quality corpora such as the power industry knowledge graph, power industry standard documents (policy documents, industry reports, dispatch logs, etc.), and industry expert experience (influencing factors - electricity consumption impact - degree of impact) to fine-tune the general large model, so that it has a high sensitivity and understanding of power-related concepts (such as orderly electricity consumption and peak summer demand), and can identify key entities in the results of step S1. S2.2 Input the results of step S1 into the domain knowledge enhancement model to identify key information events affecting electricity consumption and the types of electricity consumption impacts, and output the electricity consumption impacts and key entities: positive (electricity consumption ↑), negative (electricity consumption ↓), neutral, uncertain; S2.3. The identified information is encapsulated in a structured manner, including: industry name, policy, region, time point, and numerical information (such as "production limit of 30%" and "electricity price increase of 0.1 yuan / kWh").

[0012] Further, step S3 includes: S3.1, Based on the results of step S2, including impact identification information (electricity impact category, key entity information, etc.), a predefined thought chain is used as a prompt to form a thought chain template, which includes the following: Industry classification system (GB / T 4754 standard); Expert experience path; Typical electrical equipment and load characteristics in the industry (e.g., electrolytic aluminum → rectifier transformer → continuous high load); The upstream and downstream relationships in the industrial chain (e.g., "steel → construction industry → cement → electricity demand"); Heat map showing regional power grid division and industry distribution; Electricity usage patterns during seasonal periods, holidays, and weekdays.

[0013] Taking typical meteorological information as an example, the model output example is as follows: Input event: "A city issued a red alert for high temperatures reaching 40℃ tomorrow." Outputting thought chains: 1) Event: Red alert for high temperatures 2) Direct impact: Residential users' demand for cooling has increased dramatically. 3) Indirect impact: The number of air conditioners, electric fans, and other cooling equipment being used and their operating time have increased significantly. 4) Indirect impact: To maintain environmental comfort, commercial venues (such as shopping malls and office buildings) increase the load on central air conditioning systems. 5) Final impact: A significant increase in electricity load for both residential and commercial users throughout the city. S3.2 Analyze the generated multiple thought chains, extract each step in the chain as a causal node, the node content is a concise description of causal facts, define the relationship between adjacent nodes as a directed edge, the direction of the edge represents the direction of causal transmission, and construct the causal element extraction and node relationship pair.

[0014] Continuing from the previous example, construct the nodes and edges: Node A: "Red Alert for High Temperatures" Node B: "Residents' demand for cooling has increased dramatically" Node C: "Increased use of cooling equipment such as air conditioners and electric fans". Node D: Increased central air conditioning load in commercial venues Node E: "Citywide residential and commercial electricity load has increased significantly." Directed edges: A→B, B→C, C→E, C→D, D→E S3.3 merges all nodes and edges inferred from different events to form a global, unified directed acyclic graph, and constructs the transmission path from information to electricity consumption based on this graph.

[0015] It utilizes the large-scale model thinking chain capability to achieve reasoning and construction of a directed acyclic graph from information events to the final change in electricity load.

[0016] Furthermore, the power impact event vector in step S4 includes the following fields: The impact paths constructed in S3 are integrated into a standardized "Power Impact Event Vector," which includes the following fields: event_id: Unique identifier for the event title: Event title (e.g., news headline, policy name) Source: Event Source publish_time: Event publication time location: geographic location information location.location: Information about the event location area. location.province: Information about the province where the event was located. industry: Industry classification information industry.industry_name: Industry name load_trend: Load trend information load_trend.load_indicator: Load indicator type load_trend.direction: Direction of load change load_trend.impact_strength: Impact strength time_scope: Time range information time_scope.bucket: The scope of time impact. labels: Label information labels.category: Event impact type labels.subcategory: Specific impact factors labels.impact_pathway: Affects the output of the logical chain. labels.original_excerpt: Supporting original text abstracts.

[0017] Furthermore, the results of S4 are output in JSON / XML format via a RESTful API.

[0018] Visualization and extended application interfaces: (1) Provide a RESTful API that supports outputting "power impact event vectors" in JSON / XML format; (2) Supports integration with power grid EMS systems, energy trading platforms, policy simulation platforms, etc.; (3) Built-in visual dashboard: industry impact heat map, regional load forecast curve, impact confidence distribution map; (4) Support threshold warnings (such as “expected load decrease > 5%” or “number of industries affected > 3”) to automatically trigger email / SMS / system pop-up notifications.

[0019] An intelligent analysis system for industrial electricity consumption impact information based on a large language model includes: The data collection module collects electricity consumption information and preprocesses the collected data. The processing module constructs a semantic understanding and impact recognition model based on domain knowledge enhancement. By inputting the results of the collection module into the constructed semantic understanding and impact recognition model, key entities, attributes, relationships, and values ​​are extracted. The construction module constructs power impact paths based on the results of the processing module; The integration module integrates the power impact paths constructed in the construction module into standardized power impact event vectors and outputs them.

[0020] A smart analysis system for industrial electricity consumption impact information based on a large language model is characterized by comprising a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the smart analysis method for industrial electricity consumption impact information based on a large language model.

[0021] Beneficial effects Compared with the prior art, the present invention has the following beneficial effects: It has strong foresight and initiative: it can capture signals from network information to achieve early warning and analysis of power consumption impacts, transforming passive response into proactive management.

[0022] The analysis is broad and deep: it breaks through the limitations of traditional numerical data, incorporates massive amounts of unstructured information into the analysis scope, and greatly enriches the information dimensions for decision-making.

[0023] High standardization and integration: The output "power impact event vector" has a unified format and complete information, and can be integrated into existing power business systems, which has extremely high engineering application value and practicality. Attached Figure Description

[0024] Figure 1 This is a flowchart of Example 1; Figure 2 This is a diagram illustrating the influence of path reasoning; Figure 3 This is a schematic diagram of the vector structure of power-related events; Figure 4 This is a schematic diagram showing the results of application example 1; Figure 5 This is a schematic diagram showing the results of application example 2. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1 A method for intelligent analysis of industrial electricity consumption impact information based on a large language model, such as Figures 1 to 3 As shown, it includes the following steps: S1, collects electricity consumption information data and preprocesses the collected data, including: S1.1 defines information sources related to electricity consumption, including but not limited to: government portals, mainstream news media, social media platforms, industry forums, listed company announcements, and meteorological information platforms; S1.2, Periodically collect heterogeneous information data from information sources related to electricity consumption; S1.3, Preprocess the collected data; S1.4 Output standardized information, which includes information ID, source, publication time, original text, and credibility weight.

[0027] S2, construct a semantic understanding and influence recognition model based on domain knowledge enhancement. By inputting the results of step S1 into the constructed semantic understanding and influence recognition model, key entities, attributes, relationships, and values ​​are extracted; including: S2.1 Construct a domain knowledge enhancement model based on a pre-trained language model. Use the power industry knowledge graph, power industry standard documents, and industry expert experience to fine-tune the general large model with instructions, which can identify key entities in the results of step S1. S2.2 Input the results of step S1 into the domain knowledge enhancement model to identify key information events affecting electricity consumption and the types of electricity consumption impacts, and output the electricity consumption impacts and key entities; S2.3, The information obtained through identification is encapsulated in a structured manner.

[0028] S3, based on the results of step S2, construct the power impact path; including: S3.1 Based on the results of step S2, a predefined thought chain is used as a prompt word to form a thought chain template; S3.2, parse the generated multiple thought chains, extract each step in the chain as a causal node, the node content is a concise description of causal facts, define the relationship between adjacent nodes as a directed edge, the direction of the edge represents the direction of causal transmission, and construct the causal element extraction and node relationship pair. S3.3 merges all nodes and edges inferred from different events to form a directed acyclic graph, and constructs the transmission path from information to electricity consumption based on this graph.

[0029] S4 integrates the power impact paths constructed in S3 with standardized power impact event vectors and outputs them in JSON / XML format via a RESTful API.

[0030] The power impact event vector includes a unique event identifier, event title, event source, event release time, geographic location information, industry classification information, load trend information, time range information, and tag information. The geographic location information includes the event's location area and province; the industry classification information includes the industry name; the load trend information includes the load index type, load change direction, and impact intensity; the time range information includes the time impact range; and the tag information includes the event impact type, specific impact items, impact logic chain output, and supporting original text summary.

[0031] Example 2 An intelligent analysis system for industrial electricity consumption impact information based on a large language model includes: The data collection module collects electricity consumption information and preprocesses the collected data. The processing module constructs a semantic understanding and impact recognition model based on domain knowledge enhancement. By inputting the results of the collection module into the constructed semantic understanding and impact recognition model, key entities, attributes, relationships, and values ​​are extracted. The construction module constructs power impact paths based on the results of the processing module; The integration module integrates the power impact paths constructed in the construction module into standardized power impact event vectors and outputs them.

[0032] Example 3 An intelligent analysis system for industrial electricity consumption impact information based on a large language model includes a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the intelligent analysis method for industrial electricity consumption impact information based on a large language model as described in Embodiment 1.

[0033] The present invention will be further described below with reference to two application examples, but these do not constitute a limitation on the present invention. The two application examples are information that affects electricity consumption and information that does not affect electricity consumption.

[0034] Application Example 1: Analysis of Related Information Step S1: The acquisition module captures the original news article and extracts the release event, source, and text. In this example, we take the news article "Weak Demand Dominates Coking Coal and Coke Futures Prices Under Pressure" published by China Steel News Network on March 21, 2024 at 15:16 as an example.

[0035] Step S2: LLM identifies keywords: "steel", "production cut", "weak demand", "below expectations", and determines "impact on industrial electricity consumption", with the direction being "negative".

[0036] Step S3: Intelligent Construction of Power Impact Paths Industries: Ferrous metal smelting and rolling processing, petroleum, coal and other fuel processing; Region: Nationwide; Impact Path: 1) Weak end-user demand → Most coking and steel enterprises are in a loss-making stage → They take the initiative to limit production → Production load decreases → Electricity demand decreases; 2) The capacity utilization rate of all independent coking enterprises in the sample decreases → Coke supply is close to the historical low → Production activities decrease → Electricity load decreases; 3) Stricter coal mine safety production supervision policies → Coal mine production is restricted → Coal supply tightens marginally → Production activities are restricted → Electricity demand decreases; 4) Weak end-user demand → Steel demand has a small pulling effect → Enterprises reduce their production enthusiasm → Electricity demand decreases; 5) The profit margin of the entire industrial chain shrinks → Enterprises take the initiative to reduce production load → Production activities decrease → Electricity demand decreases.

[0037] Step S4: Output of structured tags Output power impact event vectors in JSON format, such as Figure 4 As shown.

[0038] Application Example 2: Analysis of Unrelated Information Step S1: The data collection module captures the original news article and extracts the publication event, source, and text. This example uses the news article "Li Zhongshuang predicts that the hot-rolled coil market may stabilize in the later period, but cold-rolled sheet prices may be subject to a correction" published by China Metallurgical News at 7:00 on February 23, 2024.

[0039] Step S2: LLM identifies keywords: No common key information was captured.

[0040] Step S3: The path reasoning module calls the knowledge graph: Industry: Ferrous metal smelting and rolling processing; Region: East China; Impact path: If the information is determined to be non-load-related, output standard unrelated information.

[0041] Step S4: Output of structured tags Output power impact event vectors in JSON format, such as Figure 5 As shown.

[0042] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent analysis of industrial electricity consumption impact information based on a large language model, characterized in that, Includes the following steps: S1, collect electricity consumption information data and preprocess the collected data; S2, construct a semantic understanding and influence recognition model based on domain knowledge enhancement, and extract key entities, attributes, relationships and values ​​by inputting the results of step S1 into the constructed semantic understanding and influence recognition model; S3, Based on the results of step S2, construct the power impact path; S4 integrates the power impact paths constructed in S3 with standardized power impact event vectors and outputs them.

2. The intelligent analysis method for industrial electricity consumption impact information based on a large language model according to claim 1, characterized in that, Step S1 includes: S1.1 defines information sources related to electricity consumption, including but not limited to: government portals, mainstream news media, social media platforms, industry forums, listed company announcements, and meteorological information platforms; S1.2, Periodically collect multimodal heterogeneous information data from information sources related to power consumption; S1.3, Preprocess the collected data; S1.4 Output standardized information, which includes information ID, source, publication time, original text, and credibility weight.

3. The intelligent analysis method for industrial electricity consumption impact information based on a large language model according to claim 1, characterized in that, Step S2 includes: S2.1 Construct a domain knowledge enhancement model based on a pre-trained language model. Use the power industry knowledge graph, power industry standard documents, and industry expert experience to fine-tune the general large model with instructions, which can identify key entities in the results of step S1. S2.2 Input the results of step S1 into the domain knowledge enhancement model to identify key information events affecting electricity consumption and the types of electricity consumption impacts, and output the electricity consumption impacts and key entities; S2.3, The information obtained through identification is encapsulated in a structured manner.

4. The intelligent analysis method for industrial electricity consumption impact information based on a large language model according to claim 1, characterized in that, Step S3 includes: S3.1 Based on the results of step S2, a predefined thought chain is used as a prompt word to form a thought chain template; S3.2, parse the generated multiple thought chains, extract each step in the chain as a causal node, the node content is a concise description of causal facts, define the relationship between adjacent nodes as a directed edge, the direction of the edge represents the direction of causal transmission, and construct the causal element extraction and node relationship pair. S3.3 merges all nodes and edges inferred from different events to form a directed acyclic graph, and constructs the transmission path from information to electricity consumption based on this graph.

5. The intelligent analysis method for industrial electricity consumption impact information based on a large language model according to claim 1, characterized in that, The power impact event vector includes a unique event identifier, event title, event source, event release time, geographic location information, industry classification information, load trend information, time range information, and tag information.

6. The intelligent analysis method for industrial electricity consumption impact information based on a large language model according to claim 5, characterized in that, The geographic location information includes the event location area information and the event location province information; the industry classification information includes the industry name; the load trend information includes the load index type, load change direction, and impact intensity; the time range information includes the time impact range; and the tag information includes the event impact type, specific impact items, impact logic chain output, and supporting original text summary.

7. The intelligent analysis method for industrial electricity consumption impact information based on a large language model according to claim 1, characterized in that, The results of S4 are output in JSON / XML format via a RESTful API.

8. A smart analysis system for industrial electricity consumption impact information based on a large language model, characterized in that, include: The data collection module collects electricity consumption information and preprocesses the collected data. The processing module constructs a semantic understanding and impact recognition model based on domain knowledge enhancement. By inputting the results of the collection module into the constructed semantic understanding and impact recognition model, key entities, attributes, relationships, and values ​​are extracted. The construction module constructs power impact paths based on the results of the processing module; The integration module integrates the power impact paths constructed in the construction module into standardized power impact event vectors and outputs them.

9. A smart analysis system for industrial electricity consumption impact information based on a large language model, characterized in that, The device includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the intelligent analysis method for industrial electricity consumption impact information based on a large language model as described in any one of claims 1 to 7.

Images