CN-121998695-A - Electric power supply and demand factor prediction method and system

CN121998695ACN 121998695 ACN121998695 ACN 121998695ACN-121998695-A

Abstract

The invention provides a power supply and demand factor prediction method and a power supply and demand factor prediction system, wherein the method comprises the steps of obtaining unstructured text of a power supply and demand factor influence event, constructing a text analysis model based on an LLM semantic understanding engine, converting the unstructured text into a structure standardization event description, analyzing the power supply and demand factor influence intensity according to the structure standardization event description through an influence intensity mapping mechanism based on context awareness, constructing a dynamic gating fusion model based on a real-time power supply and demand situation, carrying out feature fusion on input parameters to obtain a power supply and demand factor feature vector, and carrying out power supply and demand factor prediction according to the power supply and demand factor feature vector by adopting a power supply and demand factor prediction model based on a composite loss function to obtain a prediction result. According to the invention, the text analysis capability realizes qualitative leap, the event influence quantification is more accurate and scientific, the feature fusion mechanism is more intelligent and flexible, and the prediction target is more fit with the actual demand of the electric power market.

Inventors

DAI CHANGCHUN
LI YONGBO
Qian Hanhan
QI HUI
ZHANG WEISHI
JI CHAO
HAO YUXING
XIE DAOQING
FU JINGYU
ZHOU TAO
JIANG HAILONG
ZHAO XUETING
ZHANG WEI
Cheng Honggu
LIN ZHEMIN

Assignees

安徽电力交易中心有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (10)

1. A method of predicting a power supply and demand factor, the method comprising: Obtaining unstructured text of an electric power supply and demand factor influence event; Constructing a text analysis model based on an LLM semantic understanding engine, and converting the unstructured text into a structure standardized event description by using the text analysis model; Analyzing the power supply and demand factor influence intensity according to the structure standardized event description through an influence intensity mapping mechanism based on context awareness; Constructing a dynamic gating fusion model based on a real-time power supply and demand situation, and carrying out feature fusion on the structural standardized event description and the power supply and demand factor influence intensity as input parameters to obtain a power supply and demand factor feature vector; And adopting a power supply and demand factor prediction model based on a composite loss function, and performing power supply and demand factor prediction according to the power supply and demand factor feature vector to obtain a prediction result.
2. The method of claim 1, wherein the constructing a text parsing model based on the LLM semantic understanding engine comprises: Constructing a power system domain knowledge constraint base containing various power professional rules; Designing a structured output template and a domain knowledge enhancement prompt, and encoding the power professional rule into the domain knowledge enhancement prompt, and/or embedding a typical sample example in the domain knowledge enhancement prompt; and performing reinforcement learning on the LLM semantic understanding engine based on the structured output template and the domain knowledge reinforcement prompt to obtain a text analysis model passing format verification and/or range check.
3. The method of claim 1, wherein the power system domain knowledge constraint library comprises at least one of power system operation rules, event classification hierarchy, event impact logic rules, pre-warning level mapping rules, unit normalization specifications, and zone name normalization rules.
4. The method of claim 1, wherein said converting said unstructured text into a structured normalized event description using said text parsing model comprises: Inputting the unstructured text into the text analysis model, and outputting a structure standardized event description containing a plurality of key event information; wherein the plurality of key event information includes at least one of event type, direction of impact, region of action, duration, and intensity level.
5. The method of claim 1, wherein the constructing a context-aware based impact strength mapping mechanism that analyzes power supply and demand factor impact strength from a structure-normalized event description comprises: carrying out fine-granularity semantic analysis on the structure standardized event description by using a large language model, and converting discrete intensity levels into continuous influence values reflecting actual influence degrees; and normalizing the continuous influence values by combining the running state of the power system to obtain the influence intensity of the electric power supply and demand factor influence event on the electric power supply and demand factor.
6. The method according to claim 1, wherein the constructing a dynamic gating fusion model based on real-time power supply and demand situation, performing feature fusion with the structure standardized event description and the power supply and demand factor influence intensity as input parameters, and obtaining a power supply and demand factor feature vector, includes: constructing an influence intensity gating model of an integrated input layer, a full connection layer, a batch normalization layer, an activation function layer and an output layer; And calculating the dynamic weight of the input parameter by using the influence intensity gating model and taking a real-time power supply and demand situation as a regulation basis, weighting event characteristics corresponding to the structure standardized event description according to the dynamic weight, and splicing continuous variable characteristics corresponding to the electric power supply and demand factor influence intensity to obtain a low-dimensional fusion vector serving as the electric power supply and demand factor characteristic vector.
7. The method of claim 6, wherein the input layer is configured to receive the input parameters, wherein the full connection layer is configured to map the input parameters to a hidden space to obtain feature vectors, wherein the batch normalization layer is configured to normalize the feature vectors, wherein the activation function layer is configured to introduce a nonlinear transformation, and wherein the output layer is configured to generate dynamic weights within a [0,1] interval for the input parameters.
8. The method of claim 6, wherein the method further comprises: Generating an auxiliary context vector of the unstructured text by using a large language model so as to judge the real-time power supply and demand situation of the current power system; wherein the real-time power supply and demand situation includes supply and demand, supply and demand and/or balance state.
9. The method according to any one of claims 1-8, wherein the composite loss function comprises a predictive loss term based on mean square error and a feature sparsity constraint term based on L1 regularization.
10. A power supply and demand factor prediction system, the system comprising: the text acquisition module is used for acquiring unstructured text of the electric power supply and demand factor influence event; the standard processing module is used for constructing a text analysis model based on the LLM semantic understanding engine and converting the unstructured text into a structure standardized event description by utilizing the text analysis model; The continuous processing module is used for analyzing the influence intensity of the power supply and demand factors according to the structure standardized event description through an influence intensity mapping mechanism based on context awareness; The feature fusion module is used for constructing a dynamic gating fusion model based on a real-time power supply and demand situation, and carrying out feature fusion on the structure standardized event description and the power supply and demand factor influence intensity as input parameters to obtain a power supply and demand factor feature vector; And the result prediction module is used for predicting the power supply and demand factors according to the power supply and demand factor characteristic vector by adopting a power supply and demand factor prediction model based on the composite loss function to obtain a prediction result.

Description

Electric power supply and demand factor prediction method and system Technical Field The invention relates to the technical field of intersection of artificial intelligence and an electric power system, in particular to a method and a system for predicting an electric power supply and demand factor. Background A Power Supply-demand factor (Power Supply-DEMAND GAP, PSDG) is defined as a core index of the operation of the Power market, as a difference between the total Power demand of the system and the Power Supply of the non-thermal Power unit, that is, bid space=total Power demand-total Power Supply of the non-thermal Power unit. The index directly reflects the market bidding space which needs to be filled by a thermal power unit or other adjustable power supplies in the power system, and is a key basis for determining the daily/real-time market price, the unit start-stop plan and auxiliary service scheduling. In recent years, power systems are undergoing profound changes. By the end of 2024, the national renewable energy installed capacity has broken through 14 hundred million kilowatts, accounting for more than 50% of the total assembly machine, wherein the wind power and photovoltaic power generation installation respectively reaches 4.4 hundred million kilowatts and 6.1 hundred million kilowatts. The large-scale and high-proportion access of new energy makes the supply and demand balance mechanism of the power system change fundamentally, on one hand, the new energy output has strong randomness and volatility under the influence of the meteorological conditions, and on the other hand, the traditional load characteristics are more complicated due to novel loads such as distributed energy sources and electric automobiles. These variations result in increased fluctuations in the power supply and demand factor, exhibiting strong uncertainties and event-driven characteristics. In this context, power market operations face unprecedented challenges. Taking a provincial power grid in the eastern China as an example, in a typical day in 2024 summer, due to the sudden drop of photovoltaic output and the high-temperature load increase, the power supply and demand factor is suddenly changed from-2000 MW (supply and demand) to +5000MW (supply and demand) within 2 hours, so that the real-time market price fluctuation exceeds 500 yuan/MWh. The severe fluctuation not only increases the decision difficulty of market participants, but also threatens the safe and stable operation of the power grid. Therefore, the power supply and demand factor change trend is accurately predicted, the supply and demand unbalance risk is identified in advance, and the method becomes an urgent need of a modern power market intelligent decision system. However, power supply and demand factor prediction faces multiple technical challenges. Besides the conventional time sequence mode, a large number of external events break the original balance by changing the two ends of supply and demand, namely, the weather early warning influences the form of a load curve, the holiday adjustment and rest change the power utilization mode, the new energy output limited bulletin directly compresses the non-thermal power supply capacity, and the power grid overhaul plan influences the power transmission capacity. Most of the event information exists in the form of unstructured text in the channels of dispatch logs, weather services, government notices and the like, and the traditional method is difficult to effectively extract and utilize. Although artificial intelligence technology has made remarkable progress in power system analysis in recent years, particularly a deep learning model is excellent in the field of load prediction, a systematic feature representation method is still lacking for a core variable reflecting a market bidding space, namely a power supply and demand factor. Breakthrough of Large Language Model (LLM) provides new ideas for solving this problem. LLM exhibits powerful capabilities in terms of semantic understanding, text generation, and knowledge reasoning, and is capable of extracting key information from unstructured text and performing context-aware semantic analysis. However, the general large model lacks the expertise of the electric power system, and the direct application to the electric power market scene can cause the problems of inaccurate information extraction, deviation of understanding of the technical terms and the like. How to effectively combine the powerful language understanding capability of a large model with the expertise of a power system and construct a special feature representation method facing to the prediction of power supply and demand factors becomes the leading direction of the current research. Disclosure of Invention Therefore, the invention provides a method and a system for predicting power supply and demand factors, which aim to solve the technical problems of insufficient utilization of unstructu