CN-122022058-A - Power energy carbon pollution cooperative optimization method and system based on supply and demand matching and large model

CN122022058ACN 122022058 ACN122022058 ACN 122022058ACN-122022058-A

Abstract

The invention relates to the technical field of clean production and pollution reduction and carbon reduction in the power industry, and discloses a power energy carbon pollution cooperative optimization method and system based on supply and demand matching and a large model, wherein the method comprises the steps of constructing a power supply and demand matching heterogeneous knowledge base, analyzing the current production state of real-time operation data, and generating process demand risk description; matching candidate regulation and control schemes through semantic retrieval, constructing multiple intelligent agents, and iteratively generating a regulation and control strategy; the invention can accurately match and reuse the history optimal regulation scheme through multi-agent game and large language model retrieval, realize the collaborative optimization of carbon pollution, ensure stable production, realize continuous self-adaptive updating of the knowledge base and effectively reduce running cost and carbon footprint.

Inventors

LIN XIAOQING
WANG REN
CHEN JIE
Tang Chenyue
YU HONG
HUANG QUNXING
LI XIAODONG
YAN JIANHUA

Assignees

浙江大学

Dates

Publication Date: 20260512
Application Date: 20260209

Claims (10)

1. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model is characterized by comprising the following steps of: Collecting historical operation data of a power plant, extracting historical operation fragments meeting evaluation of carbon pollution benefits, and constructing a power supply and demand matching heterogeneous knowledge base containing working condition requirements, technical supply and carbon pollution benefit triples; Collecting real-time operation data of a power plant, carrying out semantic analysis on the current production state by using a large language model, and generating process demand risk description; Converting the process demand risk description into a query vector, carrying out semantic retrieval on the power supply and demand matching heterogeneous knowledge base, and matching and adapting to a candidate regulation and control technical scheme set of the process demand risk description; constructing intelligent agents comprising energy efficiency guarantee, pollution control and carbon emission optimization and objective functions corresponding to the intelligent agents based on a candidate regulation and control technical scheme set, and generating a regulation and control strategy of the intelligent agents through iterative solution; based on the arbitration big model, conflict resolution and balance are carried out on the regulation and control strategies of all the intelligent agents, and a cooperative optimization instruction considering clean production and pollution reduction and carbon reduction targets is generated; Executing the collaborative optimization instruction, acquiring the actual carbon-pollution benefits after executing the instruction, and if the actual carbon-pollution benefits are smaller than a preset threshold, recording the current corresponding process demand risk description, the collaborative optimization instruction and the actual carbon-pollution benefits so as to update the power supply and demand matching heterogeneous knowledge base and realize self-adaptive life learning and carbon-pollution collaborative optimization of the power system.
2. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 1, wherein the power plant history operation data is collected, history operation fragments meeting energy and pollution benefit evaluation are extracted, and a power supply and demand matching heterogeneous knowledge base comprising working condition requirements, technical supply and energy and pollution benefit triplet mapping is constructed by the following steps: collecting historical data of a distributed control system of a power plant generator set, emission data of a smoke emission continuous monitoring system and carbon accounting report data, and obtaining historical operation data of the power plant; After cleaning the power plant historical operation data, extracting historical operation fragments meeting the evaluation of carbon and pollution benefits; The method comprises the steps of carrying out structural extraction on historical operation fragments by using a large language model, extracting working condition characteristics as working condition demand side vectors, extracting operation parameter adjustment and process means as technical supply side vectors, extracting steam yield change, pollutant concentration change and carbon emission intensity change as carbon pollution benefit labels, storing the working condition demand side vectors, the technical supply side vectors and the carbon pollution benefit labels into a vector database, and generating an electric power supply and demand matching heterogeneous knowledge base with working condition demand, technical supply and carbon pollution benefit triples.
3. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 1, wherein the real-time operation data of a power plant is collected, the large language model is utilized to carry out semantic analysis on the current production state, and the process of generating the process demand risk description is as follows: Configuring a large language model with data and text semantic mapping, carrying out knowledge enhancement on the large language model by loading a historical operation log and a fault report of the power industry, generating an enhanced large language model, and simultaneously creating a process diagnosis expert role; Collecting time sequence data containing the temperature of a hearth, the oxygen content of flue gas, the speed of a fire grate and the real-time concentration of pollutants in a current time window, and converting the time sequence data into a structured prompt text containing the names of working condition parameters and corresponding time sequence values; And inputting the structured prompt text into a process diagnosis expert role, analyzing the coupling risk among working condition requirements, technical supply and carbon pollution benefits in the structured prompt text, and outputting the industrial requirement risk description in a natural language form.
4. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 1, wherein the process of converting process demand risk description into query vectors, performing semantic retrieval on a power supply and demand matching heterogeneous knowledge base, and matching a candidate regulation and control technical scheme set of the process demand risk description is as follows: Converting the process demand risk description into a query vector; calculating cosine similarity of the query vector and all working condition demand side vectors in the power supply and demand matching heterogeneous knowledge base, and selecting Top-N working condition demand side vectors with highest cosine similarity score; and screening the technical supply side vectors corresponding to the Top-N working condition demand side vectors to obtain a candidate regulation and control technical scheme set.
5. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 1, wherein the process of constructing the intelligent agents comprising energy efficiency guarantee, pollution control and carbon emission optimization and the objective functions corresponding to the intelligent agents based on a candidate regulation and control technical scheme set and generating the regulation and control strategy of the intelligent agents through iterative solution is as follows: constructing an energy efficiency guarantee agent and a corresponding energy efficiency objective function, and generating a regulation strategy of the energy efficiency guarantee agent through iterative solution, wherein the regulation strategy comprises the following specific steps: Taking the regulation parameters in the candidate regulation technical scheme set as optimization variables and maximizing the load and steam stability as optimization targets, and constructing an energy efficiency objective function, namely: Wherein, the In order to be an energy efficient objective function, In order to take the minimum value of the values, 、 Are all the weight numbers of the energy efficiency, In order to effectively ensure the steam flow predicted value of the intelligent agent under the current regulation strategy, Is a steam flow set value under the rated working condition, The real-time load rate of the unit is set; Adopting a gradient descent method or a heuristic search method to iteratively adjust each optimized variable, and outputting an aggressive combustion regulation and control parameter when an energy efficiency objective function reaches a minimum value, wherein the aggressive combustion regulation and control parameter is used as a regulation and control strategy for energy efficiency assurance intelligent bodies; constructing a pollution control intelligent agent and a corresponding pollution control objective function, and generating a regulation strategy of the pollution control intelligent agent through iterative solution, wherein the regulation strategy comprises the following specific steps: Taking the minimum pollutant emission concentration as an optimization target, taking an aggressive combustion regulation parameter as input, taking the correction quantity of the aggressive combustion regulation parameter as an optimization variable, applying punishment to the aggressive combustion regulation parameter which leads to the pollutant emission concentration exceeding standard, and constructing a pollution control objective function, namely: Wherein, the In order to control the objective function of the pollution, In order to be able to determine the type of contaminant, Is the first The hazard weight coefficient of the seed pollutant, Is the first The predicted emission concentration of the species of the pollutant, Is the first Environmental limits for seed contaminants; Iteratively solving a pollution control objective function, and outputting correction quantity of the aggressive combustion regulation and control parameter when the pollution control objective function reaches a minimum value to generate the combustion regulation and control parameter which takes energy efficiency and environmental protection into consideration, wherein the combustion regulation and control parameter is used as a regulation and control strategy of a pollution control intelligent body; Constructing a carbon bank optimization intelligent agent and a corresponding carbon bank optimization objective function, and generating a regulation strategy of the carbon bank optimization intelligent agent through iterative solution, wherein the regulation strategy comprises the following specific steps: Taking the carbon dioxide equivalent emission and the material consumption of the minimum unit generated energy as optimization targets, taking the combustion regulation and control parameters as optimization variables, and constructing a carbon emission optimization objective function; And iteratively solving a carbon bank optimization objective function, and outputting energy-saving regulation parameters when the carbon bank optimization objective function reaches a minimum value, wherein the energy-saving regulation parameters are used as a regulation strategy of the carbon bank optimization intelligent body.
6. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 5, wherein a calculation formula of a carbon bank optimization objective function is as follows: Wherein, the The objective function is optimized for the carbon number, For the amount of direct carbon emissions produced by the combustion of the fuel, In order to regulate the kind of the medicament, Is the first The consumption of the medicament is regulated and controlled, Is the first The carbon emission factor of the medicament is regulated and controlled, Is the power consumption of the factory, Is the average carbon emission factor of the power grid.
7. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 1, wherein the process of carrying out conflict resolution and balance on the regulation and control strategy of each agent based on the arbitration large model and generating the collaborative optimization instruction considering clean production and pollution reduction and carbon reduction targets is as follows: Taking the environment-friendly limit value and the equipment safe operation limit as hard constraints, judging whether the predicted value of the equipment operation state parameter violates the hard constraints or not by each regulation and control parameter after the regulation and control strategy of each intelligent agent is executed, if so, judging that the regulation and control strategy touches the conflict boundary, triggering a fusing mechanism, setting the game weight coefficient of the regulation and control strategy to be 0, and executing the next comprehensive benefit value calculation, otherwise, not processing and executing the next comprehensive benefit value calculation; According to the objective function corresponding to each agent, calculating the comprehensive profit value under the current strategy combination; Under the hard constraint, when the comprehensive benefit value reaches the maximum value through multiple rounds of iterative search, the regulation and control parameter combination corresponding to the regulation and control strategy is used as a cooperative optimization instruction for considering clean production and pollution reduction and carbon reduction targets.
8. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 7, wherein a formula for calculating a comprehensive benefit value under a current strategy combination is: Wherein, the In order to integrate the value of the benefit, 、、 Are all the game weight coefficients of the game, 、、 And respectively optimizing objective function values for energy efficiency, pollution control and carbon emission after normalization treatment.
9. The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model according to claim 1, wherein a calculation formula of actual energy and carbon pollution benefits is as follows: Wherein, the In order to achieve the practical benefit of carbon pollution, 、、 Are the weight coefficients of the benefit and the weight coefficient of the benefit, In order to achieve an energy efficiency improvement rate, For the carbon emission reduction rate, the method has the advantages of high yield, Is the reduction rate of the concentration of the pollutant.
10. A power energy and carbon pollution cooperative optimization system based on supply and demand matching and a large model, which is characterized by being applied to the method as claimed in any one of claims 1-9, and comprising the following steps: the multi-mode data perception module is used for collecting historical operation data of the power plant and constructing a power supply and demand matching heterogeneous knowledge base; The supply and demand matching engine module is used for carrying out semantic interpretation on the current production state by utilizing a large language model based on real-time operation data of the power plant, generating process demand risk description, and matching a candidate regulation and control technical scheme set of the process demand risk description by carrying out semantic retrieval in a power supply and demand matching heterogeneous knowledge base; The multi-agent decision center module is used for constructing an agent and arbitration large model comprising energy efficiency guarantee, pollution control and carbon emission optimization based on a candidate regulation and control technical scheme set, performing multi-agent game and inference decision, and generating a collaborative optimization instruction considering clean production and pollution reduction and carbon reduction targets; And the self-evolution closed-loop module is used for monitoring the instruction execution effect in real time based on the collaborative optimization instruction so as to realize self-adaptive updating of the power supply and demand matching heterogeneous knowledge base and collaborative optimization of carbon pollution.

Description

Power energy carbon pollution cooperative optimization method and system based on supply and demand matching and large model Technical Field The invention relates to the technical field of clean production, pollution reduction and carbon reduction in the power industry, in particular to a power energy carbon pollution cooperative optimization method and system based on supply and demand matching and a large model. Background The current power industry (covering the thermal power generation fields such as coal burning, biomass and garbage incineration power generation) faces great challenges, on one hand, the load requirement and the power generation efficiency (energy efficiency) are guaranteed, on the other hand, the emission of pollutants such as nitrogen oxides NO x, sulfur dioxide SO 2 and particulate matters is strictly controlled, pollution reduction is realized, and meanwhile, carbon emission caused by desulfurization and denitrification agent consumption and fossil energy auxiliary combustion is also required to be reduced to the greatest extent, SO that carbon reduction is realized. However, in the actual production process of electric power production, there is often a complex nonlinear coupling relationship between Energy efficiency (Energy), carbon emission (Carbon) and Pollution control (Pollution), and even a significant conflict of interests is expressed. For example, to reduce NO x emissions (environmental objectives), it is often necessary to control the combustion temperature or increase the ammonia injection amount, but this may lead to incomplete combustion and thus reduced thermal efficiency (impaired energy efficiency), or increased indirect carbon emissions (increased carbon emissions) due to excessive ammonia injection and ammonia slip. The traditional power distributed control system DCS mainly depends on PID control or manual experience adjustment, so that the problem of dynamic balance of multi-objective and strong coupling is difficult to solve, and comprehensive optimization of energy-carbon-pollution (ECP) is difficult to realize. In recent years, with the rapid development of artificial intelligence technology, large Language Models (LLM) and multi-modal technologies are beginning to be applied to the industrial field, and have been well effective in the industry complex problem. At present, large language model intelligent research in the field of electric power and related combustion power generation mainly focuses on two layers of 'knowledge extraction' and 'state identification'. The existing method solves the problem that a large model is difficult to understand industrial terms such as a sliding grate, SNCR denitration and the like, but does not relate to how to solve complex control strategy conflicts in the production process by utilizing the knowledge, and cannot directly guide real-time process optimization, and meanwhile, the existing method also gives a working condition state (such as left bias burning) and operation advice generated by utilizing the large model, but is still a unidirectional linear mode of 'perception-response', and the generated advice is always static retrieval based on a single rule and lacks comprehensive trade-off of energy efficiency, environmental protection and low carbon, so that pollution reduction and carbon reduction of a power plant cannot be realized. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a power energy and carbon pollution cooperative optimization method and system based on supply and demand matching and a large model, which are used for solving the problems that the pollution reduction and carbon reduction targets of a power plant cannot be realized due to the fact that the energy and carbon pollution cooperative optimization cannot be realized in the existing method. In order to achieve the aim of the invention, the invention adopts the following technical scheme: The power energy and carbon pollution collaborative optimization method based on supply and demand matching and a large model comprises the following steps of: Collecting historical operation data of a power plant, extracting historical operation fragments meeting evaluation of carbon pollution benefits, and constructing a power supply and demand matching heterogeneous knowledge base containing working condition requirements, technical supply and carbon pollution benefit triples; Collecting real-time operation data of a power plant, carrying out semantic analysis on the current production state by using a large language model, and generating process demand risk description; Converting the process demand risk description into a query vector, carrying out semantic retrieval on the power supply and demand matching heterogeneous knowledge base, and matching and adapting to a candidate regulation and control technical scheme set of the process demand risk description; constructing intelligent agents comprising energy efficienc