CN-121981427-A - Intelligent decision-making and control method and platform for multi-energy heterogeneous coupling energy system of coupling power spot market

CN121981427ACN 121981427 ACN121981427 ACN 121981427ACN-121981427-A

Abstract

The invention relates to an intelligent decision and control method and a platform for a multi-energy heterogeneous coupling energy system of a coupling power spot market, wherein the method comprises the following steps of step 1, multi-source data fusion and multi-time scale prediction; step 2, generating a multi-target optimization strategy, step 3, enhancing fuzzy decision by a large model, step 4, carrying out hierarchical real-time coordination control, and step 5, carrying out closed loop feedback iterative optimization. The platform comprises a data access and prediction module, a multi-target optimization engine module, a large model decision-making agent module, a real-time coordination controller module and a feedback learning module. The invention utilizes the deep reasoning capability and multi-time scale prediction data of the large model intelligent body and combines the real-time information of the electric power market and the running state of the equipment to construct a closed-loop mechanism of prediction, optimization, decision-making, control and learning, thereby effectively solving the limitation of the traditional method on the insufficient adaptability of the complex electric power market environment.

Inventors

Qin Manting
QI CHUNKAI
ZHANG MENG
WEI KAIHUA
MA KANG
CAO BO
SONG YIN
JIAO YANG
GUO YING
LI CHENG

Assignees

中国大唐集团科技创新有限公司
清华大学

Dates

Publication Date: 20260505
Application Date: 20251210

Claims (8)

1. The intelligent decision and control method for the multi-energy heterogeneous coupling energy system of the coupling power spot market is characterized by comprising the following steps: Step 1, multi-source data fusion and multi-time scale prediction, comprising: Acquiring equipment operation data from an SCADA system through an OPC UA protocol, acquiring environment monitoring data from a weather station through an MQTT protocol, and acquiring market transaction data from an electric power market transaction center through a RESTful API; Based on a long-short-term memory network fusion attention mechanism, constructing a multi-source data fusion prediction model, generating a daily available power prediction and a short-term available power prediction, and outputting a prediction interval for quantifying uncertainty; Step 2, generating a multi-objective optimization strategy, which comprises the following steps: Constructing models of different optimization targets by utilizing the data in the step 1, performing scheduling optimization of the day-ahead and the short-term, and generating a preliminary market electricity price reporting plan and an internal scheduling instruction optimization strategy; Step 3, large model enhanced fuzzy decision, comprising: Based on the optimization strategy generated in the step 2, obtaining an optimal strategy by adopting a multi-objective fuzzy comprehensive evaluation and large-model agent analysis method; step 4, hierarchical real-time coordination control, which comprises the following steps: Calculating a power deviation signal by taking the optimal strategy generated in the step 3 as a reference and combining real-time electricity price data of an electric power market, current equipment state and a scheduling instruction, inputting the data into a coordination control algorithm, and respectively calculating control instructions of an electric power system, a thermodynamic system, a chemical system and a hydrogen production system by adopting model prediction control; Step 5, closed loop feedback iterative optimization, comprising: Collecting market clearing results and equipment running state data, and generating systematic evaluation by an agent according to preset evaluation standards; Performing expected target comparison analysis on the evaluation result and the strategy, and performing dynamic diagnosis and self-adaptive adjustment on the evaluation standard based on the analysis result; And (3) performing gradient update for 1 time every 1 hour of new data by adopting an incremental learning mechanism, automatically pushing the alternative scheme to the manual scheduling table when the AI decision confidence coefficient is less than 75%, and recording expert selection actions as reinforcement learning reward signals to realize continuous optimization of the evaluation system.
2. The intelligent decision and control method for the multi-energy heterogeneous coupled energy system of the coupled power spot market according to claim 1, wherein the equipment operation data in the step1 comprises real-time operation parameters of a power system, a thermodynamic system, a chemical system and a hydrogen production system, the environment monitoring data comprises wind speed, temperature, illumination, weather forecast, satellite cloud image, cloud height, cloud speed and ground shadow, and the market transaction data comprises new energy online price, new energy offline price and auxiliary service price of the power market.
3. The intelligent decision and control method for the multi-energy heterogeneous coupled energy system of the coupled power spot market according to claim 1, wherein the input feature dimension of the multi-source data fusion prediction model in step 1 is extended to 47 dimensions, including cloud picture pixel grid data, shadow movement vectors and policy factors.
4. The intelligent decision and control method for the multi-energy heterogeneous coupled energy system of the coupled power spot market according to claim 1, wherein the step 2 specifically comprises: Constructing a day-ahead scheduling optimization model by using the equipment operation data, the market transaction data and the day-ahead available power prediction in the step 1, and calling a plurality of algorithms to generate day-ahead market quotations and day-ahead scheduling plans under different algorithms, wherein the day-ahead scheduling optimization model adopts mixed integer linear programming, and the solving time is less than 120 seconds; And (3) constructing a short-term scheduling optimization model by using the equipment operation data, the market transaction data and the short-term available power prediction in the step (1), and calling a plurality of algorithms to generate short-term market quotations and short-term scheduling plans under different algorithms, wherein the short-term scheduling optimization model adopts rolling time domain optimization, and the solving time is less than 15 seconds.
5. The intelligent decision and control method for the multi-energy heterogeneous coupled energy system of the coupled power spot market according to claim 4, wherein the step 3 specifically comprises: Aiming at strategies generated by different algorithms, a multi-target fuzzy comprehensive evaluation method is adopted to construct a candidate strategy set Evaluation index set : ; Evaluation index set The method comprises the following steps: ; Wherein, the economic index Calculated by expected profit of unit electric quantity, safety index Represented by negative values of system out-of-limit risk, low-carbon index Then the ratio of the clean energy power in the total power consumption is determined; Construction of an evaluation matrix The method comprises the following steps: ; Wherein the method comprises the steps of Representing policies At the index Is a quantized value of (2); Performing fuzzification processing on each index in the evaluation index set by adopting a trapezoidal membership function, wherein the membership is The calculation is as follows: ; Wherein the method comprises the steps of , Respectively as indexes The lower threshold and the upper threshold of (2) are dynamically determined by 5% and 95% quantiles of the statistical features of the historical data; Dynamically generating weight vectors by analyzing historical data and knowledge base text information by large model agents Calculating comprehensive evaluation value of each strategy : ; Selecting a strategy with the maximum comprehensive evaluation value as an optimal strategy : ; And feeding back the optimal strategy selection result, and outputting the power market quotation and energy scheduling instruction of the optimal strategy to the coordination control module.
6. The method for intelligently deciding and controlling a multi-energy heterogeneous coupled energy system of a coupled power spot market according to claim 1, wherein the coordination control logic in step 4 is: In the power surplus scene, if the real-time electricity price is lower than 80% of average price before the day, the residual electricity is transferred to the loads of energy storage, hydrogen production and heat supply, and if the real-time electricity price is higher than 120% of average price before the day, the residual electricity is sold on the internet; Under the power shortage scene, if the real-time electricity price is lower than 80% of the average price before the day, electricity is purchased from the power grid, and if the real-time electricity price is higher than 120% of the average price before the day, the power balance is realized by discharging energy storage or reducing hydrogen production and heat supply loads.
7. An intelligent decision and control platform for a multi-energy heterogeneous coupled energy system for coupling an electric power spot market, characterized in that it performs the method of any one of claims 1 to 6, comprising: The data access and prediction module integrates a multi-source heterogeneous data interface and deploys an LSTM-attribute prediction model; The multi-objective optimization engine module is internally provided with an objective function solver and supports MILP and rolling time domain optimization; a large model decision-making agent module adopts Qwen-14B-LoRA field fine tuning models, deploys the fine tuning models on a local GPU server and configures a MOE hybrid expert framework; the real-time coordination controller module is deployed at the edge computing node and adopts an MPC algorithm to support the self-adaptive control period from the second level to the minute level; And the feedback learning module is used for realizing incremental learning and A/B testing mechanism and supporting the online evolution of the model.
8. The intelligent decision and control platform for the multi-energy heterogeneous coupled energy system of the coupled power spot market according to claim 7, wherein each module works cooperatively through a standardized data bus, and wherein the large model decision intelligent module and the mainstream PLC, the energy storage EMS and the electrolytic cell control system realize standardized interfaces through Modbus TCP/OPC UA protocol.

Description

Intelligent decision-making and control method and platform for multi-energy heterogeneous coupling energy system of coupling power spot market Technical Field The invention belongs to the technical field of automatic control of comprehensive energy systems, and particularly relates to an intelligent decision and control method and platform of a multi-energy heterogeneous coupled energy system of a coupled power spot market. Background As the evolution of the power market has advanced deeply, multi-energy systems are facing highly complex and uncertain market environments. The electricity price signal has been converted from a fixed cost element to a dynamic decision variable, and the conventional energy management system has the following fundamental technical drawbacks: First, static rule driven policy rigidifies fail. The existing quotation strategies depend on manual experience rules or simple optimization models, and cannot adapt to complex game environments of spot markets. The measured data shows that in the Shanxi electric power spot market 2023 data, the policy gain fluctuation rate based on static rules is as high as + -35%, and the market Nash equilibrium point is difficult to capture. When the renewable energy permeability exceeds 40%, the decision error rate of the conventional method increases exponentially. Second, multi-time scale decisions are severely disjoint. The daily market declaration (24 hours scale) and real-time control (minutes scale) adopt a hierarchical independent optimization architecture, so that the time scale conversion loss is obvious. Under a typical scene, the daily gain loss of a certain 30MW photovoltaic +5MW hydrogen production power station caused by the day-ahead-real-time power deviation reaches 1.2 ten thousand yuan, and the annual loss exceeds 400 ten thousand yuan. The traditional method does not establish a trans-scale feedback mechanism, and decisions at each stage are disjointed with each other. Third, unstructured knowledge utilization goes to zero. The existing system only depends on structured SCADA data and historical statistics rules, the utilization rate of unstructured text information such as electric power spot market operation rules, weather bureau strong convection weather early warning, policy adjustment notification and the like is 0%, and the contribution degree of the information to electricity price prediction can reach 18% -23% through calculation. The large model technology represented by the generated pre-training transducer makes breakthrough progress in the aspects of multi-source information processing, time sequence prediction and semantic reasoning, but the coupling application of the large model technology and the electric power market is blank, and a novel energy management and control system integrating the large model cognitive ability, fuzzy decision robustness and industrial control instantaneity is needed to be constructed. Disclosure of Invention The invention aims to provide an intelligent decision-making and control method and platform for a multi-energy heterogeneous coupling energy system of a coupling power spot market, which are used for constructing a closed-loop mechanism of prediction-optimization-decision-control-learning by combining real-time information of the power market and equipment running state by utilizing deep reasoning capacity and multi-time scale prediction data of a large-model intelligent body, so that the limitation of the traditional method on the insufficient adaptability of complex power market environment is effectively solved. The invention provides an intelligent decision and control method of a multi-energy heterogeneous coupling energy system of a coupling power spot market, which comprises the following steps: Step 1, multi-source data fusion and multi-time scale prediction, comprising: Acquiring equipment operation data from an SCADA system through an OPC UA protocol, acquiring environment monitoring data from a weather station through an MQTT protocol, and acquiring market transaction data from an electric power market transaction center through a RESTful API; Based on a long-short-term memory network fusion attention mechanism, constructing a multi-source data fusion prediction model, generating a daily available power prediction and a short-term available power prediction, and outputting a prediction interval for quantifying uncertainty; Step 2, generating a multi-objective optimization strategy, which comprises the following steps: Constructing models of different optimization targets by utilizing the data in the step 1, performing scheduling optimization of the day-ahead and the short-term, and generating a preliminary market electricity price reporting plan and an internal scheduling instruction optimization strategy; Step 3, large model enhanced fuzzy decision, comprising: Based on the optimization strategy generated in the step 2, obtaining an optimal strategy by adopting a multi-obj