CN-122022227-A - Multi-energy supply and demand balance method and system

CN122022227ACN 122022227 ACN122022227 ACN 122022227ACN-122022227-A

Abstract

The application provides a multi-energy supply and demand balance method and system, which comprises the steps of carrying out data preprocessing on collected electric energy data, heat energy data, hydrogen energy data and environment parameters to generate a standardized data matrix, sequentially carrying out gray correlation analysis, clustering division, quantitative energy complementary relation and piecewise linear coding on the basis of the standardized data matrix to screen out key contradictory features, constructing an electric-thermal-hydrogen dynamic equation on the basis of the key contradictory features, carrying out space-time parameter configuration on the basis of the electric-thermal-hydrogen dynamic equation to generate a digital twin multi-energy coupling model, extracting model constraint parameters on the basis of the digital twin multi-energy coupling model, inputting the model constraint parameters into a deep reinforcement learning network, updating a reward function weight in combination with a real-time state, generating a dynamic scheduling strategy and coding into an equipment executable scheduling instruction. The application has the advantages of accurate optimized scheduling strategy, balanced and efficient supply and demand, strong self-adaptive capacity of the system and improved energy utilization rate.

Inventors

QI CHENGFEI
Zhong kan
LI SHAN
GUO JIAO
YAN KAI
ZHANG WEI
GAO GE
WANG HONGYU
YANG DONGSHENG
CHEN YUE
CUI KAI
JU HANJI
WANG YACHAO
GAO SHUAI
JIANG ZHENYU
WANG YANQIN
GAO JIAHAO
BAI DAICHENG
Bi Chaoran
WANG YAOYU
ZHANG XIWEI
ZHU YUHUAN
YI ZHONGLIN
GONG HONGXIN
LIU YAN
WANG HONG
LI WENWEN
JIAO DONGXIANG
ZHANG XINYUE
ZHANG ZHIJIE

Assignees

国网冀北电力有限公司计量中心

Dates

Publication Date: 20260512
Application Date: 20251205

Claims (13)

1. A method of multi-energy supply and demand balancing, the method comprising: performing data preprocessing on the collected electric energy data, heat energy data, hydrogen energy data and environmental parameters to generate a standardized data matrix; Sequentially carrying out gray correlation analysis, clustering division and quantization energy complementary relation and piecewise linear coding based on the standardized data matrix to screen out key contradiction characteristics; constructing an electric-thermal-hydrogen dynamic equation based on the key contradictory features, and performing space-time parameter configuration based on the electric-thermal-hydrogen dynamic equation to generate a digital twin multi-functional coupling model, wherein the space-time parameter configuration comprises geographic area division, time scale division and equipment operation parameter mapping; extracting model constraint parameters based on the digital twin multi-energy coupling model; And inputting the model constraint parameters into a deep reinforcement learning network, updating the weight of the reward function in combination with the real-time state, generating a dynamic scheduling strategy and encoding the dynamic scheduling strategy into equipment executable scheduling instructions to control the supply and demand balance of the multi-energy system.
2. The multi-energy supply and demand balancing method according to claim 1, further comprising: calculating deviation between the scheduling instruction and real-time data, generating correction parameters by federal learning in combination with long-term prediction data, and updating space-time configuration parameters of the digital twin multi-energy coupling model.
3. The method of claim 1, further comprising monitoring a mean square error and a thermal entropy rate of increase of a grid frequency, optimizing model parameters of the key contradictory features and the digital twin-type multi-energy coupling model by using an adaptive particle swarm, and feeding back the model parameters to the digital twin-type multi-energy coupling model in a closed loop.
4. The multi-energy supply and demand balancing method according to claim 1, wherein the data preprocessing of the collected electric energy source data, thermal energy source data, hydrogen energy source data and environmental parameters to generate a standardized data matrix comprises: carrying out data cleaning and abnormal value elimination on the collected electric energy data, heat energy data, hydrogen energy data and environmental parameters to obtain preliminary pretreatment data; carrying out normalization processing and feature extraction on the preliminary pretreatment data to generate a multidimensional feature vector; And carrying out edge calculation fusion on the multidimensional feature vectors to generate a standardized data matrix.
5. The multi-energy supply and demand balancing method according to claim 1, wherein the screening out key contradictory features based on the standardized data matrix sequentially performs gray correlation analysis and cluster division, quantized energy complementation relationship and piecewise linear coding comprises: Extracting a fluctuation index and a response time index from the standardized data matrix, and carrying out joint analysis and clustering division to obtain a joint analysis result of the fluctuation index and the response time index; Quantizing the energy complementary relation according to the joint analysis result to generate supply-demand unbalance threshold parameters; And carrying out feature coding on the supply and demand unbalance threshold parameters to generate key contradiction features.
6. The multi-energy supply and demand balance method according to claim 1, wherein the deep reinforcement learning network adopts a neural network architecture based on an Actor-Critic framework, an Actor network is used for generating a dynamic scheduling strategy, the Critic network is used for evaluating the value of the dynamic scheduling strategy, the model constraint parameters comprise boundary conditions of an electric-thermal-hydrogen dynamic equation, energy conversion efficiency thresholds and space-time limitation of a supply and demand coupling relation, and the model constraint parameters are transmitted to an input layer of the deep reinforcement learning network through a data interface and are subjected to deep fusion with an initialization weight parameter.
7. The multi-energy supply and demand balancing method of claim 6, wherein the deep reinforcement learning network comprises a constraining layer that dynamically limits the output range, response speed, and energy conversion efficiency of the electrical, thermal, and hydrogen devices using piecewise linear functions, the method further comprising: When the power grid frequency deviation exceeds a preset threshold value, the constraint layer adjusts the upper limit of the charge and discharge power of the energy storage equipment; The Actor network generates scheduling actions according to the current state space and the movable domain output by the constraint layer, wherein the scheduling actions comprise a generator output adjustment instruction, a heat pump power adjustment instruction and an electrolyzer hydrogen production rate instruction.
8. The multi-energy supply and demand balancing method of claim 7, wherein the step of inputting the model constraint parameters into a deep reinforcement learning network further comprises: In an offline stage, pre-training the deep reinforcement learning network by utilizing historical data to establish a strategy library; in an online stage, dynamically adjusting the weight of the deep reinforcement learning network through a real-time data stream so as to adapt to the time-varying characteristic of the electric-thermal-hydrogen coupling relation; and the model constraint parameters are integrated to the weight updating flow of the deep reinforcement learning network through a back propagation algorithm and a gradient descent mechanism.
9. The multi-energy supply and demand balancing method according to claim 1, wherein the step of updating the bonus function weights in combination with the real-time status includes: the rewarding function weight is updated through the establishment of multi-objective optimization indexes, and comprises a power grid frequency deviation suppression weight, a heat supply network entropy increase suppression coefficient and a hydrogen energy storage system charge-discharge energy balance factor.
10. The multi-energy supply and demand balancing method of claim 1, wherein the process of generating the dynamic scheduling policy comprises: inputting the real-time supply and demand state data and the updated rewarding function to an online reasoning module of the deep reinforcement learning network to generate a candidate action strategy set; Performing feasibility verification on the candidate action strategy set through the model constraint parameters, and eliminating actions violating physical constraints of the multi-energy flow coupling model; and selecting an optimal strategy by adopting an epsilon-greedy algorithm, and dynamically adjusting an epsilon value according to the real-time supply and demand volatility index.
11. The multi-energy supply and demand balancing method of claim 1, wherein the process of encoding the device-executable scheduling instructions comprises: after feasibility verification, the scheduling instruction is converted into a scheduling instruction executable by equipment through an instruction encoder, and the instruction encoder adopts a three-layer coding architecture to perform instruction conversion: The first layer decomposes the strategy actions into control parameters of three independent energy subsystems, namely electricity, heat and hydrogen; the second layer carries out protocol conversion on control parameters of each subsystem according to the type of the equipment; The third layer generates an instruction code stream in a binary or Modbus protocol format according to the device communication interface standard.
12. The multi-energy supply and demand balancing method of claim 11, wherein the process of encoding the device-executable scheduling instructions further comprises: converting the generator output adjustment strategy into an active power set value and a reactive compensation instruction for the electric energy subsystem, and adding a device address code and a timestamp label; Converting a heat pump power regulation strategy into a temperature set value and a flow control instruction for a heat energy subsystem, and inserting feedforward compensation parameters; And converting the hydrogen production rate strategy of the electrolytic cell into a current density set value and a pressure control instruction for the hydrogen energy subsystem, and adjusting the instruction priority according to the real-time pressure data of the hydrogen storage tank.
13. A multi-energy supply and demand balancing system, comprising: the multi-source data acquisition module is used for carrying out data preprocessing on the acquired electric energy source data, heat energy source data, hydrogen energy source data and environmental parameters so as to generate a standardized data matrix; The joint analysis module is used for sequentially carrying out gray correlation analysis, clustering division, quantized energy complementary relation and piecewise linear coding based on the standardized data matrix to screen out key contradiction characteristics; The digital twin modeling module is used for constructing an electric-thermal-hydrogen dynamic equation based on the key contradictory characteristics, and carrying out space-time parameter configuration based on the electric-thermal-hydrogen dynamic equation to generate a digital twin multi-functional coupling model, wherein the space-time parameter configuration comprises geographic area division, time scale division and equipment operation parameter mapping; the constraint parameter extraction module is used for extracting model constraint parameters based on the digital twin multi-energy coupling model; And the reinforcement learning scheduling module is used for inputting the model constraint parameters into the deep reinforcement learning network, updating the bonus function weight in combination with the real-time state, generating a dynamic scheduling strategy and encoding the dynamic scheduling strategy into equipment executable scheduling instructions so as to control the supply and demand balance of the multi-energy system.

Description

Multi-energy supply and demand balance method and system Technical Field The application relates to the technical field of energy management and optimization, in particular to a multi-energy supply and demand balancing method and system. Background With the transformation of global energy structures and the wide application of various energy forms, the cooperative scheduling of multi-energy systems such as electric-thermal-hydrogen and the like becomes a key for improving the energy utilization efficiency and the system stability. However, conventional multi-energy scheduling strategies typically rely on fixed rules or empirical models, which are difficult to accommodate for dynamically changing energy demands and supply conditions, resulting in inefficiency and increased instability of the system. In addition, the existing scheduling method often lacks enough flexibility and adaptability when processing the complexity and nonlinear characteristics of the multi-energy system, and cannot effectively realize the real-time balance of energy supply and demand. In recent years, development of deep reinforcement learning technology provides a new solution for dynamic scheduling of a multi-energy system, but how to effectively apply the solution to multi-energy supply and demand balance, and generating a dynamic collaborative scheduling strategy to optimize system performance still remains a technical problem to be solved currently. Disclosure of Invention In view of the above, the present application provides a multi-energy supply and demand balancing method and system to solve at least one of the above-mentioned problems. In order to achieve the above purpose, the present application adopts the following scheme: according to a first aspect of the present application, there is provided a multi-energy supply-demand balancing method, the method comprising: performing data preprocessing on the collected electric energy data, heat energy data, hydrogen energy data and environmental parameters to generate a standardized data matrix; Sequentially carrying out gray correlation analysis, clustering division and quantization energy complementary relation and piecewise linear coding based on the standardized data matrix to screen out key contradiction characteristics; constructing an electric-thermal-hydrogen dynamic equation based on the key contradictory features, and performing space-time parameter configuration based on the electric-thermal-hydrogen dynamic equation to generate a digital twin multi-functional coupling model, wherein the space-time parameter configuration comprises geographic area division, time scale division and equipment operation parameter mapping; extracting model constraint parameters based on the digital twin multi-energy coupling model; And inputting the model constraint parameters into a deep reinforcement learning network, updating the weight of the reward function in combination with the real-time state, generating a dynamic scheduling strategy and encoding the dynamic scheduling strategy into equipment executable scheduling instructions to control the supply and demand balance of the multi-energy system. As an embodiment of the present application, the above method further includes: calculating deviation between the scheduling instruction and real-time data, generating correction parameters by federal learning in combination with long-term prediction data, and updating the space-time configuration parameters of the digital twin multi-energy coupling model. The method further comprises the steps of monitoring the mean square error and the thermal entropy acceleration rate of the power grid frequency, optimizing the model parameters of the key contradiction characteristics and the digital twin multi-energy coupling model by adopting the self-adaptive particle swarm, and feeding back the model parameters to the digital twin multi-energy coupling model in a closed loop mode. As an embodiment of the present application, the data preprocessing for the collected electric energy source data, heat energy source data, hydrogen energy source data and environmental parameters in the above method to generate a standardized data matrix includes: carrying out data cleaning and abnormal value elimination on the collected electric energy data, heat energy data, hydrogen energy data and environmental parameters to obtain preliminary pretreatment data; carrying out normalization processing and feature extraction on the preliminary pretreatment data to generate a multidimensional feature vector; And carrying out edge calculation fusion on the multidimensional feature vectors to generate a standardized data matrix. As one embodiment of the application, the method for screening out key contradictory features based on gray correlation analysis, cluster division, quantized energy complementary relation and piecewise linear coding sequentially based on the standardized data matrix comprises the following steps: Extr