New energy power system multi-energy complementary optimal configuration method, system and equipment
Through the hierarchical federated learning architecture and carbon-constrained federated reinforcement learning, the problems of data security and carbon constraints in the new energy power system are solved, efficient multi-energy complementary optimization configuration is achieved, data privacy is protected and system operation efficiency is improved.
Patent Information
- Application Number
- CN202510567960.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-30
- Publication Date
- 2025-09-19
AI Technical Summary
The existing multi-energy complementary optimization configuration method for new energy power systems cannot effectively consider carbon constraints under the condition of data security, resulting in poor optimization configuration effect.
A hierarchical federated learning architecture is adopted to aggregate the model parameters of edge nodes through homomorphic encryption, and weighted average aggregation is performed in combination with global regularization and gradient correction terms. The spatiotemporal attention mechanism and carbon-constrained federated reinforcement learning are used to optimize policy information to meet carbon constraints.
Effectively protect data privacy, reduce the risk of privacy leakage, reduce computational complexity, improve optimization configuration efficiency, and achieve carbon emission reduction goals.
Smart Images
Figure CN120675172A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of new energy power systems, for example, to a method, system, and device for optimizing the configuration of multi-energy complementary configurations in new energy power systems. Background Art
[0002] Current approaches to optimizing multi-energy complementarity in new energy power systems primarily employ centralized architectures, enabling resource scheduling through global data sharing. Furthermore, these approaches often employ static carbon emission quota allocation mechanisms and coordinate electricity, heat, gas, and other energy sources through mathematical programming models. These technologies provide a foundational framework for energy synergy, but there remains room for improvement in areas such as privacy protection, spatial and temporal feature mining, and dynamic carbon market responsiveness.
[0003] Existing centralized optimization methods require the aggregation of energy data from various regions to a central server, which leads to the risk of leakage of sensitive data (such as user load curves and distributed energy locations); cross-regional communication delays are high, making it difficult to achieve minute-level policy updates.
[0004] The static carbon constraint model requires carbon quotas to be used as fixed constraints and fails to consider the impact of price fluctuations in the carbon trading market on the strategy, resulting in economic losses.
[0005] Therefore, in the existing multi-energy complementary optimization configuration method, carbon constraints cannot be effectively considered under the condition of data security, resulting in poor optimization configuration effect.
[0006] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of this application. Summary of the Invention
[0007] In order to provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not an extensive review, nor is it intended to identify key / critical elements or delineate the scope of protection of these embodiments, but rather serves as a prelude to the detailed description that follows.
[0008] The multi-energy complementary optimization configuration method, system, equipment and storage medium of the new energy power system disclosed in this invention solve the problem that the existing multi-energy complementary optimization configuration method cannot effectively consider carbon constraints while ensuring data security, which leads to poor optimization configuration effect.
[0009] The present disclosure provides a method for optimizing the configuration of multiple energy complementarity in a new energy power system, the method comprising:
[0010] Based on the regional aggregation layer of the hierarchical federated learning architecture, the model parameter θ of the edge node n (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region, for multiple regions to set their respective θ region Upload to the central coordination layer of the hierarchical federated learning architecture;
[0011] The central coordination layer combines the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global ;
[0012] Based on the spatiotemporal attention mechanism, according to the global model parameter θ global The associated data determines the fused high-level feature representation H;
[0013] Based on the data related to the high-level feature representation H, carbon-constrained federated reinforcement learning is performed to obtain optimized policy information;
[0014] The optimized strategy information obtained by carbon-constrained federated reinforcement learning is used as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraints and is used for actual energy system scheduling.
[0015] In some embodiments, based on the regional aggregation layer of the hierarchical federated learning architecture, the model parameter θ of the edge node n is (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , satisfying the formula:
[0016]
[0017] In some embodiments, the central coordination layer combines the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global , satisfying the formula:
[0018]
[0019] Among them, N total =∑N r .
[0020] In some embodiments, based on the spatiotemporal attention mechanism, according to the global model parameter θ global The associated data determines the fused high-level feature representation H, including:
[0021] According to the global model parameter θ global The associated data, extract the time feature vector Xtime , spatial eigenvector X space , multi-energy feature vector X energy , and based on the spatiotemporal attention mechanism, calculate the attention weight and obtain the weighted feature representation;
[0022] Based on the spatiotemporal feature fusion mechanism, the results processed by the multi-head attention mechanism are fused with the original feature X, and the fused high-level feature representation H is obtained through layer normalization LayerNorm.
[0023] In some embodiments, based on the spatiotemporal attention mechanism, the attention weight is calculated to satisfy the formula:
[0024]
[0025] Among them, Attention(Q,K,V) is the attention weight, Q=W q [X time ;X space ;X energy ],K=W k [X time ;X space ;X energy ],V=W v [X time ;X space ;X energy ],W q is the preset q matrix, W k is the preset k matrix, W v is the preset v matrix.
[0026] In some embodiments, based on data related to the high-level feature representation H, carbon-constrained federated reinforcement learning is performed to obtain optimized policy information, including:
[0027] According to the electricity purchase cost C grid , natural gas cost C gas , Carbon cost C carbon And the economic weight coefficient α, low carbon weight coefficient β calculate the reward value R, satisfying the formula Among them, E grid is the carbon emission value of the power grid, E gas is the carbon emission value of natural gas, E max is the rated maximum carbon emission value;
[0028] Based on historical carbon prices Carbon quota remaining amount S (t) Determine carbon price forecasts Satisfy the formula Where, τ is the time window length;
[0029] Based on the reward value R and the carbon price forecast value Determine the optimized strategy information.
[0030] In some embodiments, the optimized policy information obtained by carbon-constrained federated reinforcement learning is used as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions, including:
[0031] According to the optimization objective function Taking the control variable u as the optimization object, considering the operating cost C op , carbon cost weight factor γ and expected carbon emissions estimated by Monte Carlo sampling Perform optimization and solve to obtain the optimal control variable u that meets the constraints.
[0032] The present disclosure provides a multi-energy complementary optimization configuration system for a new energy power system, the system comprising:
[0033] The regional model parameter determination module is used for the regional aggregation layer based on the hierarchical federated learning architecture to determine the model parameters θ of the edge node n. (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , for multiple regions to set their respective θ region Upload to the central coordination layer of the hierarchical federated learning architecture;
[0034] The global model parameter determination module is used by the central coordination layer to combine the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global ;
[0035] The high-level feature representation determination module is used to determine the global model parameters θ based on the spatiotemporal attention mechanism. global The associated data determines the fused high-level feature representation H;
[0036] The strategy information determination module is used to perform carbon-constrained federated reinforcement learning based on data related to the high-level feature representation H to obtain optimized strategy information;
[0037] The optimization module is used to use the optimized strategy information obtained by carbon-constrained federated reinforcement learning as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions for use in actual energy system scheduling.
[0038] An embodiment of the present disclosure provides an electronic device, the device including at least one processor;
[0039] and a memory communicatively coupled to the at least one processor;
[0040] The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor so that the at least one processor can execute the above-mentioned new energy power system multi-energy complementary optimization configuration method.
[0041] An embodiment of the present disclosure provides a storage medium storing program instructions, which, when run, execute the above-mentioned multi-energy complementary optimization configuration method for a new energy power system.
[0042] The method, system, device, and storage medium for optimizing the configuration of multi-energy complementary new energy power systems provided by the embodiments of the present disclosure can achieve the following technical effects:
[0043] The disclosed embodiment uses a hierarchical federated learning architecture, in which the energy data of each region or entity is processed and trained at the local node, and only the model parameters rather than the original data are uploaded to the upper node. In this way, sensitive data does not need to be exposed to other regions or the coordination center of the entire power system, effectively protecting data privacy and reducing the risk of privacy leakage. Moreover, compared with the traditional centralized optimization algorithm, the hierarchical federated learning architecture of the present invention disperses the computing tasks and reduces the computational complexity. Nodes at all levels perform data processing and model training in parallel, reducing the overall optimization computing time and improving the efficiency of multi-energy complementary optimization configuration, thereby helping to improve the operating efficiency of the entire new energy power system. Furthermore, the collaborative optimization mechanism of federated reinforcement learning and carbon constraints uses carbon trading prices and carbon emissions as key parameters of the reward function. This encourages energy production and consumption nodes to consider carbon emission factors when optimizing their own behavior, thereby effectively achieving the carbon emission reduction goals of the entire new energy power system.
[0044] The above general description and the following description are exemplary and explanatory only and are not intended to limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] One or more embodiments are exemplarily described by corresponding drawings. These exemplary descriptions and drawings do not limit the embodiments. Elements with the same reference numerals in the drawings are shown as similar elements. The drawings do not constitute a scale limitation. In addition,
[0046] Figure 1 This is a flow chart of a multi-energy complementary optimization configuration method for a new energy power system provided by an embodiment of the present disclosure;
[0047] Figure 2 This is a structural diagram of a multi-energy complementary optimization configuration system for a new energy power system provided by an embodiment of the present disclosure;
[0048] Figure 3 It is a structural diagram of a multi-energy complementary optimization configuration device for a new energy power system provided by an embodiment of the present disclosure. DETAILED DESCRIPTION
[0049] In order to be able to understand the features and technical content of the embodiments of the present disclosure in more detail, the implementation of the embodiments of the present disclosure is described in detail below in conjunction with the accompanying drawings. The accompanying drawings are for reference only and are not used to limit the embodiments of the present disclosure. In the following technical description, for the sake of convenience of explanation, a full understanding of the disclosed embodiments is provided through multiple details. However, one or more embodiments can still be implemented without these details. In other cases, to simplify the drawings, well-known structures and systems can be simplified for display.
[0050] The terms "first," "second," and the like in the embodiments of the present disclosure are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate to facilitate the description of the embodiments of the present disclosure herein. Furthermore, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions.
[0051] Unless otherwise stated, the term "plurality" means two or more.
[0052] In the embodiment of the present disclosure, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B.
[0053] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.
[0054] The term "correspondence" may refer to an association relationship or a binding relationship. The correspondence between A and B means that there is an association relationship or a binding relationship between A and B.
[0055] The following describes the multi-energy complementary optimization configuration method, system, device and storage medium for the new energy power system provided by the embodiments of the present disclosure in conjunction with the accompanying drawings.
[0056] Figure 1 It is a flow chart of a method for optimizing the configuration of multiple energy complementarity in a new energy power system provided by an embodiment of the present disclosure.
[0057] Combine Figure 1 As shown, the multi-energy complementary optimization configuration method of the new energy power system may include:
[0058] S101, the regional aggregation layer based on the hierarchical federated learning architecture, calculates the model parameters θ of the edge node n (n)According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , for multiple regions to set their respective θ region Upload to the central coordination layer of the hierarchical federated learning architecture;
[0059] S102, the central coordination layer combines the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global ;
[0060] S103, based on the spatiotemporal attention mechanism, according to the global model parameter θ global The associated data determines the fused high-level feature representation H;
[0061] S104, based on the data related to the high-level feature representation H, carbon-constrained federated reinforcement learning is performed to obtain optimized policy information;
[0062] S105, the optimized strategy information obtained by carbon-constrained federated reinforcement learning is used as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions for actual energy system scheduling.
[0063] In some embodiments, the regional aggregation layer based on the hierarchical federated learning architecture calculates the model parameter θ of the edge node n. (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , satisfying the formula:
[0064]
[0065] In some embodiments, the central coordination layer combines the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global , satisfying the formula:
[0066]
[0067] Among them, N total =∑N r .
[0068] In some embodiments, the above-mentioned spatiotemporal attention mechanism is based on the global model parameter θ globalThe associated data determines the fused high-level feature representation H, including:
[0069] According to the global model parameter θ global The associated data, extract the time feature vector X time , spatial eigenvector X space , multi-energy feature vector X energy , and based on the spatiotemporal attention mechanism, calculate the attention weight and obtain the weighted feature representation;
[0070] Based on the spatiotemporal feature fusion mechanism, the results processed by the multi-head attention mechanism are fused with the original feature X, and the fused high-level feature representation H is obtained through layer normalization LayerNorm.
[0071] In some embodiments, the above-mentioned spatiotemporal attention mechanism is based on which the attention weight is calculated to satisfy the formula:
[0072]
[0073] Among them, Attention(Q, K, V) is the attention weight, Q=W q [X time ;X space ;X energy ],K=W k [X time ;X space ;X energy ],V=W v ]X time ;X space ;X energy ],W q is the preset q matrix, W k is the preset k matrix, W v is the preset v matrix.
[0074] In some embodiments, the carbon-constrained federated reinforcement learning is performed based on the data related to the high-level feature representation H to obtain optimized policy information, including:
[0075] According to the electricity purchase cost C grid , natural gas cost C gas , Carbon cost C carbon And the economic weight coefficient α, low carbon weight coefficient β calculate the reward value R, satisfying the formula Among them, E grid is the carbon emission value of the power grid, E gas is the carbon emission value of natural gas, E max is the rated maximum carbon emission value;
[0076] Based on historical carbon prices Carbon quota remaining amount S (t) Determine carbon price forecasts Satisfy the formula Where, τ is the time window length;
[0077] Based on the reward value R and the carbon price forecast value Determine the optimized strategy information.
[0078] In some embodiments, the optimized policy information obtained by the carbon-constrained federated reinforcement learning is used as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions, including:
[0079] According to the optimization objective function Taking the control variable u as the optimization object, considering the operating cost C op , carbon cost weight factor γ and expected carbon emissions estimated by Monte Carlo sampling Perform optimization and solve to obtain the optimal control variable u that meets the constraints.
[0080] In a specific example, the present disclosure may include three parts: hierarchical federated learning architecture design, spatiotemporal feature modeling based on attention mechanism, and collaborative optimization of federated reinforcement learning and carbon constraints. Specifically:
[0081] 1. Hierarchical Federated Learning Architecture Design
[0082] Three-tier architecture:
[0083] Central coordination layer: aggregates regional model parameters and generates global optimization strategies.
[0084] Regional aggregation layer: collects edge node model updates and performs model training within the region.
[0085] Edge node layer: Locally train the model and only upload encrypted gradient information.
[0086] Communication mechanism: Homomorphic encryption and differential privacy technologies are used to ensure data transmission security.
[0087] 2. Spatiotemporal Feature Modeling Based on Attention Mechanism
[0088] Input features:
[0089] Time characteristics (season, time period, historical load);
[0090] Spatial characteristics (regional meteorological data, status of cross-regional power transmission lines);
[0091] Multi-energy features (wind and solar power output prediction, heat / electricity storage status).
[0092] Attention Mechanism:
[0093] Design a spatiotemporal attention module to dynamically adjust the feature weights of different regions and time slices;
[0094] Combined with the Transformer architecture, it captures long-range spatiotemporal dependencies.
[0095] 3. Co-optimization of Federated Reinforcement Learning and Carbon Constraints
[0096] Status definition:
[0097] Energy status (power supply and demand in each region, energy storage SOC);
[0098] Carbon status (carbon trading price, regional carbon quota remaining).
[0099] Action space: cross-regional power dispatch strategies, multi-energy conversion equipment operation modes (such as electricity-to-gas start and stop).
[0100] Reward function:
[0101] Economic incentives: electricity purchase costs and carbon trading income;
[0102] Low-carbon incentive: the product of carbon emission intensity and carbon price;
[0103] Penalty items: wind and solar power curtailment rate, and load shortfall.
[0104] Training mechanism:
[0105] Each regional edge node independently trains a reinforcement learning agent and updates the global strategy through federated averaging;
[0106] Introduce carbon price prediction models (such as LSTM) to enhance adaptability to future carbon market fluctuations.
[0107] Specifically, 1. Layered federated learning data processing and output input data
[0108] (1) The edge node has local data. At the regional aggregation layer, the regional model aggregation formula is:
[0109]
[0110] Model parameter θ for edge node n (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region .
[0111] (2) Multiple regions will have their own θ region Upload to the central coordination layer, which updates the formula based on the global model:
[0112]
[0113] Combined with the sample size N of each region r (N total =∑N r ) for regional model parameters θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term (Privacy-preserving difference) to obtain the global model parameters θ global , this θ global As output, enter the next link.
[0114] 2. Data Processing and Output of the Spatiotemporal Attention Mechanism
[0115] The global model parameters θ obtained by hierarchical federated learning global The associated data is used to extract the time feature vector X time (such as seasonal coding), spatial feature vector X space (such as regional coordinates), multi-energy feature vector X energy (such as energy storage SOC), according to the formula
[0116] Q=W q [X time ;X space ;X energy ]
[0117] K=W k [X time ;X space ;X energy ]
[0118] V=W v [X time ;X space ;X energy ],
[0119] Calculate query Q, key K, and value V respectively.
[0120] Then according to the spatiotemporal attention weight calculation formula Calculate the attention weights and obtain the weighted feature representation.
[0121] Finally, through the spatiotemporal feature fusion formula H = LayerNorm(X + MultiHeadAttn(X)), the results processed by the multi-head attention mechanism MultiHeadAttn are fused with the original feature X, and the fused high-level feature representation H is obtained through layer normalization LayerNorm. H is used as the output to enter the carbon constraint federated reinforcement learning link.
[0122] 3. Data Processing and Output of Carbon Constrained Federated Reinforcement Learning
[0123] The high-level features obtained by the spatiotemporal attention mechanism represent the relevant data H and are used for carbon-constrained federated reinforcement learning. The formula is designed based on the reward function. Combined with the electricity purchase cost C grid , natural gas cost C gas , Carbon cost C carbon =P carbon ·(E grid +E gas )(where E grid For grid carbon emissions, E gas is natural gas carbon emissions), as well as the economic weight coefficient α and the low-carbon weight coefficient β to calculate the reward value R.
[0124] At the same time, using the carbon price prediction model Based on historical carbon prices (time window length is τ) and policy signal S (t) (Remaining carbon quota) Forecast carbon price Through the above-mentioned reward value calculation and carbon price prediction, the optimized strategy-related information is obtained and enters the multi-energy complementary optimization model link as output.
[0125] 4. Data processing and output of multi-energy complementary optimization model
[0126] (1) The strategy-related information obtained by carbon constraint federated reinforcement learning is used as the input of the multi-energy complementary optimization model. According to the optimization objective function Taking the control variable u (power dispatch, equipment status, etc.) as the optimization object, considering the operating cost C op (including electricity purchase and equipment maintenance), carbon cost weight factor γ, and expected carbon emissions estimated by Monte Carlo sampling Perform optimization solution.
[0127] (2) The power balance constraint ∑ g P g +P import -P export =L+P conversion (Local power generation P g , cross-regional power exchange P import / export , load L, power-to-gas / heat power P conversion ) and energy storage constraint S min ≤S t ≤S max 、 (Storage state of charge S t , charge and discharge efficiency η ch / dch ).
[0128] Finally, the optimal control variable u that meets the constraints is obtained, which is the strategy output of multi-energy complementary optimization and is used for actual energy system scheduling and other operations.
[0129] In this disclosure, the hierarchical federated learning architecture: its hierarchical division, node function definition, data interaction and collaborative training mechanism are the core of achieving multi-energy complementary optimization configuration, ensuring the effective collaboration of multi-level nodes under the premise of protecting privacy.
[0130] Integration of federated reinforcement learning and carbon constraints: The rational integration of carbon trading prices and carbon emissions into the reinforcement learning reward function, as well as the collaborative optimization algorithm under the federated learning architecture, are the key to achieving low-carbon multi-energy complementary optimization configuration.
[0131] The present disclosure provides a multi-energy complementary optimization configuration method for a new energy power system. Through a hierarchical federated learning architecture, energy data from each region or entity is processed and trained at a local node, and only model parameters, not raw data, are uploaded to the upper node. For example, sensitive data such as production plans and energy reserves of small energy production enterprises do not need to be exposed to other regions or the coordination center of the entire power system. This effectively protects data privacy and reduces the risk of privacy leakage. Moreover, compared with traditional centralized optimization algorithms, the hierarchical federated learning architecture of the present invention disperses computing tasks and reduces computational complexity. Nodes at all levels perform data processing and model training in parallel, reducing the overall optimization calculation time and improving the efficiency of multi-energy complementary optimization configuration, thereby helping to improve the operating efficiency of the entire new energy power system. Furthermore, the collaborative optimization mechanism of federated reinforcement learning and carbon constraints uses carbon trading prices and carbon emissions as key parameters of the reward function. This encourages energy production and consumption nodes to consider carbon emissions when optimizing their own behavior. For example, energy production enterprises are more inclined to adopt low-carbon production technologies or adjust production plans to reduce carbon emissions because this will obtain higher rewards (such as returns in the carbon trading market), thereby effectively achieving the carbon emission reduction goals of the entire new energy power system.
[0132] and Figure 1 Corresponding to the multi-energy complementary optimization configuration method of the new energy power system, the present disclosure also provides a multi-energy complementary optimization configuration system of the new energy power system, such as Figure 2 As shown, the system may specifically include:
[0133] The regional model parameter determination module 201 is used for determining the model parameter θ of the edge node n in the regional aggregation layer based on the hierarchical federated learning architecture. (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , for multiple regions to set their respective θ region Upload to the central coordination layer of the hierarchical federated learning architecture;
[0134] The global model parameter determination module 202 is used by the central coordination layer to combine the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global ;
[0135] The high-level feature representation determination module 203 is used to determine the global model parameters θ based on the spatiotemporal attention mechanism. global The associated data determines the fused high-level feature representation H;
[0136] A strategy information determination module 204 is configured to perform carbon-constrained federated reinforcement learning based on data related to the high-level feature representation H to obtain optimized strategy information;
[0137] The optimization module 205 is used to use the optimized strategy information obtained by carbon-constrained federated reinforcement learning as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions for use in actual energy system scheduling.
[0138] In some embodiments, based on the regional aggregation layer of the hierarchical federated learning architecture, the model parameter θ of the edge node n is (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , satisfying the formula:
[0139]
[0140] In some embodiments, the central coordination layer combines the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global , satisfying the formula:
[0141]
[0142] Among them, N total =∑N r .
[0143] In some embodiments, based on the spatiotemporal attention mechanism, according to the global model parameter θ global The associated data determines the fused high-level feature representation H, including:
[0144] According to the global model parameter θ globalThe associated data, extract the time feature vector X time , spatial eigenvector X space , multi-energy feature vector X energy , and based on the spatiotemporal attention mechanism, calculate the attention weight and obtain the weighted feature representation;
[0145] Based on the spatiotemporal feature fusion mechanism, the results processed by the multi-head attention mechanism are fused with the original feature X, and the fused high-level feature representation H is obtained through layer normalization LayerNorm.
[0146] In some embodiments, based on the spatiotemporal attention mechanism, the attention weight is calculated to satisfy the formula:
[0147]
[0148] Among them, Attention(Q,K,V) is the attention weight, Q=W q [X time ;X space ;X energy ],K=W k [X time ;X space ;X energy ],V=W v [X time ;X space ;X energy ],W q is the preset q matrix, W k is the preset k matrix, W v is the preset v matrix.
[0149] In some embodiments, based on data related to the high-level feature representation H, carbon-constrained federated reinforcement learning is performed to obtain optimized policy information, including:
[0150] According to the electricity purchase cost C grid , natural gas cost C gas , Carbon cost C carbon And the economic weight coefficient α, low carbon weight coefficient β calculate the reward value R, satisfying the formula Among them, E grid is the carbon emission value of the power grid, E gas is the carbon emission value of natural gas, E max is the rated maximum carbon emission value;
[0151] Based on historical carbon prices Carbon quota remaining amount S (t) Determine carbon price forecasts Satisfy the formula Where, τ is the time window length;
[0152] Based on the reward value R and the carbon price forecast value Determine the optimized strategy information.
[0153] In some embodiments, the optimized policy information obtained by carbon-constrained federated reinforcement learning is used as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions, including:
[0154] According to the optimization objective function Taking the control variable u as the optimization object, considering the operating cost C op , carbon cost weight factor γ and expected carbon emissions estimated by Monte Carlo sampling Perform optimization and solve to obtain the optimal control variable u that meets the constraints.
[0155] Combine Figure 3 As shown, the embodiment of the present disclosure also provides a new energy power system multi-energy complementary optimization configuration device 300, including a processor 304 and a memory (memory) 301. Optionally, the system may also include a communication interface (Communication Interface) 302 and a bus 303. Among them, the processor 304, the communication interface 302, and the memory 301 can communicate with each other through the bus 303. The communication interface 302 can be used for information transmission. The processor 304 can call the logic instructions in the memory 301 to execute the new energy power system multi-energy complementary optimization configuration method of the above embodiment.
[0156] In addition, the logic instructions in the memory 301 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.
[0157] Memory 301, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the methods in the embodiments of the present disclosure. Processor 304 executes the program instructions / modules stored in memory 301 to perform functional applications and data processing, thereby implementing the multi-energy complementary optimization configuration method for a new energy power system in the above-mentioned embodiments.
[0158] The memory 301 may include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data generated based on the use of the terminal device. Furthermore, the memory 301 may include high-speed random access memory and non-volatile memory.
[0159] An embodiment of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to execute a multi-energy complementary optimization configuration method for a new energy power system.
[0160] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.
Claims
1. A new energy power system multi-energy complementary optimization configuration method, characterized in that: The method comprises: Based on the regional aggregation layer of the hierarchical federated learning architecture, the model parameter θ of the edge node n (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , for multiple regions, their respective θ region Upload to the central coordination layer of the hierarchical federated learning architecture; The central coordination layer combines the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global ; Based on the spatiotemporal attention mechanism, according to the global model parameter θ global The associated data determines the fused high-level feature representation H; Based on the data related to the high-level feature representation H, carbon-constrained federated reinforcement learning is performed to obtain optimized policy information; The optimized strategy information obtained by carbon-constrained federated reinforcement learning is used as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraints and is used for actual energy system scheduling.
2. The method according to claim 1, characterized in that The regional aggregation layer based on the hierarchical federated learning architecture has the model parameter θ of the edge node n (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , satisfying the formula:
3. The method according to claim 1, characterized in that The central coordination layer combines the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global , satisfying the formula: Among them, N total =∑N r .
4. The method according to claim 1, wherein The spatiotemporal attention mechanism is based on the global model parameter θ global The associated data determines the fused high-level feature representation H, including: According to the global model parameter θ global The associated data, extract the time feature vector X time , spatial eigenvector X space , multi-energy feature vector X energy , and based on the spatiotemporal attention mechanism, calculate the attention weight and obtain the weighted feature representation; Based on the spatiotemporal feature fusion mechanism, the results processed by the multi-head attention mechanism are fused with the original feature X, and the fused high-level feature representation H is obtained through layer normalization LayerNorm.
5. The method according to claim 4, characterized in that The attention weight is calculated based on the spatiotemporal attention mechanism, satisfying the formula: Among them, Attention(Q, K, V) is the attention weight, Q=W q [X time ;X space ;X energy ],K=W k [X time ;X space ;X energy ],V=W v [X time ;X space ;X energy ],W q is the preset q matrix, W k is the preset k matrix, W v is the preset v matrix.
6. The method according to claim 1, characterized in that The carbon-constrained federated reinforcement learning is performed based on the data related to the high-level feature representation H to obtain optimized policy information, including: According to the electricity purchase cost C grid 、Natural gas cost C gas , Carbon cost C carbon And the economic weight coefficient α, low carbon weight coefficient β calculate the reward value R, satisfying the formula Among them, E grid is the carbon emission value of the power grid, E gas is the carbon emission value of natural gas, E max is the rated maximum carbon emission value; Based on historical carbon prices Carbon quota remaining amount S (t) Determine carbon price forecasts Satisfy the formula Where, τ is the time window length; Based on the reward value R and the carbon price forecast value Determine the optimized strategy information.
7. The method according to claim 1, characterized in that The optimized strategy information obtained by carbon constraint federated reinforcement learning is used as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions, including: According to the optimization objective function Taking the control variable u as the optimization object, considering the operating cost C op , carbon cost weight factor γ and expected carbon emissions estimated by Monte Carlo sampling Perform optimization and solve to obtain the optimal control variable u that meets the constraints.
8. A new energy power system multi-energy complementary optimization configuration system, characterized by: The system comprises: The regional model parameter determination module is used for the regional aggregation layer based on the hierarchical federated learning architecture to determine the model parameters θ of the edge node n. (n) According to the local data weight w n Perform homomorphic encryption aggregation to obtain the regional model parameter θ region , for multiple regions, their respective θ region Upload to the central coordination layer of the hierarchical federated learning architecture; The global model parameter determination module is used by the central coordination layer to combine the sample size N of each region r For the regional model parameter θ (r) Perform weighted average aggregation and consider the global regularization coefficient λ and gradient correction term Get the global model parameter θ global ; The high-level feature representation determination module is used to determine the global model parameters θ based on the spatiotemporal attention mechanism. global The associated data determines the fused high-level feature representation H; The strategy information determination module is used to perform carbon-constrained federated reinforcement learning based on data related to the high-level feature representation H to obtain optimized strategy information; The optimization module is used to use the optimized strategy information obtained by carbon-constrained federated reinforcement learning as the input of the multi-energy complementary optimization model to obtain the optimal control variable u that meets the constraint conditions for use in actual energy system scheduling.
9. An electronic device, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 7.