A Multi-Agent Collaborative Optimization Scheduling Method for Medium and Low Voltage Distribution Networks in Extreme Scenarios

Abstract

This invention discloses a multi-agent collaborative optimization scheduling method for medium- and low-voltage distribution networks under extreme scenarios, relating to the field of energy dispatching technology. The method includes: S1, unified modeling based on probability distribution, which constructs random state vectors and uses probabilistic statistical methods to uniformly characterize the random characteristics of different types of resources; S2, extreme scenario generation based on conditional variational autoencoders, which generates a set of scenario samples satisfying conditional distributions by learning the mapping relationship between historical meteorological data and operating states; S3, collaborative optimization based on Lagrange price signals, which achieves multi-agent collaborative decision-making driven by constructing a revenue function coupled with the scenario; S4, multi-agent coupled modeling based on graph neural networks, which constructs a distribution network topology graph and uses graph convolution operations to aggregate features of node states and extract the interaction relationship between electrical distance and power; and S5, a distributed optimization solution method based on the alternating direction multiplier method, which decomposes the global optimization problem into sub-problems of each agent.

Description

Technical Field

[0001] This invention relates to the field of energy dispatching technology, specifically to a multi-entity collaborative optimization dispatching method for medium and low voltage distribution networks under extreme scenarios. Background Technology

[0002] In recent years, with the continuous advancement of the "dual carbon" target, the penetration rate of distributed renewable energy in medium and low voltage distribution networks has been increasing, and various types of flexible resources such as energy stations and shared energy storage have gradually participated in grid operation and dispatch. Against this backdrop, the distribution network is transforming from a traditional unidirectional power supply structure to a multi-source access, bidirectional power flow, and multi-entity collaborative operation mode, significantly increasing the complexity of system operation.

[0003] With a high proportion of renewable energy connected to the grid, wind and solar power output is significantly affected by weather conditions, exhibiting strong randomness and volatility. Coupled with load-side uncertainties, this makes the distribution network prone to power imbalances, voltage exceeding limits, and localized overloads under extreme weather or abnormal operating scenarios, severely impacting the safe and stable operation of the system. Furthermore, the participation of multiple entities such as shared energy storage, energy stations, and adjustable loads in the dispatching process presents challenges due to inconsistent operational objectives and significantly different response characteristics, making effective coordination difficult through traditional centralized dispatching methods.

[0004] In existing technologies, multi-agent scheduling methods mostly rely on deterministic models or simple scenario analysis, making it difficult to accurately reflect the uncertainty characteristics under extreme scenarios. Furthermore, most methods do not fully consider the distribution network topology and the electrical coupling relationships between nodes, leading to deviations between optimization results and actual operation. In addition, as the number of participating agents increases, the size of the optimization model expands rapidly, and centralized solution methods suffer from high computational complexity and slow response speed, making it difficult to meet the real-time and online scheduling requirements of actual systems. Summary of the Invention

[0005] The purpose of this invention is to address the problems of insufficient uncertainty modeling, difficulties in multi-agent coordination, and low optimization efficiency in existing medium- and low-voltage distribution networks under conditions of high renewable energy integration. This invention provides a multi-agent collaborative optimization scheduling method for medium- and low-voltage distribution networks in extreme scenarios. By designing a unified state modeling method based on probability distribution and combining it with a conditional variational autoencoder and a multi-agent collaborative optimization model, the scheduling efficiency and computation speed of medium- and low-voltage distribution networks in extreme scenarios are significantly improved, providing reliable technical support for the safe and stable operation of distribution networks with high renewable energy penetration.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a multi-entity collaborative optimization scheduling method for medium and low voltage distribution networks under extreme scenarios, comprising the following steps: S1. Unified modeling based on probability distribution: By constructing random state vectors and using probabilistic statistical methods, the random characteristics of different types of resources are uniformly characterized, providing consistent state input for subsequent multi-agent collaborative scheduling. S2. Extreme scene generation based on conditional variational autoencoder: By learning the mapping relationship between historical meteorological data and operating status, a set of scene samples that meet the conditional distribution is generated. S3. Collaborative optimization based on Lagrange price signals: Multi-agent collaborative decision-making is achieved by constructing a revenue function coupled with the scenario. S4. Multi-agent coupling modeling based on graph neural network: By constructing a distribution network topology and using graph convolution operation to aggregate features of node states, the interaction relationship between electrical distance and power is extracted. S5. A distributed optimization solution method based on the alternating direction multiplier method decomposes the global optimization problem into sub-problems of each subject; and introduces the augmented Lagrangian function for iterative solution, thereby realizing the rapid solution of multi-subject collaborative optimization scheduling and online rolling optimization control under the premise of satisfying the risk constraints of extreme scenarios.

[0007] As a preferred embodiment of the present invention, the specific implementation steps of S1 are as follows:

[0008] S11 Stochastic State Vector Construction: Define a state vector for each subject, including power output, energy storage state, and operating characteristics.

[0009] S12 Single-Subject Probability Distribution Modeling: Establish a probability distribution for each variable in the state vector to reflect force fluctuations and uncertainty characteristics.

[0010] S13 Multi-agent joint distribution construction: The states of each agent are jointly represented, and independence is initially assumed to provide a basis for subsequent coupling constraints.

[0011] S14 State Update Mechanism: Defines a method for updating the state of a time series as the scheduling cycle changes, including dynamic changes that take into account random disturbances.

[0012] S15 Standardization Processing: Normalizes different subject states to a unified dimension to ensure consistent input for subsequent optimization.

[0013] S16 Database Storage and Interface: Stores state vectors in a unified data interface to provide data support for extreme scenario generation and scheduling optimization calls.

[0014] As a preferred embodiment of the present invention, the specific implementation steps of S2 are as follows: S21 Condition Variable Definition: Determine the condition vector required to generate extreme scenarios, including meteorological characteristics and load characteristics.

[0015] S22 encoder modeling: The conditional variational autoencoder is used to encode the historical states and conditional vectors to extract latent variable representations.

[0016] S23 Decoder generates extreme scenarios: Input latent variables and condition vectors into the decoder to generate state samples of each subject under different extreme conditions.

[0017] S24 Scene Set Construction: Combine the state samples generated by each subject to form a system-level extreme scene set.

[0018] S25 Risk Measurement Function Construction: Conduct risk assessment for each scenario, such as loss of load or cost loss, and quantify the risk under extreme conditions.

[0019] S26 Scene Sample Standardization and Database Storage: The generated scene set is standardized and stored in the database for subsequent scheduling.

[0020] As a preferred embodiment of the present invention, the specific implementation steps of S3 are as follows: S31 Revenue Function Construction: Define a revenue function for each subject in each scenario, combining electricity price and subject power output.

[0021] S32 Cost Function Definition: Defines the operating cost function for each entity, including fixed and variable costs.

[0022] S33 System Power Balance Constraint: Ensures that the total system power output equals the total load in each scenario.

[0023] S34 Node Capacity Constraint: Constrains the power of each host within its minimum and maximum capacity.

[0024] S35 Multi-agent Coordination Constraints: Utilizing Lagrange price signals to coordinate the behavior of multiple agents and achieve system-level optimization.

[0025] S36 Scenario Coupling Processing: Weighted averaging of gains from multiple scenarios improves the robustness of the strategy under extreme conditions.

[0026] As a preferred embodiment of the present invention, the specific implementation steps of S4 are as follows: S41 Distribution network topology construction: The distribution network nodes and lines are constructed into a graph structure, with each node corresponding to the main status.

[0027] S42 Adjacency Matrix Definition: Defines the connection relationship between nodes, which can be combined with line impedance weighting.

[0028] S43 Graph Convolution Feature Aggregation: Aggregates node state features through graph convolution to extract neighbor information.

[0029] S44 Node Coupling Calculation: Quantifies the electrical coupling relationship between nodes as a constraint input.

[0030] S45 Scene Coupling Constraints: Introduce extreme scene information into the coupling relationship and dynamically adjust the coupling strength between nodes.

[0031] S46 Structured Constraint Generation: Forms multi-agent scheduling optimization constraints to ensure that power interaction between nodes conforms to distribution network characteristics.

[0032] As a preferred embodiment of the present invention, the specific implementation steps of S1 are as follows: S51 Global Optimization Model Construction: Integrating the revenue function and coupling constraints to form a multi-agent global optimization problem.

[0033] S52 Construction of Augmented Lagrange Function: Introducing Lagrange multipliers and augmented terms decomposes global constraints, facilitating distributed solution.

[0034] S53 Local Subproblem Solving: Decompose the global problem into local subproblems of each subject and solve them iteratively to update the scheduling strategy.

[0035] S54 Dual Variable Update: Iteratively update the Lagrange multipliers to ensure that the global power balance is gradually satisfied.

[0036] S55 Scenario Weighted Average Processing: The scheduling outputs under multiple scenarios are weighted and averaged to form the final strategy.

[0037] S56 Convergence Judgment and Iteration Stop: Determines the convergence condition for iteration. If the condition is met, the final multi-agent scheduling strategy is output for online rolling optimization control.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) This invention first constructs a unified state modeling method based on probability distribution to realize the standardized description of multiple types of flexible resources such as distributed new energy, energy stations and shared energy storage, thereby improving the unified modeling capability of multi-source heterogeneous resources.

[0040] (2) Based on the conditional variational autoencoder, the present invention generates extreme scenario samples, effectively characterizing the random characteristics of new energy output and load demand under extreme conditions, and improving the robustness of the scheduling strategy; at the same time, it constructs a multi-subject collaborative optimization model and introduces Lagrange price signals to realize coordinated decision-making among multiple subjects and improve the overall economic efficiency of the system operation.

[0041] (3) The present invention further combines the distribution network topology and uses graph neural networks to characterize the electrical coupling relationship between nodes, thereby enhancing the model's ability to express the actual operation characteristics of the distribution network. Finally, through a distributed optimization solution method based on the alternating direction multiplier method, the large-scale optimization problem can be rapidly decomposed and solved in parallel, significantly improving computational efficiency and meeting the requirements of online scheduling.

[0042] Through the above technical solutions, the present invention can effectively improve the operating efficiency and dispatch level of medium and low voltage distribution networks under extreme scenarios, enhance the economy and robustness of the system, and strengthen its adaptability to complex operating environments, providing reliable technical support for the safe and stable operation of distribution networks with a high proportion of new energy sources. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating a multi-entity collaborative optimization scheduling method for medium- and low-voltage distribution networks under extreme scenarios in this embodiment.

[0044] Figure 2 This is a schematic diagram of the extreme scene generation method based on conditional variational autoencoder in this embodiment;

[0045] Figure 3 This is a schematic diagram of the multi-agent coupling modeling method based on graph neural networks in this embodiment. Detailed Implementation

[0046] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] like Figure 1-3 As shown, this embodiment discloses a multi-entity collaborative optimization scheduling method for medium and low voltage distribution networks under extreme scenarios, including the following steps: S1 addresses the issues of strong output fluctuations and significant differences in statistical characteristics of flexible resources such as distributed renewable energy, energy stations, and shared energy storage in medium- and low-voltage distribution networks. It proposes a unified modeling method based on probability distribution. By constructing random state vectors and utilizing probabilistic statistical methods, it uniformly characterizes the random characteristics of different resource types, achieving standardized modeling of multi-source heterogeneous resources within a unified optimization framework. This provides consistent state input for subsequent multi-entity collaborative scheduling. The specific implementation steps are as follows: S11 defines the first The state vectors of each subject: , in, This indicates that the subject has made contributions; Indicates the energy storage status; This represents the operational feature vector, including temperature, wind speed, and solar radiation intensity; S12 establishes a probability distribution for each state variable: , in, This represents the mean and standard deviation of active power output; The mean and standard deviation represent the energy storage status. This represents the mean and covariance matrix of the eigenvectors. S13 for the entire system Individual entities construct a joint distribution : ; S14 Defines the time series state update formula: , in, Represents the dynamic update function; Indicates the scheduling time step; This represents random noise that follows a zero-mean Gaussian perturbation, reflecting prediction errors and environmental randomness; S15 normalizes the state variables of each entity to the same dimension, forming the optimization input: , A unified scale is beneficial for subsequent multi-agent optimization solutions and reinforcement learning algorithms; S16 Normalizes the state of all subjects Store in a unified database Used for S2–S5 calls: , The data interface supports time steps. The query provides input for generating and optimizing solutions for extreme scenarios.

[0048] S2 addresses the issue of highly randomized renewable energy output and load demand under extreme weather conditions by proposing an extreme scenario generation method based on a conditional variational autoencoder. This method learns the mapping relationship between historical meteorological data and operational status to generate a set of scenario samples that satisfy the conditional distribution, thereby improving the robustness and security of scheduling results under extreme conditions. S21 Defines the conditional input variable vector : , in, This represents a meteorological feature vector, including wind speed, temperature, and light intensity; This represents the load characteristic vector, including historical load and short-term forecast load; S22 constructs a conditional variational autoencoder, which converts the state vector and condition vector Mapping to latent variables :

[0049] in, Indicates encoder parameters; This represents the mean of the latent variable; Represents the covariance of latent variables; S23 will include latent variables With condition vector Input to decoder to generate scene samples :

[0050] in, Indicates the generated scene number; Indicates decoder parameters; Indicates the first The subject in the first State vectors under extreme scenarios; S24 combines the extreme scenario sample sets generated by each subject to form a system scenario set. : ; S25 introduces entropy risk value to measure risk in each scenario:

[0051] in, The risk weight coefficient represents the extreme scenario and is used to characterize the degree of impact of different scenarios on the system optimization results; Representing a scene The system loss function; Used for subsequent optimization constraints to ensure the robustness of scheduling in extreme scenarios; S26 normalizes the generated extreme scenario vectors:

[0052] Normalized scene set Store in database This is used for S3–S5 scheduling optimization.

[0053] S3 addresses the issue of inconsistent objectives among multiple stakeholders, including shared energy storage operators, distribution networks, and load sides, during dispatching. It proposes a collaborative optimization method based on Lagrange price signals. By constructing a scenario-coupled revenue function, it achieves multi-stakeholder collaborative decision-making, thereby improving overall economic efficiency while ensuring system power balance. The specific implementation steps are as follows: S31 to the The subject in the first An extreme scenario Define the payoff function:

[0054] in, Indicates the first Individual subject in the scene The scheduling power is below; Indicates the nodal price signal; This represents the main operating cost function; S32 defines a quadratic function for the main operating cost:

[0055] in Indicates the main operating parameters; S33 for each scenario Define global power balance constraints:

[0056] in, This indicates the total system load in this scenario; S34 defines the power output capacity limit for each module:

[0057] in, These represent the minimum and maximum adjustable power of the main body, respectively. S35 via Lagrange multipliers Coordinating the behavior of the main body:

[0058] right Taking the partial derivative yields the optimal solution:

[0059] Achieve collaborative optimization among multiple stakeholders in a given scenario; S36 will generate the scene set from S2. Introducing a payoff function for calculation:

[0060] Optimize the average revenue across multiple scenarios to improve scheduling robustness under extreme conditions.

[0061] S4 addresses the complex power flow coupling and topological associations among nodes in medium- and low-voltage distribution networks. It proposes a multi-agent coupling modeling method based on graph neural networks. By constructing a distribution network topology graph and using graph convolution operations to aggregate node state features, the method extracts the interaction between electrical distance and power, accurately characterizing the spatial coupling characteristics among multiple agents. This provides a structured constraint expression for the multi-agent optimal scheduling model. The specific implementation steps are as follows: S41 represents the distribution network nodes and lines as a diagram:

[0062] Each node The state vector corresponding to S1 ; Each edge This indicates the wiring and electrical coupling between nodes; S42 Constructs the adjacency matrix of the graph :

[0063] Based on line impedance Weighting:

[0064] S43 uses a graph convolutional network to extract features from node states:

[0065] in, , The identity matrix is ​​used to introduce node self-connection in graph convolution operations to preserve the node's own state information; Degree matrix; Indicates the first Layer node characteristics; Represents the weight matrix; Indicates the activation function; S44 uses graph convolution output to calculate the power cross-coupling between nodes:

[0066] in, For the first The feature vector of each node in the last layer; This is the Euclidean distance function, used to quantify the coupling strength between nodes; S45 will be used in extreme scenarios Introducing coupling relationships:

[0067] in, This is a scenario-sensitive function used to adjust coupling weights and reflect changes in line or node constraints under extreme conditions. Indicates element-wise multiplication; S46 Ultimately, this leads to multi-agent scheduling optimization constraints:

[0068] in, Represents a node The neighbor set; used to ensure that the node output satisfies network coupling constraints during the scheduling process.

[0069] To address the issues of large scale, strong coupling, and difficulty in centralized solution of the aforementioned multi-agent optimization scheduling model, S5 proposes a distributed optimization solution method based on the alternating direction multiplier method. This method decomposes the global optimization problem into sub-problems for each agent and introduces an augmented Lagrangian function for iterative solution. This enables rapid solution and online rolling optimization control of multi-agent collaborative optimization scheduling while satisfying extreme scenario risk constraints. The specific implementation steps are as follows: S51 integrates the aforementioned S3 and S4 to construct a multi-agent global optimization problem:

[0070] Constraints:

[0071] in, Indicates the number of subjects; Indicates the number of extreme scenarios; This represents the payoff function defined in S3; This represents the coupling constraints defined in S4; S52 introduces Lagrange multipliers And augmented terms, decompose the global constraints:

[0072] in, These are augmentation coefficients used to accelerate the convergence of the alternating direction multiplier method; S53 For each subject In each scene Construct local optimization subproblems:

[0073] in, This is the global average value from the previous iteration; S54 Updated Lagrange Multipliers :

[0074] Ensure that global power balance constraints are gradually satisfied; S55 performs a weighted average of scheduling outputs from multiple scenarios to form the final strategy:

[0075] in, For scene probabilities or weights; S56 Determine the convergence condition of the iteration:

[0076] in, Preset tolerance; output after convergence As the final strategy for multi-entity collaborative scheduling.

[0077] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-entity collaborative optimization scheduling method for medium- and low-voltage distribution networks under extreme scenarios, characterized in that, Includes the following steps: S1. Unified modeling based on probability distribution: By constructing random state vectors and using probabilistic statistical methods, the random characteristics of different types of resources are uniformly characterized, providing consistent state input for subsequent multi-agent collaborative scheduling. S2. Extreme scene generation based on conditional variational autoencoder: By learning the mapping relationship between historical meteorological data and operating status, a set of scene samples that meet the conditional distribution is generated. S3. Collaborative optimization based on Lagrange price signals: Multi-agent collaborative decision-making is achieved by constructing a revenue function coupled with the scenario. S4. Multi-agent coupling modeling based on graph neural network: By constructing a distribution network topology and using graph convolution operation to aggregate features of node states, the interaction relationship between electrical distance and power is extracted. S5. A distributed optimization solution method based on the alternating direction multiplier method decomposes the global optimization problem into various main sub-problems; An augmented Lagrangian function is introduced for iterative solution, thereby enabling rapid solution and online rolling optimization control of multi-agent collaborative optimization scheduling while satisfying the risk constraints of extreme scenarios.

2. The multi-entity collaborative optimization scheduling method for medium and low voltage distribution networks under extreme scenarios as described in claim 1, characterized in that, The specific implementation steps of S1 are as follows: S11, Definition of the The state vectors of each subject: ， in, This indicates that the subject has made contributions; Indicates the energy storage status; This represents the operational feature vector, including temperature, wind speed, and solar radiation intensity; S12. Establish a probability distribution for each state variable: ， in, This represents the mean and standard deviation of active power output; The mean and standard deviation represent the energy storage status. This represents the mean and covariance matrix of the eigenvectors. S13, For the entire system Individual entities construct a joint distribution : ； S14. Define the time series state update formula: ， in, Represents the dynamic update function; Indicates the scheduling time step; This represents random noise that follows a zero-mean Gaussian perturbation, reflecting prediction errors and environmental randomness; S15. Normalize all principal state variables to the same dimension to form the optimized input: ； S16, Normalize the state of all subjects Store in a unified database Used for S2–S5 calls: ； The data interface supports time steps. The query provides input for generating and optimizing solutions for extreme scenarios.

3. The multi-entity collaborative optimization scheduling method for medium and low voltage distribution networks under extreme scenarios as described in claim 1, characterized in that, The specific implementation steps of S2 are as follows: S21. Define the conditional input variable vector : ， in, This represents a meteorological feature vector, including wind speed, temperature, and light intensity; This represents the load characteristic vector, including historical load and short-term forecast load; S22. Construct a conditional variational autoencoder and convert the state vector... and condition vector Mapping to latent variables : ， in, Indicates encoder parameters; This represents the mean of the latent variable; Represents the covariance of latent variables; S23, latent variables With condition vector Input to decoder to generate scene samples : ； in, Indicates the generated scene number; Indicates decoder parameters; Indicates the first The subject in the first State vectors under extreme scenarios; S24. Combine the extreme scenario sample sets generated by each subject to form a system scenario set. : ； S25. Introduce entropy risk value to measure risk in each scenario: ； in, The risk weight coefficient represents the extreme scenario and is used to characterize the degree of impact of different scenarios on the system optimization results; Representing a scene The system loss function; Used for subsequent optimization constraints to ensure the robustness of scheduling in extreme scenarios; S26. Normalize the generated extreme scene vectors: ； Normalized scene set Store in database This is used for S3–S5 scheduling optimization.

4. The multi-entity collaborative optimization scheduling method for medium and low voltage distribution networks under extreme scenarios as described in claim 1, characterized in that, The specific implementation steps of S3 are as follows: S31, regarding the first The subject in the first An extreme scenario Define the payoff function: ； in, Indicates the first Individual subject in the scene The scheduling power is below; Indicates the nodal price signal; This represents the main operating cost function; S32. Define a quadratic function for the main operating cost: ； in Indicates the main operating parameters; S33, For each scenario Define global power balance constraints: ； in, This indicates the total system load in this scenario; S34. Define the power output capacity limit for each main unit: ； in, These represent the minimum and maximum adjustable power of the main body, respectively. S35, via Lagrange multipliers Coordinating the behavior of the main body: ； right Taking the partial derivative yields the optimal solution: ； Achieve collaborative optimization among multiple stakeholders in a given scenario; S36. The scene set generated in S2 Introducing a payoff function for calculation: ； Optimize the average revenue across multiple scenarios to improve scheduling robustness under extreme conditions.

5. The multi-entity collaborative optimization scheduling method for medium and low voltage distribution networks under extreme scenarios according to claim 1, characterized in that, The specific implementation steps of S4 are as follows: S41. Represent the distribution network nodes and lines as follows: ； Each node The state vector corresponding to S1 ; Each edge This indicates the wiring and electrical coupling between nodes; S42, Constructing the adjacency matrix of the graph : ， Based on line impedance Weighting: ； S43. Use graph convolutional networks to extract features from node states: ； in, , The identity matrix is ​​used to introduce node self-connection in graph convolution operations to preserve the node's own state information; Degree matrix; Indicates the first Layer node characteristics; Represents the weight matrix; Indicates the activation function; S44. Calculate the power interaction coupling between nodes using graph convolution output: ； in, For the first The feature vector of each node in the last layer; This is the Euclidean distance function, used to quantify the coupling strength between nodes; S45 will be used in extreme scenarios Introducing coupling relationships: ； in, This is a scenario-sensitive function used to adjust coupling weights and reflect changes in line or node constraints under extreme conditions. Indicates element-wise multiplication; S46. Finally, the multi-agent scheduling optimization constraints are formed: ； in, Represents a node The neighbor set; used to ensure that the node output satisfies network coupling constraints during the scheduling process.

6. The multi-entity collaborative optimization scheduling method for medium and low voltage distribution networks under extreme scenarios according to claim 1, characterized in that, The specific implementation steps of S5 are as follows: S51. Integrate S3 and S4 to construct a multi-agent global optimization problem: ， Constraints: ， in, Indicates the number of subjects; Indicates the number of extreme scenarios; This represents the payoff function defined in S3; This represents the coupling constraints defined in S4; S52, Introducing Lagrange multipliers And augmented terms, decompose the global constraints: ， in, These are augmentation coefficients used to accelerate the convergence of the alternating direction multiplier method; S53, For each subject In each scene Construct local optimization subproblems: ， in, This is the global average value from the previous iteration; S54, Update Lagrange Multipliers : ； Ensure that global power balance constraints are gradually satisfied; S55. Perform a weighted average of the scheduling outputs from multiple scenarios to form the final strategy: ； in, For scene probabilities or weights; S56. Determine the convergence condition of the iteration: ； in, Preset tolerance; output after convergence As the final strategy for multi-entity collaborative scheduling.

Images