Method, system and device for constructing main-grid and distribution network collaborative planning intelligent agent based on thought chain, and medium

Abstract

The application discloses a main and distribution network collaborative planning intelligent agent construction method and system based on a thinking chain, equipment and a medium, and belongs to the field of intelligent agent construction, and comprises the following steps: acquiring power grid structure data, identifying the boundary of a main network and a distribution network, dividing the power grid into multiple power grid sub-regions, instantiating each power grid sub-region into an intelligent agent object, assigning a task type to each intelligent agent object to generate a task set, and constructing a reasoning chain structure; extracting an output boundary variable from the reasoning chain structure of the main network intelligent agent, extracting a to-be-injected variable bit from the reasoning chain structure, pairing the two, and establishing a cross-chain reference path between the intelligent agents through a cross-chain function to output an enhanced intelligent agent set; and aiming at constructing a main and distribution network overall collaborative planning strategy, combining, conflict coordinating and globally optimal solution screening the local strategies output by the chain intelligent agents.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent construction technology, specifically to a method, system, device, and medium for constructing intelligent agents based on the main distribution network collaborative planning of the thinking chain. Background Technology

[0002] The current collaborative planning of the main and distribution networks faces insufficient systemic and structural coordination capabilities, especially against the backdrop of high-proportion distributed energy integration, drastic load fluctuations in power grid areas, and increasingly complex boundary relationships between the main and distribution networks. Traditional methods often treat the main and distribution networks as two independent optimization systems, linearly coupling them only at the boundary conditions or capacity lower limits level. This lack of process modeling and behavioral link expression leads to a lack of dynamic consistency, planning objective coordination, and structural control flexibility between the two-level networks. Although multi-agent technology has been gradually introduced into power system modeling in recent years, most models still only remain at the level of "parallel planning and independent solution," lacking truly effective boundary data injection mechanisms and inter-policy conflict resolution mechanisms. Furthermore, it is difficult to perform refined modeling and state reconstruction of the reasoning process within the agents.

[0003] This not only makes system strategy combination difficult, but also leads to a lack of transparency and controllability in the entire collaborative planning process, making it particularly difficult to meet the requirements of the current new power system centered on regional governance for the coordinated unity of "structural stability, boundary flexibility, and implementation feasibility." Therefore, there is an urgent need to propose a construction scheme for a main and distribution network collaborative intelligent agent system with chain-like reasoning capabilities, variable boundary injection mechanisms, and system-level strategy fusion capabilities, in order to achieve the engineering transformation from decentralized decision-making to structural collaborative evolution. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is: how to solve the problem that existing main and distribution network collaborative planning treats the main network and distribution network as two independent optimization systems, only performs linear coupling at the boundary condition level, lacks process modeling and variable-level coupling mechanisms, resulting in a lack of dynamic consistency and structural coordination capabilities between the two-level networks.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for constructing a main-distribution network collaborative planning intelligent agent based on a thought chain, comprising, Acquire power grid structure data, identify the boundary between the main grid and the distribution network, divide the power grid into multiple power grid sub-regions starting from the boundary, instantiate each power grid sub-region into an intelligent agent object, assign a task type to each intelligent agent object, and generate a task set; For each intelligent agent object, a reasoning chain structure with the characteristics of a thought chain is constructed. The reasoning chain structure is used to realize the step-by-step decision-making ability of the intelligent agent in the execution of the planning task. Based on the boundary node pairs between the main network and the distribution network, output boundary variables are extracted from the inference chain structure of the main network agent, and variable bits to be injected are extracted from the inference chain structure of the distribution network agent. The output boundary variables are paired with the variable bits to be injected, and cross-chain reference paths between agents are established through cross-chain functions. The enhanced set of agents is then output. Based on the enhanced set of intelligent agents, with the goal of constructing a coordinated planning strategy for the main distribution network, the local strategies output by each chain of intelligent agents are combined, conflict coordinated, and global optimal solutions are selected.

[0007] As a preferred embodiment of the intelligent agent construction method for main and distribution network collaborative planning based on the thought chain described in this invention, the steps of acquiring power grid structure data, identifying the boundary between the main grid and the distribution network, dividing the power grid into multiple power grid sub-regions starting from the boundary, and instantiating each power grid sub-region into an intelligent agent object include: Acquire power grid structure data, which includes at least the power grid topology and node voltage levels; Identify the boundary nodes between the main grid and the distribution network based on changes in node voltage levels, and generate a set of boundary nodes; Starting from the nodes in the boundary node set, several power grid sub-regions are generated by expanding to both sides based on the power grid topology; Each sub-region of the power grid is instantiated as a corresponding intelligent agent object.

[0008] As a preferred embodiment of the main distribution network collaborative planning intelligent agent construction method based on the thinking chain described in this invention, wherein: the step of assigning task types to each intelligent agent object and generating a task set includes, Based on the characteristics of each power grid sub-region, a planning task type is assigned to the corresponding intelligent agent object; The planning task type is associated with the load data and equipment distribution information of the power grid sub-region to generate a task set.

[0009] As a preferred embodiment of the main distribution network collaborative planning intelligent agent construction method based on the thinking chain described in this invention, wherein: the step of constructing an inference chain structure with thinking chain characteristics for each intelligent agent object includes, For each agent object, construct a chain structure containing several sequentially executed inference stages; The chain structure is used to decompose the planning task into a chain of logically ordered behaviors, with each inference stage executing a subtask of the planning task corresponding to the current agent. Each inference phase defines an externally visible state vector.

[0010] This invention addresses the technical problems of uncontrollable and opaque internal reasoning processes within agents and the lack of interaction interfaces with external agents in existing main-distribution network collaborative planning by constructing a chain structure containing multiple sequentially executed inference stages for each agent object and defining the output of each inference stage as an externally visible state vector. It achieves a shift from end-to-end optimization of the planning task to step-by-step causal evolution, making the decision-making process interpretable and traceable, and providing a structured entry interface for subsequent variable injection between main-distribution network agents. As a preferred embodiment of the main network and distribution network collaborative planning agent construction method based on the thought chain described in this invention, the steps of extracting output boundary variables from the inference chain structure of the main network agent and extracting variable bits to be injected from the inference chain structure of the distribution network agent based on the boundary node pairs of the main network and distribution network include: pairing the output boundary variables with the variable bits to be injected, and establishing cross-chain reference paths between agents through cross-chain functions. Based on boundary node pairs, output boundary variables, which are externally visible state variables, are extracted from the inference chain structure of the main network agent. Extract the variable bits to be injected as externally visible state variables from the inference chain structure of the distribution network agent; The output boundary variables are paired with the variable bits to be injected, and a cross-chain reference path between agents is established through a cross-chain function. The cross-chain function is used to inject the output boundary variables as constraints into the inference chain structure of the distribution network agent. Based on cross-chain reference paths, variable consistency constraints are constructed between the main network agent and the distribution network agent.

[0011] This invention addresses the technical problem in existing technologies where there is a lack of boundary data injection mechanisms and policy conflict resolution mechanisms between main and distribution network agents. These mechanisms allow for linear coupling at the boundary condition level, preventing true variable-level collaboration. The invention achieves mutual perception and constraint transmission between distributed agents, ensuring dynamic consistency and collaborative controllability of the main and distribution network planning strategies across boundary variables. It extracts output boundary variables from the inference chain structure of the main network agent and injects variable bits from the inference chain structure of the distribution network agent. By pairing these two components and establishing cross-chain reference paths between agents through cross-chain functions, and constructing variable consistency constraints, this invention solves the problem of insufficient boundary data injection mechanisms and policy conflict resolution mechanisms between main and distribution network agents. As a preferred embodiment of the main-distribution network collaborative planning intelligent agent construction method based on the thinking chain described in this invention, the step of combining, conflict coordinating, and selecting the global optimal solution for the local strategies output by each chain-like intelligent agent includes: The local policies output by each chain agent are merged by region, and a policy dependency graph is constructed based on cross-chain reference paths; Based on the policy dependency graph, local strategies are screened for feasibility to determine whether they meet the requirements of boundary variable consistency and resource non-conflict. Local strategies that do not meet the feasibility screening are marked as infeasible candidates and then adjusted accordingly. The optimal solution for the strategy combination after feasibility screening is selected by using the objective screening function.

[0012] As a preferred embodiment of the main distribution network collaborative planning intelligent agent construction method based on the thinking chain described in this invention, the target selection function includes: The weighted local cost is obtained by multiplying the local cost of each agent's local policy by the corresponding regional importance coefficient and summing the results. The weighted summation of the deviations between the boundary variables output by the main network agent and the variable bits to be injected by the distribution network agent yields the boundary variable consistency penalty value. The number of conflicting behaviors in the statistical strategy combination is multiplied by the tolerance weight to obtain the conflict penalty value; The global cost is obtained by summing the weighted local cost, the boundary variable consistency penalty, and the conflict penalty. The goal is to minimize the global cost value and then select the globally optimal solution.

[0013] This invention provides a main distribution network collaborative planning intelligent agent construction system based on the thinking chain.

[0014] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a main distribution network collaborative planning intelligent agent construction system based on thinking chain, comprising: a data collection module, a structure construction module, an extraction module, and an output module; The data collection module acquires power grid structure data, identifies the boundary between the main grid and the distribution network, divides the power grid into multiple sub-regions starting from the boundary, instantiates each sub-region as an intelligent agent object, assigns a task type to each intelligent agent object, and generates a task set. The construction structure module is to build a reasoning chain structure with the characteristics of a thought chain for each intelligent agent object. The reasoning chain structure is used to realize the step-by-step decision-making ability of the intelligent agent in the execution of the planning task. The extraction module is based on the boundary node pairs between the main network and the distribution network. It extracts the output boundary variables from the inference chain structure of the main network agent and the variable bits to be injected from the inference chain structure of the distribution network agent. It pairs the output boundary variables with the variable bits to be injected and establishes cross-chain reference paths between agents through cross-chain functions, and outputs the enhanced agent set. The output module is based on the enhanced set of intelligent agents and aims to construct an overall collaborative planning strategy for the main distribution network. It combines, coordinates conflicts, and selects the global optimal solution for the local strategies output by each chain of intelligent agents.

[0015] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the main distribution network collaborative planning intelligent agent construction method based on the thinking chain.

[0016] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the main distribution network collaborative planning intelligent agent construction method based on the thinking chain.

[0017] The beneficial effects of this invention are as follows: This invention proposes a method and system for constructing intelligent agents for collaborative planning of main and distribution networks based on thought chains. By structurally dividing the tasks of the main and distribution network areas, instantiating distributed intelligent agents, and constructing a chain-like reasoning structure with causal paths and state visibility for each intelligent agent, explicit modeling and traceable evolution of the task process are realized. By analyzing the physical connection relationships and task dependency logic between the main and distribution networks, boundary coupling variables are identified and injected into the intelligent agent chains, establishing an inter-chain variable interaction mechanism, so that the intelligent agents are no longer information islands, but behavioral units with mutual perception capabilities. Furthermore, based on the policy candidate structure output by each intelligent agent chain, a multi-agent policy combination model is constructed. Under the condition of satisfying boundary consistency and physical feasibility, a multi-objective evolutionary optimization mechanism is used to realize the generation of global collaborative policies, and finally outputs the structured joint planning results of the main and distribution networks. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the overall process of constructing a main distribution network collaborative planning intelligent agent based on a thought chain, as provided in one embodiment of the present invention.

[0020] Figure 2 The flowchart shows the enhanced intelligent agent set output of the main distribution network collaborative planning intelligent agent construction method based on the thinking chain provided in one embodiment of the present invention.

[0021] Figure 3 This is a flowchart illustrating the combination of local strategies, conflict coordination, and global optimal solution selection in a main-distribution network collaborative planning intelligent agent construction method based on a thought chain, as provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0023] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for constructing a main distribution network collaborative planning intelligent agent based on a thought chain, including: S1. Obtain power grid structure data, identify the boundary between the main grid and the distribution network, divide the power grid into multiple power grid sub-regions starting from the boundary, instantiate each power grid sub-region into an intelligent agent object, assign a task type to each intelligent agent object, and generate a task set; S2. Construct a reasoning chain structure with the characteristics of a thought chain for each intelligent agent object. The reasoning chain structure is used to realize the step-by-step decision-making ability of the intelligent agent in the execution of the planning task. S3. Based on the boundary node pair between the main network and the distribution network, extract the output boundary variables from the inference chain structure of the main network agent, extract the variable bits to be injected from the inference chain structure of the distribution network agent, pair the output boundary variables with the variable bits to be injected, and establish cross-chain reference paths between agents through cross-chain functions, and output the enhanced agent set. S4. Based on the enhanced agent set, with the goal of constructing an overall collaborative planning strategy for the main and distribution networks, the local strategies output by each chain agent are combined, conflict coordinated, and global optimal solutions are selected. Example 2, refer to Figures 2-3 As an embodiment of the present invention, based on the previous embodiment, a method for constructing a main distribution network collaborative planning intelligent agent based on a thought chain is provided, including: The steps in step S1, including acquiring power grid structure data, identifying the boundaries between the main grid and the distribution network, dividing the power grid into multiple sub-regions based on these boundaries, instantiating each sub-region as an intelligent agent object, assigning task types to each intelligent agent object, and generating a task set, include the following steps: S11 acquires power grid structure data, which includes at least the power grid topology and node voltage levels.

[0024] Specifically, based on the static structural data and historical operational data of the actual power system, the data sources and formats are as follows: Power Grid Structure Data The data is obtained from the primary wiring diagram topology of the power grid exported from the municipal-level power dispatch center. The format is GeoJSON or XML structured graphical data, containing the location, number, and voltage level of each node; node voltage level. Attribute fields from the GIS system, such as 220kV, 110kV, 10kV, etc.; node load data. Exported from the load management system, with a time granularity of 15 minutes, in the form of normalized values; equipment type identifier. Extract tags such as "substation" and "feeder outgoing line" from the asset management system.

[0025] S12 identifies the boundary nodes between the main grid and the distribution network based on changes in node voltage levels, and generates a set of boundary nodes.

[0026] Specifically, by node voltage level Changes in the data allow for the identification of the physical boundary between the main network and the distribution network.

[0027] Any two connected nodes with different voltage levels are identified as boundary nodes. The set of boundary nodes is constructed as follows: ; in, Represents the set of boundary nodes. Indicates candidate boundary nodes, This represents a connection edge in the network. and These are the voltage levels of the nodes at both ends of the edge. This boundary set is determined by... The structure is obtained through traversal calculation and used for subsequent main and distribution network division.

[0028] S13 starts from the nodes in the boundary node set and expands to both sides based on the power grid topology to generate several power grid sub-regions.

[0029] Each sub-region of the power grid is instantiated as a corresponding intelligent agent object.

[0030] by Starting from each boundary node, a breadth-first search algorithm is used to search the structure graph. Extending from the center to both sides, it generates several structurally connected and functionally distinct power grid sub-regions. .

[0031] It should be further noted that this object contains the following functional units: Perceptual model unit, used to read and The input state is standardized. Inference initialization model units are used, setting up depth placeholders for the chained structure. Template matching model units are used based on task type. Choose an appropriate model structure. For example, for the energy storage capacity estimation task, a three-layer feedforward neural network model is used, with 16, 32, and 16 neurons in each layer. The input is the typical daily load curve (standardized 24-hour load vector), and the output is the recommended energy storage capacity (normalized value).

[0032] The data interface unit defines the output structure and calling standards; for example, the agent's output format must be in the form of... The data packet structure is provided for subsequent inference and policy optimization.

[0033] S14 assigns planning task types to the corresponding intelligent agents based on the characteristics of each power grid sub-region.

[0034] Specifically, based on regional characteristics, specific planning task types are assigned to each region. ,like This indicates the task of selecting a new site. This indicates a topology reconfiguration task. This indicates the task of assessing energy storage capacity.

[0035] The task type is preset based on the regional load structure and equipment configuration, and does not depend on external annotations. The regional task definition is as follows: ; in, Indicates the first The first region Class task definition, It is a region obtained by structural division. It is a task type tag. It is the load set of this region. It is a statistical vector of equipment types in this area.

[0036] S15 associates the planning task type with the load data and equipment distribution information of the power grid sub-region to generate a task set.

[0037] Node load data for each region Integrating into a load set Furthermore, it calculates statistical characteristics such as maximum load, average daily load, and typical electricity consumption periods, which serve as inputs for subsequent reasoning. Equipment type set Aggregates within the region to form a regional equipment distribution vector. It is used to determine the load type and structural characteristics of a region.

[0038] It should be noted that this invention structurally breaks down the entire main and distribution network collaborative planning task into multiple regional task units based on physical boundaries and functional objectives, and instantiates an independent intelligent agent object for each task unit. Each intelligent agent will independently perform chained reasoning, policy formulation, and collaborative modeling operations in subsequent steps. The key work in this step includes: dividing the region based on the main and distribution network structural boundaries, clustering tasks with the help of load evolution laws, and constructing intelligent agents based on task type templates. The final output is a set of regional tasks. and collection of intelligent agent objects .

[0039] In step S2, a reasoning chain structure with thought chain characteristics is constructed for each intelligent agent object. The reasoning chain structure is used to realize the agent's step-by-step decision-making ability in the execution of planning tasks, including the following steps: S21 constructs a chain structure for each agent object, which contains several sequentially executed inference stages; The chain structure is used to decompose the planning task into a logically ordered chain of actions, with each inference phase executing a subtask corresponding to the current agent's planning task.

[0040] Specifically, to meet the special requirements of the coordinated planning of the main and distribution networks, each stage in the reasoning structure must not only consider the local task logic, but also reserve space for the intervention of external boundary parameters.

[0041] For example, when site selection is the final output of the inference chain, its input must be able to receive the power capacity of the upper-level main grid boundary, and the intermediate state of the energy storage configuration must expose its time-series regulation potential for subsequent chains to share and call.

[0042] Therefore, each link in the chain of reasoning All must output externally visible state variables, which are used to characterize the equipment operating conditions, topology node adjustability, voltage levels, and other information formed in the current stage, and can be called by subsequent stages and related chains.

[0043] For example, in the deployment of energy storage tasks ( In this process, the chain structure sequentially includes identification of typical peak load sections, evaluation of peak shaving effects, site constraint screening, and capacity determination. Among these, Used to identify typical peak periods from regional load sequences and output the load sequence within the corresponding peak periods, as... Input; Used to calculate load fluctuation regulation residuals under different energy storage configuration capacities. It also outputs the peak shaving effect corresponding to each capacity scheme; Read The output peak shaving effect, combined with geographical location and power supply radius constraints, filters candidate sites and outputs a set of candidate sites that meet the constraints and their corresponding feasible capacity range. Read further The output set of candidate sites and feasible capacity range are used to determine the final energy storage configuration capacity under construction cost constraints. Specifically, in Phase, load fluctuation adjustment residual The expression is: , in, Indicates the area In the The load value at any given time. Indicates the current energy storage configuration capacity. Indicates by The length of typical peak periods identified. This indicates the intensity of load fluctuation at the corresponding moment. This is the fluctuation penalty weighting coefficient.

[0044] In this expression, Indicates the current energy storage configuration capacity Next, the The residual load remaining after peak shaving at a given time point is used to characterize the capacity scheme's effect on reducing the actual load. This represents a penalty term for the intensity of load fluctuations at that moment, used to reflect the stability requirements of the energy storage configuration scheme during periods of high fluctuation. The two parts together constitute the capacity scheme. Comprehensive peak shaving evaluation value during typical peak periods .

[0045] In conventional methods, load shaving only considers the average residual, while this formula... Including peak shaving stability as an additional indicator further aligns with the requirements for safe operation of the distribution network.

[0046] Additionally, for the site selection task ( The last step The following evaluation formula is designed: ; in, Indicates candidate website address The rating value, Represents a region node Arrive at the station Geographical distance, For node load weights, This indicates the estimated construction cost at that point. This represents the cost penalty coefficient.

[0047] S22 defines an externally visible state vector for each inference phase.

[0048] Unlike general chain structures, to meet the specific needs of coordinated planning between the main grid and distribution grid, each stage of the inference structure must not only consider local task logic but also reserve space for external boundary parameters to intervene. For example, when site selection is the final output of the inference chain, its input must be able to receive the boundary power capacity of the upper-level main grid, and the intermediate state of the energy storage configuration must expose its time-series regulation potential for subsequent chains to share and call upon.

[0049] Therefore, each inference stage defines an externally visible state vector, whose structure corresponds one-to-one with physical attributes such as device operating conditions, topology node adjustability, and voltage levels. Each inference function must define its "externally visible state vector," which is used for reference by subsequent inference stages or for inference chains of other agents.

[0050] Once generated, all inference chain structures are serialized and written into the chained execution module within the agent, enabling it to independently complete step-by-step decision-making for task planning. Explicit state variables output by each node in the chain are registered as "system-callable variables" for subsequent boundary coupling modeling, facilitating mutual constraints and feedback between different chains.

[0051] Finally, the output of this step is the enhanced set of agents. Each object contains a reasoning chain structure. Each chain has a complete state definition, function structure, visible variable interface, and execution boundary control logic.

[0052] It should be further clarified that the current step aims to improve upon each agent constructed in the previous step. Design a multi-stage reasoning structure with "thinking chain" characteristics. This enables the system to make progressive decisions during the execution of planning tasks. Unlike traditional integrated optimization functions, this chain-like structure simulates the "step-by-step thinking + causal evolution" process of a human planner, advancing the task execution state layer by layer. It is interpretable, traceable, and interruptible. Its greatest value lies not in the model reasoning itself, but in transforming the complex and non-decoupled main and distribution network planning tasks into a logically ordered chain of behaviors, thus providing "state entry points" and "constraint intervention points" for collaborative modeling between main and distribution networks.

[0053] Reference Figure 2In step S3, based on the boundary node pairs between the main network and the distribution network, output boundary variables are extracted from the inference chain structure of the main network agent, and variable bits to be injected are extracted from the inference chain structure of the distribution network agent. The output boundary variables are paired with the variable bits to be injected, and cross-chain reference paths between agents are established through cross-chain functions. The enhanced agent set is then output, including the following steps: S31 extracts output boundary variables as externally visible state variables from the inference chain structure of the main network agent based on boundary node pairs.

[0054] The specific input data is the output of the second step, which is the completed set of intelligent agents with a chain structure. Each agent contains its task definition. Reasoning chain structure It also clarified the various reasoning nodes in the chain. The type, input structure, and state propagation path. This step will be based on these structures, combined with the main distribution network area boundaries identified in the first step. Construct a variable dependency graph between agents, inject boundary variables, and establish cross-chain coupling paths.

[0055] The first step in this process is boundary variable identification. This involves identifying the set of boundary nodes in the main and distribution networks. For reference, in the topology diagram Find node pairs in the path that have resource, power, or voltage influence. ,in The node in the region where the main network intelligent agent is located. For the nodes in the area where the distribution network intelligent agent is located, and For each pair of boundary nodes, extract the main network chain. The output variables responsible for modules such as power allocation, load response, and node voltage prediction serve as their "output boundary variables". .

[0056] S32 extracts the bits of the variable to be injected as externally visible state variables from the inference chain structure of the distribution network agent.

[0057] The system will scan all distribution network smart agent chains. The inference node, used to perform tasks such as site deployment, energy storage regulation, and load access determination, identifies slots with input variables to be constrained, such as maximum access power. Minimum voltage support These variable slots constitute the variable bits to be injected, and are matched and paired with the aforementioned output boundary variables.

[0058] S33 pairs the output boundary variables with the variable bits to be injected, and establishes a cross-chain reference path between agents through a cross-chain function. The cross-chain function is used to inject the output boundary variables as constraints into the inference chain structure of the distribution network agent.

[0059] Once the boundary variables are successfully matched, the output variables are injected as constraint inputs into the target agent chain, and a cross-chain function is established to reference the path, thus forming actual data coupling between the two agents. To express this coupling relationship, we design the following consistency regular expression formula for the collaborative transmission of boundary constraints between the master and slave chains: ; in, This indicates that there are boundary-dependent agent ID pairs between the main and distribution network agents; Power allocation from the end of the mainnet inference chain. Input slots for load access computing nodes of distribution network intelligent agents; To meet the distribution network boundary voltage requirements, This refers to the predicted voltage of the main network at the corresponding nodes. The weight of the coupling term expresses the priority of resource consistency and voltage matching. This regularization term not only defines the consistency objective of variable constraints but also forms an optimizable system coordination structure, which will serve as one of the key evaluation terms in subsequent strategy combinations.

[0060] S34 constructs variable consistency constraints between the main network agent and the distribution network agent based on cross-chain reference paths.

[0061] Specifically, to further enhance the relevance of boundary variable modeling, a state dependency filtering mechanism is introduced. Not all agents have strong dependencies; excessive injection will increase model complexity and reduce interpretability.

[0062] Therefore, the system will filter cross-chain variables based on a dual criterion of "variable stability + task relevance," retaining only variables that have a significant and stable predictable impact on the outcome during specific task phases. For example, if a node on the mainnet... of The standard deviation of historical annual load fluctuations exceeds If the variable is not specified, it will be marked as a "high uncertainty" variable and used only as a reference input in the post-inference node of the distribution network agent to avoid directly interfering with the early decision-making.

[0063] After the injection operation is complete, all cross-chain boundary variables will be in the injected chain. The "source citation path" is marked in the middle, such as Indicates origin from intelligent agent No. The output of each inference node. This reference path can be used during chained execution scheduling to read data sequentially, ensuring the synchronization of chain execution and the validity of variables.

[0064] The final output is an enhanced set of agents. The chain structure of each agent It contains at least one cross-chain injection variable, and the variable has a readable path, injection timing, and inference node binding information.

[0065] The system implements a closed-loop mechanism of "coupled modeling + state transmission + variable injection" between main and distribution network agents within a chain structure. Compared with traditional centralized collaborative optimization methods, this mechanism no longer relies on a global constraint model to express the main-distribution linkage, but instead pushes the linkage expression down to the internal chain of each agent, truly realizing the chain-based agent design concept of distributed structure and information collaboration.

[0066] It should be noted that this step, based on the previous two steps, further completes the coupling modeling and boundary variable injection operations between intelligent agents in the main distribution network collaborative planning system.

[0067] Its core objective is to enable each agent with a fully constructed chain-based reasoning structure to... Internal inference chain structure In this process, key boundary variables output from other intelligent agent chains are identified and introduced to ensure that the constraint relationships between multiple chains are explicitly expressed within the model, ultimately forming a coupled chain-type intelligent agent system with "variable consistency and upstream-downstream interaction capabilities".

[0068] This step plays a crucial role in the invention system, bridging the preceding and following steps: it not only preserves the "distributed interpretability" of the chained reasoning in the previous step, but also provides a linkage channel for the next step of collaborative strategy optimization. Without this step, each agent in the system can only make optimal decisions within a local scope, completely failing to consider the real resource linkages, boundary constraints, and information transmission between the main and distribution networks, thus losing the core innovative value of the invention in "coordination, structure, and agent integration".

[0069] Reference Figure 3 In step S4, based on the enhanced set of agents, with the goal of constructing an overall collaborative planning strategy for the main distribution network, the local strategies output by each chain of agents are combined, conflict-coordinated, and the global optimal solution is selected, including the following steps: S41 merges the local policies output by each chain agent by region and constructs a policy dependency graph based on cross-chain reference paths.

[0070] The specific input for the current step is the set of agents output from step three. Each object contains a reasoning chain structure. And the inference nodes within them that have been injected with boundary variables. Each chain, after inference, will generate a local policy candidate output. For example: for energy storage configuration tasks The output strategy might include "suggesting the deployment of energy storage at nodes". "The capacity is 20MWh", where the suggested capacity value is derived from the second step of chain reasoning. The optimized nodes are calculated; and the deployable nodes are determined by... That is in Incorporating boundary variables (such as main grid power supply capacity) After that, it is determined according to the site selection rules. All Submitted to the policy integration scheduler through a unified data structure, the structure of which is defined as follows: The format is stored in a structured JSON file.

[0071] During the data processing phase, all input strategies will be merged by region, and the coupling boundaries between the primary and distribution networks will be automatically identified. For example, distribution network regions. The output energy storage site location strategy depends on the main grid area. The voltage support capability, this boundary relationship has been injected in step three, and its associated record is " of Quoted of Output voltage The system will build a policy dependency graph structure based on these reference records for subsequent consistency checks and conflict screening.

[0072] S42 uses a policy dependency graph to screen the feasibility of local policies, determine whether they meet the requirements of boundary variable consistency and resource non-conflict, and mark local policies that do not meet the feasibility screening as infeasible candidates and then provide feedback for adjustment.

[0073] The first stage of specific strategy combination is the input policy set. Conduct feasibility screening and preliminary assessment to determine whether the requirements of boundary variable consistency and resource non-conflict are met.

[0074] The specific boundary consistency judgment is based on the cross-chain reference path recorded in the policy dependency graph. It reads the predicted value of the boundary variable output by the main network agent and the required value of the variable bit to be injected by the distribution network agent, and judges whether the two match. If the output value of the main network required by the distribution network strategy exceeds the actual capacity that the main network can provide, the strategy does not meet the boundary variable consistency requirement, is marked as an infeasible candidate, and feedback is provided for adjustment.

[0075] Boundary consistency is based on the power grid topology and checks whether multiple agent strategies conflict in the use of the same physical resources (such as feeders, nodes, and substation sites). If resource usage conflicts exist, the importance indicators and load priorities of each strategy are evaluated, and strategies with high importance and low volatility are retained first, while conflicting strategies are coordinated or eliminated.

[0076] Boundary consistency is verified by reading the predicted values ​​of output and input variables from the reference path. If the distribution network strategy requires the main network power... The capacity is 15MW, while the corresponding node in the main network output strategy is... If the maximum is only 12MW, this strategy will be marked as an "infeasible candidate" and pushed into the refactoring strategy pool for variable feedback adjustment.

[0077] Resource conflict checking relies on the topology graph. For example, two distribution network intelligent agents. and It is recommended to use the feeder When used as a main supply branch, the system will evaluate conflict weights and importance indicators, and prioritize strategies that are highly important to the load and have low volatility.

[0078] S43 uses an objective screening function to select the globally optimal solution for the strategy combination after feasibility screening.

[0079] The specific strategy optimization phase employs a multi-objective evolutionary algorithm to perform collaborative selection, and the system objective function is designed as follows: ; in, Represents intelligent agents Corresponding strategy The local cost, such as the unit capacity cost (ten thousand yuan / MWh) in energy storage configuration tasks, and the length of newly built lines in topology reconfiguration tasks; this value is determined by multiple nodes in the chain structure (such as... The inference output is accumulated to obtain the result. The regional importance coefficient, representing local cost, can be set based on indicators such as node load level and number of connected devices. The actual value is as follows: This indicates that the strategy has a high contribution.

[0080] The second item is the boundary variable consistency penalty item, which expresses the variable transmission error between the main and distribution networks. If the difference exceeds the limit, the weight of this item will be increased to force the strategy to be adjusted. As a conflict indicator, it counts the number of all known conflict behaviors (such as site overlap, node access overload, etc.). The tolerance weights set for the system. These three sub-items together constitute the combined strategy evaluation index. The ultimate goal is to minimize the global cost while ensuring system boundary consistency and conflict-free operation.

[0081] In practice, this function is used as the objective function input into the evolutionary strategy searcher. The searcher employs an improved non-dominated sorting genetic algorithm (NSGA-II), generating a strategy combination encoding vector each time. This is a candidate global strategy. During strategy combination, the combination constraints of strategies for different task types are controlled by encoded constraint templates. For example, site location strategy and access strategy must appear in pairs, and energy storage capacity strategy must rely on the prediction results of existing power supply capacity. Each round of evolution generates 50 sets of candidate solutions, with an evolution limit of 100 rounds. Finally, the optimal solution in the non-dominated solution set is retained. indivual.

[0082] Final output results This is a set of structured strategies, in the form of multiple main and distribution network joint strategies, with the following format: "Region" ,Task It is recommended to Deploy energy storage, capacity MWh, power supply node Construction period "Month" or "Mainnet" ,Task It is recommended to build a new line. ,length kilometers, used for support "Energy storage access". This structure can be directly imported into the automatic power distribution design platform or power supply capacity simulation platform to perform preliminary design verification.

[0083] The current step, as further explained, is directed towards a set of agents for which the chain structure construction and boundary variable injection have been completed. With the goal of constructing a coordinated planning strategy for the main distribution network, the system completes the combination of output strategies of each chain of intelligent agents, conflict coordination, and global optimal solution selection.

[0084] Throughout the process, it is no longer merely about integrating local agent strategies, but rather about systematically, constrainedly, and preferentially reconstructing all local strategies based on the boundary variable coupling relationships, regional physical connection constraints, and task objective conflicts between the various chains. This results in a collaborative planning strategy that is consistent and executable across the system. .

[0085] Example 3 is an embodiment of the present invention, which provides a main distribution network collaborative planning intelligent agent construction system based on thinking chain, including a data collection module, a structure construction module, an extraction module, and an output module; The data collection module acquires power grid structure data, identifies the boundary between the main grid and the distribution network, divides the power grid into multiple sub-regions starting from the boundary, instantiates each sub-region as an intelligent agent object, assigns a task type to each intelligent agent object, and generates a task set. The construction structure module is to build a reasoning chain structure with the characteristics of a thought chain for each intelligent agent object. The reasoning chain structure is used to realize the step-by-step decision-making ability of the intelligent agent in the execution of the planning task. The extraction module is based on the boundary node pairs between the main network and the distribution network. It extracts the output boundary variables from the inference chain structure of the main network agent and the variable bits to be injected from the inference chain structure of the distribution network agent. It pairs the output boundary variables with the variable bits to be injected and establishes cross-chain reference paths between agents through cross-chain functions, and outputs the enhanced agent set. The output module is based on the enhanced set of intelligent agents and aims to construct an overall collaborative planning strategy for the main distribution network. It combines, coordinates conflicts, and selects the global optimal solution for the local strategies output by each chain of intelligent agents.

[0086] This embodiment also provides an electronic device applicable to the construction method of a main distribution network collaborative planning intelligent agent based on the thinking chain, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the construction method of a main distribution network collaborative planning intelligent agent based on the thinking chain proposed in the above embodiment.

[0087] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the main distribution network collaborative planning intelligent agent construction method based on the thinking chain proposed in the above embodiment.

[0088] The storage medium proposed in this embodiment and the method for constructing a main distribution network collaborative planning intelligent agent based on the thinking chain proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0089] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0090] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for constructing a main-distribution network collaborative planning intelligent agent based on the thinking chain, characterized in that: include, Acquire power grid structure data, identify the boundary between the main grid and the distribution network, divide the power grid into multiple power grid sub-regions starting from the boundary, instantiate each power grid sub-region into an intelligent agent object, assign a task type to each intelligent agent object, and generate a task set; For each intelligent agent object, a reasoning chain structure with the characteristics of a thought chain is constructed. The reasoning chain structure is used to realize the step-by-step decision-making ability of the intelligent agent in the execution of the planning task. Based on the boundary node pairs between the main network and the distribution network, output boundary variables are extracted from the inference chain structure of the main network agent, and variable bits to be injected are extracted from the inference chain structure of the distribution network agent. The output boundary variables are paired with the variable bits to be injected, and cross-chain reference paths between agents are established through cross-chain functions. The enhanced set of agents is then output. Based on the enhanced set of intelligent agents, with the goal of constructing a coordinated planning strategy for the main distribution network, the local strategies output by each chain of intelligent agents are combined, conflict coordinated, and global optimal solutions are selected.

2. The method for constructing a main-distribution network collaborative planning intelligent agent based on thinking chain as described in claim 1, characterized in that: The process involves acquiring power grid structure data, identifying the boundaries between the main grid and the distribution network, dividing the power grid into multiple sub-regions based on these boundaries, and instantiating each sub-region as an intelligent agent object. Acquire power grid structure data, which includes at least the power grid topology and node voltage levels; Identify the boundary nodes between the main grid and the distribution network based on changes in node voltage levels, and generate a set of boundary nodes; Starting from the nodes in the boundary node set, several power grid sub-regions are generated by expanding to both sides based on the power grid topology; Each sub-region of the power grid is instantiated as a corresponding intelligent agent object.

3. The method for constructing a main-distribution network collaborative planning intelligent agent based on thinking chain as described in claim 2, characterized in that: The process of assigning task types to each intelligent agent object and generating a task set includes, Based on the characteristics of each power grid sub-region, a planning task type is assigned to the corresponding intelligent agent object; The planning task type is associated with the load data and equipment distribution information of the power grid sub-region to generate a task set.

4. The method for constructing a main-distribution network collaborative planning intelligent agent based on thinking chain as described in claim 3, characterized in that: The construction of a reasoning chain structure with thought chain characteristics for each intelligent agent object includes, For each agent object, construct a chain structure containing several sequentially executed inference stages; The chain structure is used to decompose the planning task into a chain of logically ordered behaviors, with each inference stage executing a subtask of the planning task corresponding to the current agent. Each inference phase defines an externally visible state vector.

5. The method for constructing a main-distribution network collaborative planning intelligent agent based on thinking chain as described in claim 4, characterized in that: The process based on the boundary node pairs between the main network and the distribution network involves extracting output boundary variables from the inference chain structure of the main network agent, extracting variable bits to be injected from the inference chain structure of the distribution network agent, pairing the output boundary variables with the variable bits to be injected, and establishing cross-chain reference paths between agents through cross-chain functions. Based on boundary node pairs, output boundary variables, which are externally visible state variables, are extracted from the inference chain structure of the main network agent. Extract the variable bits to be injected as externally visible state variables from the inference chain structure of the distribution network agent; The output boundary variables are paired with the variable bits to be injected, and a cross-chain reference path between agents is established through a cross-chain function. The cross-chain function is used to inject the output boundary variables as constraints into the inference chain structure of the distribution network agent. Based on cross-chain reference paths, variable consistency constraints are constructed between the main network agent and the distribution network agent.

6. The method for constructing a main-distribution network collaborative planning intelligent agent based on thinking chain as described in claim 5, characterized in that: The process of combining, coordinating conflicts, and selecting the global optimal solution for the local strategies output by each chain of agents includes... The local policies output by each chain agent are merged by region, and a policy dependency graph is constructed based on cross-chain reference paths; Based on the policy dependency graph, local strategies are screened for feasibility to determine whether they meet the requirements of boundary variable consistency and resource non-conflict. Local strategies that do not meet the feasibility screening are marked as infeasible candidates and then adjusted accordingly. The optimal solution for the strategy combination after feasibility screening is selected by using the objective screening function.

7. The method for constructing a main-distribution network collaborative planning intelligent agent based on thinking chain as described in claim 6, characterized in that: The target filtering function includes, The weighted local cost is obtained by multiplying the local cost of each agent's local policy by the corresponding regional importance coefficient and summing the results. The weighted summation of the deviations between the boundary variables output by the main network agent and the variable bits to be injected by the distribution network agent yields the boundary variable consistency penalty value. The number of conflicting behaviors in the statistical strategy combination is multiplied by the tolerance weight to obtain the conflict penalty value; The global cost is obtained by summing the weighted local cost, the boundary variable consistency penalty, and the conflict penalty. The goal is to minimize the global cost value and then select the globally optimal solution.

8. A main distribution network collaborative planning intelligent agent construction system based on thinking chain, using the main distribution network collaborative planning intelligent agent construction method based on any one of claims 1 to 7, characterized in that, include: The module includes a data collection module, a structure building module, an extraction module, and an output module. The data collection module acquires power grid structure data, identifies the boundary between the main grid and the distribution network, divides the power grid into multiple sub-regions starting from the boundary, instantiates each sub-region as an intelligent agent object, assigns a task type to each intelligent agent object, and generates a task set. The construction structure module is to build a reasoning chain structure with the characteristics of a thought chain for each intelligent agent object. The reasoning chain structure is used to realize the step-by-step decision-making ability of the intelligent agent in the execution of the planning task. The extraction module is based on the boundary node pairs between the main network and the distribution network. It extracts the output boundary variables from the inference chain structure of the main network agent and the variable bits to be injected from the inference chain structure of the distribution network agent. It pairs the output boundary variables with the variable bits to be injected and establishes cross-chain reference paths between agents through cross-chain functions, and outputs the enhanced agent set. The output module is based on the enhanced set of intelligent agents and aims to construct an overall collaborative planning strategy for the main distribution network. It combines, coordinates conflicts, and selects the global optimal solution for the local strategies output by each chain of intelligent agents.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the main distribution network collaborative planning intelligent agent construction method based on any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the main distribution network collaborative planning intelligent agent construction method based on any one of claims 1 to 7.

Images