Power system simulation analysis intelligent agent system and simulation analysis method

Abstract

The invention discloses a power system simulation analysis intelligent agent system and a simulation analysis method. According to the method, the power system simulation analysis intelligent agent system comprises an interface layer used for receiving a user instruction and feeding back a task result in multiple modes; the workflow layer comprises a main agent and is used for analyzing a user instruction and generating and driving workflow for executing simulation analysis according to the task knowledge base; the skill layer comprises a plurality of sub-agents and is used for packaging and executing specific sub-tasks in the workflow; the tool / knowledge layer is used for providing a standardized calling interface for external tools and storing data and knowledge required by task execution; and the business system layer comprises one or more existing power system simulation business systems and is used for providing basic data, computing services and interaction components for the tool / knowledge layer.

Description

Technical Field

[0001] This invention relates to the field of simulation analysis technology, and more specifically, to an intelligent agent system and simulation analysis method for power system simulation analysis. Background Technology

[0002] Power system simulation analysis is a key supporting technology for the operation of large power grids, and grid dispatching, planning, and security defense all rely on simulation calculations. For a long time, power grid dispatching at all levels has formed a basic working mode combining fixed-cycle centralized calculations with temporary emergency calculations, ensuring the safe and stable operation of my country's power grid. With the construction of new power systems, power grid operation modes are becoming increasingly diverse and their operating characteristics increasingly complex, placing higher demands on power grid simulation analysis.

[0003] From the perspectives of intelligence, automation level, and human-computer interaction, power system simulation analysis technology can be divided into three development stages. The first stage is the computational tool stage, represented by PSASP and BPA. The simulation software is built around the core computational program, mainly providing data preparation and result output functions, with users organizing the workflow themselves. The advantage of this model is that the software has rich functions and can perform very detailed operations, making it suitable for expert users. The disadvantage is that it has a high barrier to entry, requiring users to understand the underlying logic of simulation calculations, and that apart from the calculation process, there is less consideration for automation and intelligence capabilities in each step.

[0004] The second stage is the scenario-based tool stage, represented by various advanced power grid simulation and analysis application systems. These systems are built around work scenarios and workflows, and functional modules are no longer divided by calculation type, but by work content and its processes. Users can complete a task within a single module without switching between several calculation modules. The advantage of this model is that it lowers the software's usage threshold and improves work efficiency; the disadvantage is that for tasks involving multiple scenarios, it is still necessary to switch between different modules, and the usage method does not break through the traditional analysis and interaction model.

[0005] The third stage is the currently booming intelligent stage. Since the introduction of next-generation artificial intelligence in 2017, in-depth research has yielded significant progress in areas closely related to power grid simulation analysis, such as power flow adjustment, operation mode generation, transient stability characteristic analysis, stability control strategy generation, and cross-sectional transmission capacity calculation. These advancements have been gradually applied to practical work. Currently, the main approach is to combine AI with simulation tools or scenario-based simulation applications, implementing intelligent technologies in areas requiring human intervention. For example, when power flow calculation fails to converge, intelligent convergence algorithms are invoked to provide adjustment solutions. While these efforts have effectively improved the efficiency and effectiveness of power grid simulation analysis, they have not fundamentally changed the working mode of power grid simulation analysis. The use of AI in simulation analysis systems can only play a supporting role, similar to an "assistant." How to achieve the automatic construction of intelligent agents for power system simulation analysis has become an urgent technical problem to be solved. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an intelligent agent system and simulation analysis method for power system simulation analysis.

[0007] According to one aspect of the present invention, a power system simulation analysis intelligent agent system is provided, comprising: The interface layer is used to receive user commands and provide task results in multiple modalities. The workflow layer includes a main intelligent agent, which is used to parse user instructions, generate and drive the execution of simulation analysis workflows based on the task knowledge base, wherein the workflow consists of multiple task execution strategies such as sequential execution, branch execution, loop execution and exception handling. The skill layer, which includes multiple sub-agents, is used to encapsulate and execute specific sub-tasks in the workflow; The tools / knowledge layer provides standardized interfaces for calling external tools and stores the data and knowledge required for task execution. The business system layer includes one or more existing power system simulation business systems, which provide basic data, computing services and interactive components for the tools / knowledge layer. The main agent parses user instructions into tasks, and decomposes the tasks into several sub-tasks, which are then assigned to the corresponding sub-agents for execution. The sub-agents complete the tasks collaboratively by calling the tool interfaces in the tool / knowledge layer and utilizing data and knowledge.

[0008] Optionally, the task knowledge base in the workflow layer stores simulation analysis workflow templates in the form of rules, and the main intelligent agent generates specific simulation analysis workflows by combining the parameters in the user instructions with the workflow templates. The strategy for loop execution is to record the return results of sub-agent objects in a list. Each sub-agent object has an independent, local variable-like result list. The main agent defines a default operation on the list that processes each item one by one. The main agent executes each item in the list according to the list returned by the sub-agent objects, forming a loop. If the sub-agent objects in subsequent steps also return a list, the main agent forms a nested loop for task execution. The anomaly handling strategy involves the agent establishing process knowledge for handling different types of anomalies. For errors in different situations, a dedicated anomaly handling subtask is established. The input and output of the erroneous tool and the tool invocation intent of the sub-agent are used as input data for the anomaly handling subtask. The data is modified, or the tool invocation object and parameters of the sub-agent are changed. After processing, the anomaly handling subtask returns to the state before the tool invocation, triggering the sub-agent to invoke the tool again. The anomaly handling includes power flow calculation failure, quasi-stability calculation failure, and AI model invocation failure.

[0009] Optionally, the workflow template uses vectorized encoding, with sine / cosine positional encoding added to the word vectors: set up, The position of the word in the sequence, The dimension of the word vector. i For dimension indexing; Introducing a multi-head attention mechanism into the workflow generation process using workflow templates: in: In the formula, , , These are matrices representing queries, keys, and values; Let be the dimension of the key vector. head i For the first i One's attention, h For the number of attention heads; W 0 To output the projection matrix; W i V For the first i The value transformation matrix of each head; W i K For the first i The key transformation matrix of the head; W i Q For the firsti The query transformation matrix for each head; Each attention sublayer is followed by a two-layer fully connected feedforward network: In the formula, x The input is the output from the attention layer; W 1 represents the first-layer weight matrix; b 1 represents the first layer bias; W 2 represents the weight matrix of the second layer; b 2 represents the second layer bias.

[0010] Optionally, the sub-agents in the skill layer include business-type sub-agents, auxiliary-type sub-agents, and knowledge-type sub-agents; Business-class sub-intelligent agents are used to execute long-term sub-tasks of power grid simulation analysis explicitly defined in the workflow; Auxiliary sub-agents are used to perform short-term sub-tasks such as data maintenance and query statistics in the workflow; Knowledge-based sub-agents are used to perform domain knowledge question-answering sub-tasks in the workflow.

[0011] Optionally, the business-class sub-agent may include a power grid control skill model: in: For control vectors; This is the gain coefficient vector; The target state vector of the power grid; The current state vector of the power grid is obtained through a knowledge-based sub-agent, the target state vector of the power grid is obtained through a business-based sub-agent, the current state of the power grid is queried through a business-based sub-agent, the gain coefficient vector is calculated or queried through a business-based sub-agent, and finally the control vector is calculated. The control execution skill applies the control vector to the power grid data to adjust the power grid data, and then the control effect is verified by the power grid simulation skill.

[0012] Optionally, the steps of the power grid control skill model for multi-step control of the power grid are as follows: The initialization step starts from the current actual operating state of the power grid, calculates the comprehensive score of the current actual operating state of the power grid to the target requirements, and puts the current actual operating state of the power grid as an adjustment path of length 1 into the initial bundle of the next round of control, while recording that no action type has been used for this path; The multi-round iterative expansion step performs the following sub-steps for each adjustment path in the current bundle controlled by the current round: (a) Dynamically generate a set of actionable actions based on the current power grid topology corresponding to the adjustment path; (b) Simulate the execution of each executable action in the action set to obtain a new state and calculate its overall score, where the overall score = target matching degree + safety margin - operation cost; (c) Group all candidate actions corresponding to the new state by action type. Within each action type group, sort the candidate actions by score and apply an exponential decay penalty to the lower-ranked ones. If an action type has never been used in the current path, give an extra reward to obtain the processed candidate actions. (d) After merging all processed candidate actions, retain the top b paths with the highest comprehensive scores to form the current bundle for the next round of input; The output result step, after the termination of multiple rounds of iterative expansion steps, returns all adjusted paths in the final current bundle as a complete solution.

[0013] Optionally, the tools / knowledge layer includes: The knowledge base is used to store work knowledge, work data, and work conclusions. Work knowledge includes workflow knowledge stored in the form of rules and domain knowledge stored in the form of triples. Work conclusions include analysis reports stored in the form of XML files. The tool service is used to encapsulate simulation calculation programs, AI models, and database operations into standardized tool interfaces based on the Model Context Protocol (MCP).

[0014] According to another aspect of the present invention, a power system simulation analysis method is provided, comprising: S1. Receive natural language task instructions from the user through the interface layer; S2. The main intelligent agent parses the task instructions, matches the corresponding workflow template from the task knowledge base, and combines the instruction parameters to form an executable workflow. The workflow consists of multiple task execution strategies, including sequential execution, branch execution, loop execution, and exception handling. S3. The main agent decomposes the workflow into multiple subtasks and schedules the corresponding sub-agents to execute them; S4. During the execution of sub-tasks, the sub-agent completes the task by calling the tool interface in the tool / knowledge layer and using the associated task data object; S5. The sub-agent returns the sub-task execution results to the main agent, which then integrates them and presents the final task results to the user through the interface layer.

[0015] Optionally, in step S4, when calling the tool interface, if the required parameters are incomplete, the system will automatically complete the default parameters through prompt words based on the context and data tags of the current task data object, or request confirmation from the user for key parameters.

[0016] Optionally, a user’s one or more consecutive interactions constitute a dialogue, and a dialogue contains one or more tasks; tasks are linked through their associated task data objects, while the content of question-and-answer subtasks within the dialogue is stored separately in the dialogue history database.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, the storage medium storing a computer program for performing the methods described in any of the above aspects of the present invention.

[0018] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the method described in any of the preceding aspects of the present invention.

[0019] Therefore, based on a large model, this invention comprehensively utilizes power system simulation analysis technology, automatic data processing technology, and related artificial intelligence technology to establish a new intelligent application system for various power grid simulation analysis businesses. This breaks through the limitations of existing simulation analysis calculation tools and scenario-based tools, and achieves a leapfrog development of power system simulation analysis in the direction of intelligence. Attached Figure Description

[0020] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures: Figure 1 This is a schematic diagram of the structure of a power system simulation and analysis intelligent agent system provided in an exemplary embodiment of the present invention; Figure 2 This is a flowchart illustrating a power system simulation analysis method provided in an exemplary embodiment of the present invention; Figure 3 This is a schematic diagram of the workflow of tasks, subtasks, and metatasks provided in an exemplary embodiment of the present invention; Figure 4 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. Detailed Implementation

[0021] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0022] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention.

[0023] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of the present invention are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0024] It should also be understood that in the embodiments of the present invention, "multiple" can refer to two or more, and "at least one" can refer to one, two or more.

[0025] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more unless explicitly defined or given contrary instructions in the context.

[0026] Furthermore, the term "and / or" in this invention is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this invention generally indicates that the preceding and following related objects have an "or" relationship.

[0027] It should also be understood that the description of the various embodiments in this invention emphasizes the differences between the various embodiments, and the similarities or similarities can be referred to each other. For the sake of brevity, they will not be described in detail.

[0028] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0029] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0030] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0031] Figure 1 This is a schematic diagram of the structure of a power system simulation and analysis intelligent agent system provided in an exemplary embodiment of the present invention. For example... Figure 1 The power system simulation analysis intelligent agent system shown includes: The interface layer is used to receive user commands and provide task results in multiple modalities. The workflow layer includes a main intelligent agent, which is used to parse user instructions, generate and drive the execution of simulation analysis workflows based on the task knowledge base, wherein the workflow consists of multiple task execution strategies such as sequential execution, branch execution, loop execution and exception handling. The skill layer, which includes multiple sub-agents, is used to encapsulate and execute specific sub-tasks in the workflow; The tools / knowledge layer provides standardized interfaces for calling external tools and stores the data and knowledge required for task execution. The business system layer includes one or more existing power system simulation business systems, which provide basic data, computing services and interactive components for the tools / knowledge layer. The main agent parses user instructions into tasks, and decomposes the tasks into several sub-tasks, which are then assigned to the corresponding sub-agents for execution. The sub-agents complete the tasks collaboratively by calling the tool interfaces in the tool / knowledge layer and utilizing data and knowledge.

[0032] Specifically, in recent years, large-scale modeling technologies, represented by ChatGPT and DeepSeek, have achieved continuous breakthroughs, elevating the interaction and understanding between humans and artificial intelligence to a new level. Based on generative methods and massive sample training, large-scale models can provide correct answers based on input text information or generate richly detailed images; DeepSeek can even provide detailed reasoning processes. Building upon large-scale modeling technology, workflow-oriented intelligent agents can be constructed, such as ByteDance's Coze and Tencent's Yuanqi. These intelligent agents can intelligently complete various complex tasks based on user input information, representing an effective way to apply artificial intelligence technology to solve practical application problems.

[0033] Based on the concept of building intelligent agents using large models, we can propose an intelligent power system analysis agent (iPSA-Agent). This technology is based on large models and comprehensively utilizes power system simulation analysis technology, automatic data processing technology, and related artificial intelligence technology to establish a new type of intelligent application system for various power grid simulation analysis businesses. It breaks through the limitations of existing simulation analysis calculation tools and scenario-based tools, and realizes a leapfrog development of power system simulation analysis in the direction of intelligence.

[0034] The technical problems to be solved by this invention include: 1. Currently, AI applications in the field of power system simulation analysis only involve replacing a few manual processes, and their overall support for simulation analysis work is limited; 2. The human-computer interaction of existing power system simulation analysis software is limited to data, the steps are complex and the content is not in-depth enough, which hinders the improvement of work efficiency; 3. Existing simulation analysis business systems are diverse in form and complex in use, making it difficult to use them together through a unified entry point.

[0035] Based on this, the construction idea of ​​this invention is as follows: Power system simulation is a highly specialized process. iPSA-Agent differs significantly from common, internet-based agent applications, with the following key characteristics: 1. Complex process. While the requirements for basic power grid simulation calculations such as power flow convergence and fault scanning are clear, and in some cases there are even standards and specifications to follow, the actual execution of these tasks is extremely complex, involving repeated adjustments to data, iterative calculations and analyses, repeated summarization of results, and handling of calculation failures. 2. Complex data. The calculations and analyses require simulation data and relevant domain knowledge as the core. Because they involve different types of calculations such as power flow, transient stability, and short circuits, the initial calculation data may generate a large amount of derived data and results, as well as a wealth of analytical conclusions, making data and knowledge management complex. 3. Complex interfaces. iPSA-Agent needs to call various simulation calculation programs, multiple types of data interfaces, AI models, and even supercomputing systems.

[0036] Based on the above characteristics, the design of iPSA-Agent mainly considers the following principles: 1. The program implementation and large-scale model application are organically combined to construct a simulation calculation workflow. Large-scale models are mainly used in areas where traditional software methods are difficult to implement, such as human-computer interaction understanding, complex task decomposition, and result summarization. Large-scale models are not used for aspects that can be easily implemented through automation and simulation calculations. 2. Organic integration of data and knowledge. In traditional simulation analysis systems, data management primarily supports functional implementation, while the knowledge behind the data is generally held by humans. iPSA-Agent executes tasks independently, requiring the establishment of a correspondence between data and knowledge. This facilitates the intelligent agent's simultaneous access to both data and required knowledge, and supports upgrading the traditional human-machine data interaction mode to a knowledge interaction mode.

[0037] 3. Integrate with existing simulation software, rather than replace it. On the one hand, existing simulation software can provide iPSA-Agent with data storage, capability implementation, and human-computer interaction assistance; on the other hand, the development and maturation of iPSA-Agent requires a long process, and traditional computing software may have more advantages for simple tasks.

[0038] like Figure 1 As shown, the present invention proposes a power system simulation and analysis intelligent agent system, comprising: 1. The interface layer consists of three parts: text, graphics, and tables. The text part is used to receive instructions, provide feedback text information, and report; the graphics part includes geographic maps and statistical charts, which are used to visually display the main information of the task execution results, and the statistical charts can be generated from large models; the tables are used to display detailed data information of the task execution results. 2. The workflow layer is used to understand user intent, form task workflows, and coordinate the various sub-agents of the skill layer based on workflow. To more accurately match the simulation analysis workflow template, the workflow template first needs to be vectorized and encoded. To preserve the template sequence order information, sine / cosine positional encoding is added to the word vectors: in: The position of the word in the sequence, The dimension of the word vector. i This is used for dimensional indexing. The encoding, added to the word embedding, is fed into subsequent layers, enabling the model to perceive relative / absolute positional information.

[0039] Then, a multi-head attention mechanism is introduced: in: In the formula, , , These are matrices representing queries, keys, and values; Let be the dimension of the key vector. head i For the first i One's attention, h For the number of attention heads; W 0 To output the projection matrix; W i V For the first i The value transformation matrix of each head; W i K For the first i The key transformation matrix of the head; W i Q For the first i The query transformation matrix for each head; Each attention sublayer is followed by a two-layer fully connected feedforward network: In the formula, x The input is the output from the attention layer; W 1 represents the first-layer weight matrix; b 1 represents the first layer bias; W 2 represents the weight matrix of the second layer; b 2 represents the second layer bias. The first linear transformation reduces the dimension from... Extended to (Usually taken as 4 times), then activated using ReLU (or GELU); the second linear transformation then maps back. .

[0040] After the above processing, the main intelligent agent can more accurately select the simulation analysis workflow template and generate a specific simulation analysis workflow.

[0041] 3. The skill layer encapsulates the execution logic of various long-term and short-term sub-tasks of iPSA-Agent through sub-intelligent agents, which are divided into business-type sub-intelligent agents, auxiliary-type sub-intelligent agents, and knowledge-type sub-intelligent agents; The business-type sub-agent includes power grid control skills, and the main control model is as follows: in: : Control vector; Gain coefficient vector; : Target state vector of the power grid; : Current state vector of the power grid.

[0042] The target state vector of the power grid is obtained through the knowledge-based sub-agent, the current state of the power grid is queried through the business-based sub-agent, the gain coefficient vector is calculated or queried through the business-based sub-agent, and finally the control vector is calculated.

[0043] After obtaining the control vector, the control execution skill applies the control vector to the power grid data to adjust the power grid data. Then, the power grid simulation skill verifies the control effect. Based on the effect, it is determined whether to proceed to the next round of control process.

[0044] For situations requiring multi-step control, follow these steps:

[0045] Here, `score()` is the state evaluation function, `get_reachable_actions()` is the reachable action generator, `violates_constraint()` is the constraint checking function, and `action_cost()` is the operation cost function. The entire process is briefly described below: (1) Initialization Starting from the current actual operating state of the power grid T0 (including generator output, bus voltage, line power flow, switch status, etc.); Calculate the overall score of this state for the target requirement q (such as total load value, voltage qualification rate, N-1 pass rate, etc.); The initial state is placed into the initial bundle B0 as an "adjustment path" of length 1, while recording that no action type has been used on this path.

[0046] (2) Multi-round iterative expansion For each candidate action in the current bundle (representing an intermediate state after several steps have been performed): (a) Dynamically generate action sets Based on the current power grid topology (e.g., which lines are closed and which areas are connected), call the get_reachable_actions function to list only the executable actions.

[0047] For example, if a substation has been disconnected, its transformer cannot be re-tuned; if a capacitor has been put into operation, the next step can only be to disconnect or keep it in operation.

[0048] (b) Simulation execution and scoring For each action, check whether it violates hard constraints (such as generator output exceeding limits, voltage exceeding limits): <1> If it is valid, then the simulation is executed to obtain a new state Tnew; <2> The overall score for the new state is calculated as follows: Target matching degree + Safety margin - Operational cost. Operational cost reflects the maintenance cost of the action (e.g., high cost for main transformer speed adjustment, low cost for capacitor switching).

[0049] (c) Increased group diversity All newly generated candidate actions are grouped by action type (e.g., "generator regulation," "capacitor switching," "line switch," etc.). Within each group, candidate actions are sorted by score, and those with lower rankings are subject to exponential decay penalties. Additionally, if an action type has never been used in the current path, an extra reward (e.g., higher score) is given. 1.1), to encourage hybrid strategies.

[0050] (d) Merging and truncation Merge all candidate actions that have undergone diversity processing; retain the top b paths with the highest total scores to form the bundle for the next round.

[0051] (3) Output results After a maximum of l steps, all paths in the bundle are returned, each path being a complete, executable multi-hop adjustment sequence.

[0052] The above steps can generate high-quality adjustment plans and provide diverse options to meet the actual needs of "multiple contingency plans and flexible selection" in power grid operation analysis.

[0053] 4. The tool / knowledge layer includes calls to external services and the storage and utilization of knowledge. The former can be based on the MCP (Model Context Protocol), while the latter stores the data and knowledge environment for task execution. 5. The business system layer provides iPSA-Agent with support in terms of data, computing, and even automatic processing functions, and can provide the necessary interactive interface.

[0054] The positioning of intelligent agents in power system simulation analysis includes the following three aspects: 1. In terms of the application of artificial intelligence technology, the current research and application mode of AI models corresponding to a few manual links will be upgraded to an application mode for complex work tasks. The large model integrated by intelligent agents will be used to understand and drive the execution of various power grid simulation analysis tasks, and coordinate the use of AI models, simulation programs, and various automated analysis methods. 2. In terms of human-computer interaction, simulation analysts interact with intelligent agents through various forms such as text and voice. The intelligent agents use the summarizing ability of large models to extract patterns from the results data and provide refined information in various forms such as text, pictures, and tables, thus promoting the current data interaction between humans and machines towards knowledge interaction. 3. In terms of constructing the simulation analysis system, as the system's "first entry point," it calls upon underlying services such as data, computation, and computing power, as well as modules and pages from various business systems to meet the needs of task execution and human-computer interaction. At this point, conventional business systems are mainly used for data creation, condition setting, and feedback information display, and are no longer the core tools for workflow execution.

[0055] It is evident that the primary goal of iPSA-Agent is to support a power grid simulation analysis workflow where "humans assign tasks, machines understand and execute those tasks" (referred to as the "human-machine task-driven mode"), thereby achieving a breakthrough in intelligent simulation analysis technology for power systems. The establishment of this mode will elevate artificial intelligence from a "helper" in power grid simulation analysis to a "colleague" capable of assisting personnel in completing tasks, enabling free human-machine collaboration in specialized fields before breakthroughs in strong artificial intelligence are achieved.

[0056] also, Figure 2 This is a flowchart illustrating a power system simulation analysis method provided in an exemplary embodiment of the present invention. (Refer to...) Figure 2 As shown, the power system simulation analysis method 200 includes: S1. Receive natural language task instructions from the user through the interface layer; S2. The main intelligent agent parses the task instructions, matches the corresponding workflow template from the task knowledge base, and combines the instruction parameters to form an executable workflow; S3. The main agent decomposes the workflow into multiple subtasks and schedules the corresponding sub-agents to execute them; S4. During the execution of sub-tasks, the sub-agent completes the task by calling the tool interface in the tool / knowledge layer and using the associated task data object; S5. The sub-agent returns the sub-task execution results to the main agent, which then integrates them and presents the final task results to the user through the interface layer.

[0057] Specifically, the implementation steps of the power system simulation analysis method implemented by the above-mentioned power system simulation analysis intelligent agent system in this embodiment include: Step 1: Build dialogue and tasks iPSA-Agent organizes work execution through dialogue and tasks. Multiple rounds of communication between the human and machine constitute a dialogue, meaning several tasks can be performed within a dialogue. The end of the dialogue and the start of a new one are determined by the human. A single input command from the user, and the corresponding execution and return process of iPSA-Agent, correspond to a task, which is one instance of human-machine communication.

[0058] In traditional large-scale model dialogues, it's typically necessary to consider the contextual relationships within the dialogue. However, in iPSA-Agent dialogues, it's possible to involve not only regular question-and-answer dialogues but also computational and data modification operations, along with related return information. Since tasks other than question-and-answer are essentially data operations, and the basic form of the execution result is also data, to avoid interference, iPSA-Agent dialogues only consider the contextual relationships of question-and-answer tasks. Other types of tasks are connected through relevant task data.

[0059] While question-and-answer content cannot be shared between different dialogues, task data objects can be used. Therefore, tasks other than question-and-answer can essentially be performed across dialogues. The significance of designing dialogues lies in generating analysis or work reports based on their content; thus, a dialogue can also be viewed as a complete work task for a person.

[0060] The tasks in the dialogue correspond to common workflows in simulation analysis, and are combined with task data objects and task execution parameters to form the specific analysis and processing process of the intelligent agent.

[0061] refer to Figure 3As shown, a single human-computer interaction or computational call is defined as a meta-task, such as power grid simulation, AI computation, and data manipulation. A sub-task is composed of a sequence of meta-tasks, and some common meta-tasks can be encapsulated for different sub-tasks to call. A user's command is a task, and a task may contain several sub-tasks. Each sub-task corresponds to a long-term task or a short-term task, and the same sub-task can be called in different tasks. Long-term sub-tasks include all analytical and computational sub-tasks with clearly defined processes, such as generating power flow methods; short-term sub-tasks include three categories: data maintenance sub-tasks (adding, deleting, and modifying simulation data), statistical analysis sub-tasks (querying and statistically analyzing simulation data and results), and question-and-answer sub-tasks. Since each scenario in simulation analysis is relatively complex, a long-term sub-task may contain multiple meta-tasks, while a short-term task may contain only a few meta-tasks.

[0062] Step 2: Establish workflows and skills Tasks, subtasks, and metatasks define the logical structure of the simulation analysis workflow. In the implementation of the intelligent agent, these are achieved through task flow orchestration and advancement, skills, and tools, respectively. During task execution, task flow orchestration is determined by the large model in conjunction with relevant knowledge, parameters, and task requirements, and then autonomously advances the execution.

[0063] For subtasks, since the operations performed are generally deterministic, the specific operations of these steps are relatively fixed and constructed as the agent's "skills." The relationship between skills and tasks here is similar to simulation analysis personnel mastering skills such as using computing programs and statistical calculation results in advance. By clarifying the workflow, i.e., task arrangement, the personnel use these abilities to complete the corresponding work tasks.

[0064] Meta-tasks correspond to specific operations, involving calls to various interfaces, algorithms, functions, and large models. These operations are encapsulated into tools, and according to the correspondence between meta-tasks and sub-tasks, the construction of agent skills is supported.

[0065] Step 3: Execution of workflow based on multi-agent systems After receiving the task instruction, the iPSA-Agent parses the instruction, retrieves the relevant workflow description from the power grid simulation analysis task knowledge base, and forms a workflow based on the parameters in the task instruction.

[0066] The execution of the workflow requires the invocation of iPSA-Agent skills. Skills are encapsulations of tools, which can take the form of a programmatically fixed process that calls the tool interface through MCP, or they can take the form of sub-agents. In this case, the sub-agent, based on the assigned task, autonomously assembles the sub-task execution process by invoking the large model and task knowledge base, and calls the tool interface through MCP during execution. iPSA-Agent adopts a multi-agent approach to implement skills for two main reasons: First, in actual work, even relatively fixed workflows can encounter various problems during execution, requiring complex human-computer interaction and process adjustments. Combined with a task knowledge base, the sub-agent is more adaptable to such situations. Second, it can track human operations and feedback, and by continuously and automatically improving the task knowledge base content related to sub-tasks, it can easily optimize the behavior of the sub-agent, achieving optimal human-machine hybrid behavior.

[0067] Step 4: Establish a data and knowledge environment framework for task execution. The execution of iPSA-Agent tasks requires the support of data and knowledge, mainly including working knowledge, working data, and working conclusions.

[0068] 1. Job-related knowledge. During the task orchestration phase, the large model needs simulation analysis workflow knowledge corresponding to the task scenario; during execution, the agent needs to invoke domain knowledge related to the task. 2. Work Data. During task execution, skill activation needs to be combined with the simulation data required for the task, and the conclusions after task completion will also be reflected in the simulation data. Effective management and storage of work data are necessary for verifying and analyzing task conclusions, as well as improving intelligent agent tools. 3. Work Conclusions. After the task is completed, the conclusions obtained are reflected in the relevant simulation data and can also be summarized as textual and graphical conclusions of the work. These conclusions need to be stored so that they can be used directly when there are similar tasks in the future, or for further analysis, querying and reference when needed.

[0069] The aforementioned work knowledge, work data, work conclusions, and task parameters together constitute the task data object, which is the data and knowledge environment required for task execution.

[0070] Step 5: Workflow Execution Strategy The execution of tasks by intelligent agents mainly includes four types: sequential execution, cyclic execution, branch execution, and exception handling. The main agent connects the outputs of each sub-agent to drive the task execution.

[0071] 1. The general model generates a workflow based on task-related knowledge. Essentially, it is a process framework. The main agent can easily and autonomously drive both sequential execution and branch execution.

[0072] 2. Looping Execution. First, the return results of the sub-Agent objects are recorded in a list. Each sub-Agent object, i.e., each sub-task stage, has an independent, local variable-like list of results. The main agent defines a default operation on the list, processing each item sequentially. The main agent executes each item in the list returned by the sub-agents, forming a loop. If subsequent sub-agents also return lists, the main agent forms a nested loop of task execution.

[0073] 3. Anomaly Handling. Anomalies during task execution typically occur during tool invocation, such as power flow calculation non-convergence. It's important to note that handling calculation or subtask execution results that don't meet requirements is not considered anomaly handling but rather a step within the workflow. The agent establishes process knowledge for handling different types of anomalies, primarily including power flow calculation failures, transient stability calculation failures, and AI model invocation failures. For different error scenarios, dedicated anomaly handling subtasks are established. These subtasks input data from the erroneous tool's inputs and outputs, as well as the sub-agent's tool invocation intent. This primarily guides the agent to modify the data or change the sub-agent's tool invocation object and parameters. After processing, the anomaly handling subtask returns to the state before the tool invocation, triggering the sub-agent to invoke the tool again.

[0074] Step 6: Management of Heterogeneous Task Data Objects The structure and size of iPSA-Agent task data objects vary greatly, therefore, the following aspects should be considered when constructing and managing them: 1. Different data structures are used to store and manage task-related data and knowledge. The power grid simulation analysis task knowledge base records workflows in rule form and records domain knowledge related to workflow execution in triple form; the power grid simulation analysis question-and-answer knowledge base adopts a document library format, based on RAG (Retrieval-Augmented Generation) technology to support human-computer question-and-answer; the task database uses a commercial database to load the simulation data required for task execution; the simulation analysis historical results library records the textual descriptions and reports of analysis results in XML file form; the dialogue history library also records question-and-answer content in XML form.

[0075] 2. Task data objects are labeled and associated using data tags to meet the needs of task execution and data querying. Data tags are stored in the task database and mainly include the task ID, referenced knowledge ID, task data ID, historical result ID, and descriptive tags related to the task, such as the name of the running mode and its characteristics, to facilitate understanding of the task content by the large model.

[0076] 3. When users input task commands, they may omit some parameters, and the input parameters may also be incomplete when calling the tool. Therefore, it is necessary to complete the parameters. Since the large model fine-tuning method requires a large amount of training and the effect has some uncertainty, prompt words are mainly used to explain the parameters and provide default parameter values. On this basis, some important parameters can be confirmed by iPSA-Agent by asking the user.

[0077] 4. Sub-task data environment centered on data labels. Sub-task execution requires a task data object, execution requirements or parameters, and an output execution result data object; these three elements constitute the task's data environment. Unless explicitly instructed by the user, the current task data object remains unchanged. Task data objects are labeled with data tags for user reselection; a data tag is a multi-dimensional vector marking data features from different dimensions. Users can specify task data object tags via text or switch between tags. If no user-specified tags are specified, the main model should automatically complete the data object tag parameters based on the context. For example, the sub-task execution result dataset should have automatically constructed data tags. For question-and-answer sub-tasks, the current task data object remains unchanged, and its execution is unrelated to the current task data object, so no data object tag is required. The basic computational dataset for the agent's runtime environment is given a default data tag, such as all computational data from simulation analysis. To simplify the logic, the system defaults to the basic computational dataset each time the agent executes a task unless a new data object is manually specified.

[0078] To facilitate user input of task execution requirements and ensure standardized input format, it is necessary to construct description templates for common tasks, which users can modify based on.

[0079] Regarding the task execution results, operations on the power grid simulation data directly affect the simulation data itself, are stored in the basic database, and are labeled; statistical analysis results, as well as the generated tables and statistical charts, are stored in files and labeled for later use; and question and answer results are directly returned to the dialogue area.

[0080] Step 6: Invoking relevant system functions iPSA-Agent calls services provided by existing simulation systems, AI platforms, and third-party software through tool interfaces. Since these systems were built at different times and used different technical approaches, it is first necessary to determine the standard specifications for the service interfaces, including data exchange formats and calling methods. Based on this, existing systems are modified to establish the service interfaces; this part involves significant software development work.

[0081] During task execution, iPSA-Agent communicates with various systems via Socket. If the instruction being executed is a single-step command, such as launching a standalone tool on an external system, iPSA-Agent returns directly to the state of waiting for input instructions after the call. If the instruction requires a return result, iPSA-Agent waits for the service interface to complete and return. To prevent failures, a termination waiting instruction is provided, after which iPSA-Agent will either continue executing the task or enter the error handling phase. This avoids the waste of a large amount of work in preceding tasks due to simple task termination.

[0082] In a specific embodiment of the present invention, Example 1 is a long-term task: User input: >Calculate the limit for cross-section A using the summer 2025 method for dataset 1. iPSA-Agent output: Get "2025 Summer Calculation Dataset 1" Based on the predefined quota calculation template for section A, set the quota calculation conditions for section A, ... > Conduct the first round of tidal flow adjustment at section A. Increase power flow at section A, start the generator unit. >Verification section A fault set The verification passed; no instability faults were found. Continue adding power flow section A. ... >Conducting tidal current adjustment at section A of the 17th section. Increase power flow at section A, start the generator unit. >Verification section A fault set Verification failed; Sanyong faulty system unstable. The limit for section A is MW, corresponding to the per-unit value pu. Example 2: Short-term task: User input: >Query the maximum active power of a single load in the North China power grid from 2023 to 2025 iPSA-Agent output: Obtain all calculation datasets for the North China Power Grid from 2023 to 2025, totaling 23 sets. The maximum active power of a single load in the North China power grid is MW, corresponding to the per-unit value pu. Therefore, based on a large model, this invention comprehensively utilizes power system simulation analysis technology, automatic data processing technology, and related artificial intelligence technology to establish a new intelligent application system for various power grid simulation analysis businesses. This breaks through the limitations of existing simulation analysis calculation tools and scenario-based tools, and achieves a leapfrog development of power system simulation analysis in the direction of intelligence.

[0083] Exemplary electronic devices Figure 4 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. For example... Figure 4 As shown, the electronic device 40 includes one or more processors 41 and memory 42.

[0084] The processor 41 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0085] The memory 42 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 41 may execute the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above, and / or other desired functions. In one example, the electronic device may also include an input device 43 and an output device 44, these components being interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0086] In addition, the input device 43 may also include, for example, a keyboard, a mouse, etc.

[0087] The output device 44 can output various information to the outside. The output device 44 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0088] Of course, for the sake of simplicity, Figure 4 Only some of the components of this electronic device relevant to the present invention are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device may include any other suitable components depending on the specific application.

[0089] Exemplary computer program products and computer-readable storage media In addition to the methods and apparatus described above, embodiments of the present invention may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the methods according to various embodiments of the present invention described in the "Exemplary Methods" section above.

[0090] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of the present invention. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0091] Furthermore, embodiments of the present invention may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps of the methods according to various embodiments of the present invention described in the "Exemplary Methods" section above.

[0092] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0093] The basic principles of the present invention have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in the present invention are merely examples and not limitations, and should not be considered as essential features of each embodiment of the present invention. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the present invention to the necessity of employing the aforementioned specific details.

[0094] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they largely correspond to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0095] The block diagrams of devices, systems, devices, and systems involved in this invention are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, systems, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0096] The methods and systems of the present invention may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of the present invention are not limited to the order specifically described above unless otherwise specifically stated. Furthermore, in some embodiments, the present invention may also be implemented as a program recorded on a recording medium, the program comprising machine-readable instructions for implementing the methods according to the present invention. Thus, the present invention also covers recording media storing programs for performing the methods according to the present invention.

[0097] It should also be noted that in the systems, apparatus, and methods of the present invention, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered equivalents of the present invention. The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the invention. Therefore, the invention is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0098] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the invention to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. An electric power system simulation analysis agent system, characterized by, Comprise: An interface layer for receiving user instructions and feeding back task results in multiple modalities; A workflow layer comprising a main agent for parsing the user instructions, generating and driving a workflow for executing simulation analysis according to a task knowledge base, wherein the workflow is composed of multiple task execution strategies in the order of execution, branch execution, loop execution and exception handling; A skill layer comprising multiple sub-agents for encapsulating and executing specific sub-tasks in the workflow; A tool / knowledge layer for providing a standardized calling interface for external tools and storing data and knowledge required for task execution; A business system layer comprising one or more existing power system simulation business systems for providing basic data, computing services and interactive components for the tool / knowledge layer; Wherein the main agent parses the user instructions into tasks and decomposes the tasks into sub-tasks for execution by the corresponding sub-agents; the sub-agents complete the tasks in a collaborative manner by calling the tool interface in the tool / knowledge layer and utilizing the data and knowledge.

2. The system of claim 1, wherein, The task knowledge base in the workflow layer stores simulation analysis workflow templates in the form of rules, and the main agent generates a specific workflow for simulation analysis by combining the parameters in the user instructions with the workflow templates; The strategy for loop execution is to record the return results of sub-agent objects in a list, where each sub-agent object has an independent, locally variable result list, and the main agent defines a default operation on the list that processes each item one by one, the main agent executes each item in the list returned by the sub-agent one by one to form a loop, if the subsequent sub-agents also return a list, the main agent forms a loop nesting of task execution; The strategy for exception handling is that the agent establishes process knowledge for different categories of exception handling for different errors, establishes a special exception handling sub-task, takes the input and output of the error tool and the tool calling intention of the sub-agent as the input data of the exception handling sub-task, modifies the data, or changes the tool calling object and parameters of the sub-agent, after processing, the exception handling sub-task returns to the tool calling before, triggering the sub-agent to call the tool again, wherein the exception handling includes power flow calculation failure, transient stability calculation failure and AI model calling failure.

3. The system of claim 2, wherein, The workflow template uses vectorization encoding, and the word vector of the vectorization encoding adds sine / cosine position encoding: set, is the position of a word in a sequence, is the dimension of a word vector, i is a dimension index; The workflow template is used to generate the multi-head attention mechanism in the workflow: Wherein: In the formula, , , These are matrices representing queries, keys, and values; Let be the dimension of the key vector. head i For the first i One's attention, h For the number of attention heads; W 0 To output the projection matrix; W i V For the first i The value transformation matrix of each head; W i K For the first i The key transformation matrix of the head; W i Q For the first i The query transformation matrix for each head; After each attention sub-layer, there is a two-layer fully connected feedforward network: wherein, x is the input, and W 1is the first layer weight matrix; b 1is the first layer bias; W 2is the second layer weight matrix; b 2is the second layer bias.

4. The system of claim 1, wherein, The sub-agents in the skill layer include business class sub-agents, auxiliary class sub-agents and knowledge class sub-agents; The business class sub-agents are used to execute explicit power grid simulation analysis long-line sub-tasks in the workflow; The auxiliary class sub-agents are used to execute short-line sub-tasks of data maintenance, query and statistics in the workflow; The knowledge class sub-agents are used to execute domain knowledge question and answer sub-tasks in the workflow.

5. The system of claim 4, wherein, The business class sub-agents include power grid control skill models: wherein: is a control vector; is a gain coefficient vector; is a grid target state vector; is a grid current state vector, wherein the grid target state vector is obtained by a knowledge class sub-agent, the grid current state is queried by a business class sub-agent, the gain coefficient vector is calculated or queried by a business class sub-agent, and finally the control vector is calculated, the control vector is acted on the grid data by a control execution skill to realize adjustment of the grid data, and then a control effect verification is performed by a grid simulation skill.

6. The system of claim 5, wherein, The power grid control skill model comprises the following steps: An initialization step, starting from the current actual operation state of the power grid, calculating a comprehensive score of the current actual operation state of the power grid on the target requirement, and putting the current actual operation state as an adjustment path with a length of 1 into the initial beam of the next round of control, while recording that the path has not used any action type; A multi-round iterative expansion step, performing the following sub-steps on each adjustment path in the current beam of the current round of control: (a) Dynamically generating an actionable action set based on the current power grid topology corresponding to the adjustment path; (b) Simulating the execution of each executable action in the actionable action set to obtain a new state and calculate its comprehensive score, wherein the comprehensive score = target matching degree + safety margin - operation cost; (c) Grouping the candidate actions corresponding to all new states according to action types, wherein in each action type group, the candidate actions are sorted according to the score, and an exponential decay penalty is applied to the lower-ranked ones, and an additional reward is given to an action type that has never been used in the current path, to obtain the processed candidate actions; (d) After combining all processed candidate actions, the top b paths with the highest comprehensive scores are retained to form the current beam for the next round of input; An output result step, after the multi-round iterative expansion step terminates, returning all adjustment paths in the final current beam as a complete scheme.

7. The system of claim 1, wherein, The tool / knowledge layer comprises: A knowledge base for storing working knowledge, working data and working conclusions, the working knowledge including working process knowledge stored in the form of rules and domain knowledge stored in the form of triples, and the working conclusions including analysis reports stored in the form of XML files; A tool service for service-based encapsulation of simulation calculation programs, AI models and database operations based on a model context protocol (MCP) to form a standardized tool interface.

8. A power system simulation analysis method based on the system of any one of claims 1-7, characterized by, Comprise: S1. Receiving a natural language task instruction of a user through the interface layer; S2. The main agent analyzes the task instruction, matches a corresponding work flow template from a task knowledge base, and forms an executable work flow in combination with instruction parameters, wherein the work flow is composed of multiple task execution strategies for sequential execution, branch execution, loop execution and exception handling; S3. The main agent decomposes the work flow into multiple sub-tasks and schedules corresponding sub-agents to execute; S4. The sub-agents complete the tasks by calling the tool interface in the tool / knowledge layer and utilizing the associated task data objects during the execution of the sub-tasks; S5. The sub-agents return the sub-task execution results to the main agent, which integrates them and feeds back the final task results to the user through the interface layer.

9. The method of claim 8, wherein, In step S4, when calling the tool interface, if the required parameters are not complete, the system automatically completes the default parameters through prompt word engineering based on the context and data tags of the current task data object, or initiates confirmation to the user for the key parameters.

10. The method of claim 8, wherein, One or more consecutive interactions of a user constitute a conversation, one said conversation containing one or more tasks; said tasks are linked by their associated task data objects, while the contents of the question-and-answer subtasks within said conversation are stored separately in a conversation history library.

11. A computer readable storage medium, characterized in that, The storage medium stores a computer program for executing the method of any one of claims 8-10.

12. An electronic device, comprising: The electronic device comprises: a processor; a memory for storing executable instructions of the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method of any one of claims 8-10.

Images