Multi-agent autonomous cooperative decision method and related device of power multi-service system

Abstract

This application pertains to collaborative decision-making methods for power systems. It addresses the challenges of accurately mapping the heterogeneous representation of the same power business object across different power-specific business systems, and the difficulty of existing formal methods in verifying the complete compliance of cross-system operation sequences, leading to difficulties in autonomous collaborative execution of cross-system operations. The application provides a multi-agent autonomous collaborative decision-making method and related apparatus for power multi-business systems. It achieves bidirectional mapping between unified object identifiers and local object identifiers in each business system through a cross-system semantic alignment graph, reducing the risk of cross-system object mismatch. It improves the compliance of candidate sub-operation sequences through hierarchical quantization of linear temporal logic formulas and formal verification of grid state traces. Finally, it relies on multi-agent collaboration to complete task parsing, object mapping, procedure verification, sub-operation execution, and result feedback, enabling autonomous collaborative operations among power multi-business systems.

Description

Technical Field

[0001] This application pertains to collaborative decision-making methods for power systems, specifically relating to a multi-agent autonomous collaborative decision-making method and related apparatus for a multi-service power system. Background Technology

[0002] Power grid companies rely on multiple proprietary power systems for collaborative operations, including production, dispatching, marketing, and centralized control. Cross-system operations involve multiple roles and various power business objects. The same power business object, such as a line, transformer, or circuit breaker, uses different identifiers, data structures, and business semantics in different business systems, resulting in significant heterogeneity in data models between power systems.

[0003] Existing cross-system collaborative operations are typically performed by personnel in different positions operating their respective responsible business systems, or by using intelligent agents with a general large language model as the inference engine to perform cross-system operations through screen simulation, interface calls, and other methods. However, general system alignment or ontology matching mainly relies on field names and attribute features, making it difficult to accurately establish the correspondence between the same power business objects in heterogeneous systems. At the same time, standard LTL (Linear Temporal Logic) and PF-LTL (Past-Future Linear Temporal Logic) are propositional logics, which are difficult to express simultaneously the quantization of object sets with topological scope, power grid topology relationships, and actual time boundaries, thus making it difficult to perform complete procedural compliance verification of cross-system operation sequences. Summary of the Invention

[0004] This application addresses the technical problem that the same power business object is expressed heterogeneously in different power proprietary business systems, making it difficult to accurately map, and that existing formal methods are unable to perform complete procedural compliance verification of cross-system operation sequences, resulting in difficulties in autonomous and collaborative execution of cross-system linkage operations. It provides a multi-agent autonomous collaborative decision-making method and related devices for power multi-business systems.

[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application proposes a multi-agent autonomous collaborative decision-making method for a power multi-service system, based on a multi-agent collaborative architecture including a coordinating agent, several system-specific agents corresponding to different power-specific service systems, a semantic alignment agent, and a procedure verification agent; including: By coordinating the intelligent agent to receive natural language descriptions of power business tasks, identifying the task intent and target power business objects of the power business tasks, retrieving applicable business procedures, and generating cross-system candidate sub-operation sequences; Based on a pre-built cross-system semantic alignment graph, the semantic alignment agent determines the cross-system unified object identifier corresponding to the target power business object, and represents the power business object in the cross-system candidate sub-operation sequence based on the cross-system unified object identifier. The procedure verification agent obtains the hierarchical quantized linear temporal logic formula corresponding to the applicable business procedure from the pre-built formal knowledge base of power business procedures. It generates the power grid state trace based on the cross-system candidate sub-operation sequence, reduces the hierarchical quantized linear temporal logic formula to the propositional level linear temporal logic formula with past tense, and performs bounded model detection to obtain the formal verification result. Based on the formal verification results, the final cross-system candidate sub-operation sequence is determined as the executable sub-operation sequence; the coordinating agent assigns sub-operations to the corresponding system-specific agents according to the executable sub-operation sequence. Based on the cross-system semantic alignment graph, the semantic alignment agent maps the cross-system unified object identifier in the sub-operation to the local object identifier in the corresponding power proprietary business system. The corresponding system-specific agent then executes the sub-operation in the corresponding power proprietary business system based on the local object identifier and returns the execution result. The semantic alignment agent reverse-maps the local object identifier in the execution result to a unified object identifier across the system. The coordination agent determines the next sub-operation in the executable sub-operation sequence based on the execution result after the reverse mapping, until the executable sub-operation sequence is completed.

[0006] Furthermore, the coordinating agent, system-specific agent, semantic alignment agent, and procedure verification agent all use the Guangming Power big model as the underlying inference engine. Each of the system-specific intelligent agents is pre-configured with the data dictionary and interface specifications of its corresponding power proprietary business system, which are used to perform read or write operations in the corresponding power proprietary business system.

[0007] Furthermore, the pre-constructed cross-system semantic alignment graph is obtained in the following way: Analyze the data dictionary of each power proprietary business system and construct the heterogeneous attribute graph corresponding to each power proprietary business system; Construct a specification ontology diagram based on the CIM-E or CIM-G standards; Based on the substation level, transmission and distribution line level, main transformer level, and distribution transformer and distribution area level in the power grid physical topology, a four-level anchor point structure of station, line, transformer and distribution area is constructed. Based on the four-level hierarchical anchor point structure, the power business object nodes in the heterogeneous attribute graph and the candidate nodes in the standard ontology graph are hierarchically anchored and aligned. For each node in the specified ontology graph, collect the power business objects aligned to that node from each power proprietary business system, generate a cross-system unified object identifier for that node, and persist the cross-system unified object identifier and its correspondence with each local object identifier to the cross-system semantic alignment graph.

[0008] Furthermore, the power grid model corresponding to the hierarchical quantization linear temporal logic formula includes a finite set of power business objects, object type functions, a set of power grid topology relationships, and a four-level hierarchical structure of the power grid; The object type function is used to map each power business object in the limited power business object set to the corresponding power business object type. The set of power grid topology relationships is used to characterize the power grid topology relationships between power business objects in a limited set of power business objects. The four-level hierarchical structure of the power grid is used to characterize the hierarchical position and hierarchical affiliation of power business objects in the limited set of power business objects within the four-level anchor point structure of station, line, transformer, and distribution station.

[0009] Furthermore, the hierarchical quantization linear temporal logic formula is based on linear temporal logic with past tense, including state predicates, numerical predicates, power grid topology relation predicates, Boolean operators, future temporal operators, past temporal operators, bounded metric time operators, and hierarchical typified quantization operators; The hierarchical quantization linear temporal logic formula is reduced to a propositional linear temporal logic formula with past tense, and bounded model testing is performed, including: Based on the power business object type and quantization scope defined by the hierarchical typified quantization operator, the hierarchical quantization linear temporal logic formula is quantized and baseified, the universal quantization is expanded into the conjunction of the corresponding power business object instance, and the existential quantization is expanded into the disjunction of the corresponding power business object instance. The power grid topology relation predicates in the quantized and basicized hierarchical quantized linear temporal logic formula are evaluated based on the power grid model, and the evaluation results are replaced with Boolean constants. The numerical predicates are evaluated based on the measured values ​​at each time step in the power grid state trace, and the evaluation results are replaced with Boolean atomic propositions. The bounded time operator is expanded into a propositional formula consisting of future or past tense operators, resulting in the propositional linear temporal logic formula with past tense. Based on the propositional linear temporal logic formula with past tense and the power grid state trace, a bounded model test is performed to obtain the formal verification result.

[0010] Furthermore, based on the formal verification results, the final cross-system candidate sub-operation sequence is determined as the executable sub-operation sequence, including: When the formal verification result is not passed, a structured counterexample is generated by the procedure verification agent; the structured counterexample includes the earliest time step in which the applicable business procedure is violated on the power grid state trace, the violated hierarchical quantized linear temporal logic formula, the power business object instance that caused the violation determined during quantization, and the violated sub-formula. The structured counterexamples are transcribed into replanning hints by a coordinating agent, and the Guangming Power big model is invoked to generate a corrected cross-system candidate sub-operation sequence based on the replanning hints. The corrected cross-system candidate sub-operation sequence is resubmitted to the procedure verification agent for formal verification until the formal verification is passed or the preset maximum number of iterations is reached. The cross-system candidate sub-operation sequences that pass formal verification are determined as the executable sub-operation sequences.

[0011] Furthermore, the next sub-operation in the sequence of executable sub-operations is determined by coordinating the agent based on the execution result after reverse mapping, including: When the execution result of the current sub-operation meets expectations, the next sub-operation is taken from the sequence of executable sub-operations; when the execution result of the current sub-operation is abnormal, the Guangming Power big model is called to evaluate the impact of the abnormality on subsequent sub-operations, and the current sub-operation is retried, the current sub-operation is skipped and subsequent sub-operations are adjusted, or the execution is terminated and the abnormality is reported based on the evaluation result. After all sub-operations in the executable sub-operation sequence have been executed, the following steps are also included: The system aggregates status changes in the unified business object view and generates an execution result report for cross-system collaborative operations. The execution result report includes the execution status of each sub-operation, details of changes to the power business objects involved, and time consumption statistics.

[0012] Secondly, this application proposes a multi-agent autonomous collaborative decision-making system for a power multi-business system, including a coordinating agent, several system-specific agents corresponding to different power proprietary business systems, a semantic alignment agent, and a procedure verification agent. The coordinating agent, system-specific agent, semantic alignment agent, and procedure verification agent operate according to the multi-agent autonomous collaborative decision-making method of the power multi-service system described above.

[0013] Thirdly, this application proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-agent autonomous collaborative decision-making method of the above-mentioned power multi-service system.

[0014] Fourthly, this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the multi-agent autonomous collaborative decision-making method for the aforementioned power multi-service system.

[0015] Compared with the prior art, this application has the following beneficial effects: This application proposes a multi-agent autonomous collaborative decision-making method for power multi-service systems. It determines a unified cross-system object identifier for target power service objects through a cross-system semantic alignment graph. Before and after sub-operation execution, a bidirectional mapping is established between the unified cross-system object identifier and the local object identifiers of each power-specific service system. This enables different power-specific service systems to transmit information and execute operations based on the same power service object, reducing cross-system object mismatches caused by inconsistencies in object identifiers and representations. Simultaneously, by reducing the hierarchical quantized linear temporal logic formula corresponding to the applicable business procedures to a propositional-level linear temporal logic formula with past tense, and combining this with bounded model detection using grid state traces, it can automatically perform formal verification of cross-system candidate sub-operation sequences before sub-operation execution, ensuring that the verified executable sub-operation sequences meet the requirements of the corresponding business procedures. Furthermore, by coordinating the collaborative processing among intelligent agents, semantic alignment intelligent agents, procedure verification intelligent agents, and system-specific intelligent agents, task parsing, object mapping, procedure verification, sub-operation assignment, business system execution, and execution result feedback are connected into a continuous work process. The next sub-operation is determined based on the execution result after reverse mapping, thereby enabling autonomous collaborative execution among multiple power-specific business systems and improving the automation level and procedure compliance of cross-system linkage operations.

[0016] This application also proposes a multi-agent autonomous collaborative decision-making system for a power multi-service system, an electronic device, and a computer-readable storage medium, which possess the advantages of the aforementioned multi-agent autonomous collaborative decision-making method for power multi-service systems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a multi-agent autonomous collaborative decision-making method for a power multi-service system as described in this application. Figure 2 This is a schematic diagram of the multi-agent collaborative architecture and cross-system operation in the embodiments of this application; Figure 3 This is a schematic diagram of the cross-system semantic alignment graph process anchored based on the four-level hierarchical structure of the power grid in the embodiments of this application; Figure 4 This is a schematic diagram of the anchor point identification of the four-level hierarchical structure of the power grid in this application embodiment; Figure 5 This is a schematic diagram of the HQ-LTL (Hierarchical Quantized Linear Temporal Logic) syntax and three-term extended structure in the embodiments of this application; Figure 6 This is a flowchart of the five-step reduction process of the HQ-LTL quantization-base-verification algorithm in the embodiments of this application; Figure 7 This is a timing diagram of multi-agent collaborative execution and semantic alignment mapping in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] The multi-agent autonomous collaborative decision-making method for power multi-service systems proposed in this application is applied to a multi-agent collaborative architecture. This architecture includes a coordinating agent, several system-specific agents, a semantic alignment agent, and a procedure verification agent. Each system-specific agent corresponds to a different power-specific service system and is used to pass corresponding sub-operations in cross-system collaborative operations to the corresponding power-specific service system for execution. The coordinating agent receives tasks, organizes cross-system candidate sub-operation sequences, controls the execution order of sub-operations, and maintains the cross-system task context. The semantic alignment agent eliminates differences in object representation between different power-specific service systems. The procedure verification agent performs formal compliance verification on the cross-system candidate sub-operation sequences before the actual execution of sub-operations.

[0022] like Figure 1The diagram shown is a flowchart illustrating a multi-agent autonomous collaborative decision-making method for a power multi-service system according to this application. Specifically, when executing power service tasks, it may include: S101, by coordinating the intelligent agent to receive the natural language description of the power business task, identify the task intent and target power business object of the power business task, retrieve the applicable business procedures, and generate a cross-system candidate sub-operation sequence.

[0023] The coordinating agent receives natural language descriptions input by business personnel, upper-level business applications, or other task initiators, parses these descriptions, and identifies the task intent and target power business objects. The task intent characterizes the power business actions expected to be completed in this power business task, while the target power business objects define the lines, transformers, circuit breakers, disconnect switches, substations, distribution areas, or other power business objects targeted by each business action.

[0024] The coordinating agent retrieves applicable business procedures based on the task intent and target power business object, and decomposes the power business task into several sub-operations spanning one or more power-specific business systems, forming a cross-system candidate sub-operation sequence. This cross-system candidate sub-operation sequence describes the execution object, target system, operation content, and sequential relationships of each sub-operation. Before formal verification is completed, this cross-system candidate sub-operation sequence is treated as a sequence to be verified and is not directly executed in the corresponding power-specific business system.

[0025] S102, the semantic alignment agent determines the cross-system unified object identifier corresponding to the target power business object based on the pre-built cross-system semantic alignment graph, and represents the power business object in the cross-system candidate sub-operation sequence based on the cross-system unified object identifier.

[0026] Since the same power business object may use different local object identifiers in different power proprietary business systems, the semantic alignment agent, based on a pre-built cross-system semantic alignment graph, queries the correspondence of the target power business object in different power proprietary business systems, determines the cross-system unified object identifier corresponding to the target power business object, and uses this cross-system unified object identifier to represent the power business object in the cross-system candidate sub-operation sequence. Thus, the cross-system candidate sub-operation sequence maintains a unified object semantics before entering a specific power proprietary business system, avoiding sub-operations pointing to the wrong power business object due to differences in object name, field structure, or local encoding.

[0027] S103: The procedure verification agent obtains the hierarchical quantized linear temporal logic formula corresponding to the applicable business procedure from the pre-built formal knowledge base of power business procedures. It generates the power grid state trace according to the cross-system candidate sub-operation sequence, reduces the hierarchical quantized linear temporal logic formula to the propositional level linear temporal logic formula with past tense, and performs bounded model detection to obtain the formal verification result.

[0028] In this embodiment, a formal knowledge base for power business procedures pre-establishes a correspondence between business procedures and formal constraints, so as to convert business requirements in natural language or clause form into logical expressions that can be formally judged. The power grid state trace is used to describe the changes in the state of power business objects over time steps during the execution of cross-system candidate sub-operation sequences. When generating the power grid state trace, the power grid state before the execution of candidate sub-operations can be used as the initial state, and the states of relevant power business objects after each sub-operation are determined sequentially according to the order of the sub-operations. The initial state and multiple subsequent states constitute a finite-length state sequence. Considering that the hierarchical quantization linear temporal logic formula contains constraints oriented towards power business object types, object sets, or object levels, and that bounded model detection requires judgment on specific states, the above hierarchical quantization linear temporal logic formula needs to be converted into a propositional expression suitable for verification on finite state sequences. The converted formula still retains the object state relationships, operation sequence relationships, and temporal relationships defined by the business procedures, thus enabling it to be used to determine whether a state violating the business procedures occurs during the execution of a cross-system candidate sub-operation sequence. Bounded model verification is performed within a finite timeframe defined by the power grid state trace. By verifying whether the state and changes of power business objects at each time step meet the corresponding logical constraints, it determines whether the cross-system candidate sub-operation sequence conforms to applicable business procedures. The resulting formal verification results are provided to the coordinating agent as the basis for retaining, adjusting, or regenerating the cross-system candidate sub-operation sequence, thereby performing an overall compliance check on the cross-system operation process before the actual execution of the sub-operation.

[0029] S104. Based on the formal verification results, determine the final cross-system candidate sub-operation sequence as the executable sub-operation sequence; and assign sub-operations to the corresponding system-specific agents according to the executable sub-operation sequence through the coordinating agent.

[0030] In practical applications, when a cross-system candidate sub-operation sequence meets the applicable business procedures, the cross-system candidate sub-operation sequence is determined as an executable sub-operation sequence; when a cross-system candidate sub-operation sequence does not meet the applicable business procedures, the cross-system candidate sub-operation sequence is corrected according to the formal verification results, and the formal verification is re-implemented on the corrected sequence until a sequence that meets the applicable business procedures is obtained or the current planning process ends.

[0031] S105, based on the cross-system semantic alignment graph, the semantic alignment agent maps the cross-system unified object identifier in the sub-operation to the local object identifier in the corresponding power proprietary business system. The corresponding system-specific agent executes the sub-operation in the corresponding power proprietary business system based on the local object identifier and returns the execution result.

[0032] After obtaining the sequence of executable sub-operations, the coordinating agent assigns the currently pending sub-operation to the system-specific agent corresponding to its target power proprietary business system, according to the order of the sub-operations in the sequence. During the coordination and assignment phases, sub-operations use a unified cross-system object identifier to maintain consistency in the task context, target object state, and execution progress maintained by the coordinating agent.

[0033] S106, the semantic alignment agent reverse maps the local object identifier in the execution result to a cross-system unified object identifier, and the coordination agent determines the next sub-operation in the executable sub-operation sequence based on the execution result after reverse mapping, until the executable sub-operation sequence is completed.

[0034] The semantic alignment agent, based on a cross-system semantic alignment graph, maps the unified cross-system object identifier in the current sub-operation to the local object identifier used by the target power proprietary business system. The corresponding system-specific agent then locates the power business object in the target power proprietary business system based on this local object identifier, executes the corresponding sub-operation, and obtains the execution result. The execution result reflects whether the sub-operation was successful and may include returned data, post-execution status, or exception information, depending on the specific operation.

[0035] The execution result returned by the system-specific intelligent agent uses the local object identifier in the corresponding power proprietary business system. The semantic alignment intelligent agent reverse maps the local object identifier in the execution result to a unified object identifier across systems, enabling the execution result to establish a consistent association with the target object currently maintained by the coordinating intelligent agent and the previous execution results. The coordinating intelligent agent determines the completion status of the current sub-operation based on the reverse-mapped execution result, and determines the next sub-operation in the executable sub-operation sequence accordingly. The above assignment, forward mapping, local execution, result feedback, reverse mapping, and next step decision process are cyclically repeated until the executable sub-operation sequence is completed.

[0036] This application utilizes a unified cross-system object identifier in cross-system candidate sub-operation sequences. When issuing sub-operations to a specific power proprietary business system, the unified cross-system object identifier is mapped to a local object identifier within that power proprietary business system. Simultaneously, the local object identifier in the execution result is mapped back to the unified cross-system object identifier. Based on the data dictionaries and interface specifications of each power proprietary business system, this allows for accurate mapping of the same power business object across different power proprietary business systems, enabling system-specific intelligent agents to accurately locate the operation object within the corresponding power proprietary business system. Furthermore, by formally validating the cross-system candidate sub-operation sequences, it is possible to determine whether the state of the power business object formed by the cross-system candidate sub-operation sequences and the order of operations conform to applicable business procedures before the actual execution of the sub-operations. This improves the accuracy and compliance of collaborative execution in cross-system coordinated operations.

[0037] In existing power multi-service collaboration scenarios, cross-system collaborative operations typically involve multiple proprietary power business systems, such as power production management systems, power dispatch automation systems, power marketing management systems, and power centralized control systems. The same 10kV line, the same distribution transformer, or the same switching station may use different identifiers, data structures, and business semantics in different proprietary power business systems. When using multi-person, multi-position manual collaboration, business personnel in different positions need to enter their respective responsible systems to perform operations, relying on manual transmission of target object information and operation results. When using intelligent agents with general large language models as inference engines for screen simulation or interface calls, it is difficult to utilize the domain priors provided by the power grid physical topology, easily leading to incorrect mappings due to differences in field names, object encoding, and semantic expressions. In addition to single-device state constraints, power business procedures also include quantitative constraints on object sets, relational constraints dependent on power grid topology relationships, and measurement time constraints with actual time boundaries. For example, procedures may require all disconnecting switches in a circuit to be in the open position, two circuit breakers in a parallel path not to be closed simultaneously, or grounding operations to be completed within a specified number of seconds after voltage testing. Using only propositional linear temporal logic makes it difficult to fully express the above constraints while preserving the object set, topological scope, and actual temporal semantics.

[0038] In response to the above problems, such as Figure 2 The diagram illustrates a multi-agent collaborative architecture and cross-system operations. In one embodiment of this application, a multi-agent collaborative architecture is constructed, comprising a coordinating agent, several system-specific agents, a semantic alignment agent, and a procedure verification agent. All of the aforementioned agents use the Guangming Power large-scale model as their underlying inference engine. Different system-specific agents can respectively interface with the power production management system (…). Figure 2 System-specific intelligent agents in production management), power dispatch automation system ( Figure 2The system's dedicated intelligent agent - automation), power marketing management system ( Figure 2 The system-specific intelligent agent (marketing) and the centralized power control system ( Figure 2 The specific system type and number of system-specific intelligent agents (centralized control) can be adjusted according to the actual deployment scope, and it is not required that all system-specific intelligent agents be called for every task.

[0039] Each system-specific intelligent agent is pre-configured with its corresponding power-specific business system's data dictionary and interface specifications. The data dictionary provides object types, field names, field semantics, local object identifiers, and object relationships, while the interface specifications provide the invocation methods, input fields, return fields, and exception return formats for read or write operations. The system-specific intelligent agent generates operation requests that conform to the requirements of the target power-specific business system based on the data dictionary and interface specifications, and parses the execution results returned by the target power-specific business system.

[0040] The coordinating agent, acting as the task scheduling hub, receives external task input, parses the natural language task, identifies the task intent, target power business object, and applicable business procedures, generates cross-system candidate sub-operation sequences, and initiates object mapping, procedure verification, and execution requests to other agents. The semantic alignment agent queries the cross-system semantic alignment graph to complete the bidirectional conversion between the unified cross-system object identifier and the local object identifier of each system. The procedure verification agent queries the formal knowledge base of power business procedures to perform formal verification on the cross-system candidate sub-operation sequences. Each system-specific agent transmits operation requests and execution results with the corresponding power-specific business system. Each agent invokes the Guangming Power Big Model as needed for task parsing, reasoning, or replanning, thus forming a collaborative process where task parsing, procedure verification, cross-system execution, and result aggregation are interconnected.

[0041] Before implementing cross-system collaborative operations, a cross-system semantic alignment graph should be constructed in advance. For example... Figure 3 The diagram illustrates the process of cross-system semantic alignment graph anchored based on the four-level hierarchical structure of the power grid. The data dictionaries of each power-specific business system are analyzed, including the marketing data dictionary, centralized control data dictionary, production management data dictionary, and automation data dictionary. A heterogeneous attribute graph G is constructed for the i-th power-specific business system. i : G i = (V i E i A i ) Among them, V i Let E be the set of object nodes in the i-th power proprietary service system. i Let A be the set of associated edges between object nodes in the i-th power proprietary business system.i This represents the set of attributes corresponding to object nodes and associated edges in the i-th power proprietary business system. Heterogeneous attribute graphs can preserve the original data organization methods and object association information of each power proprietary business system.

[0042] Simultaneously, a specification ontology graph G is constructed based on either the CIM-E (Common Information Model based efficient model exchange format) standard or the CIM-G (Common Information Model based graphic exchange format) standard. CIM : G CIM = (V CIM E CIM A CIM ) Among them, V CIM E represents the set of standard object nodes in the i-th power proprietary business system. CIM Let A be the set of standard association edges between object nodes in the i-th power proprietary business system. CIM For the i-th power proprietary business system, it is the standard attribute set corresponding to the object node and associated edge. The nodes in the standard ontology graph can cover power grid object types such as Substation, PowerTransformer, AC Line Segment, Breaker, Disconnector, and DistributionTransformer. It provides a common reference for objects in different power proprietary business systems through standardized object types, attributes and topological relationships.

[0043] like Figure 4 The diagram shown illustrates the anchor point identification in a four-level hierarchical anchor point structure within a power grid topology. Further, the four-level hierarchical anchor point structure (station, line, transformer, and substation) is extracted from the power grid physical topology. The four-level hierarchical set H is represented as: H = {Sta, Line, Trans, Dist} Where Sta represents the substation level, Line represents the transmission and distribution line level, Trans represents the main transformer level, and Dist represents the distribution transformer and substation level. The hierarchical relationships between each level are established through the parent-child mapping function Parent, and satisfy the following conditions: Parent(Dist) = Trans, Parent(Trans) ∈ {Sta, Line}, Parent(Line) = Sta, Parent(Sta) = ⊥ Here, Parent(Dist), Parent(Trans), Parent(Line), and Parent(Sta) are the parent-child mapping functions corresponding to Dist, Trans, Line, and Sta, respectively. ⊥ indicates empty or non-existent.

[0044] Additionally, Level(v) is the label function for the specific level of the power business object v within the four-level hierarchical anchor structure. h' (v) is a function to obtain the unique identifier of the parent node of the power business object v at level h. Level(v) ∈ H is used to record the level where the power business object v is located. Anchor h (v) is used to record the upper-level anchor point of the power business object v at the h level. The substation level is located at the top of the four-level anchor point structure. Transmission and distribution lines are connected to the substation, and the main transformer can be connected to either the substation or the transmission and distribution lines. Distribution transformers and transformer substations are located at the lower level. Anchors are queried along the above hierarchical structure. h (v) can obtain the affiliation anchor point of the power business object in one or more superior levels.

[0045] In a preferred embodiment of this application, a strong constraint on hierarchical anchor point consistency is used to limit the candidate alignment range of heterogeneous attribute graph nodes. For the i-th power proprietary business system, an alignment function φ for the i-th power proprietary business system is defined. i : φ i V i → V CIM v ∈ V i At that time, candidate alignment node φ i (v) satisfies: h' ∈ {h : Parent (h) Level(v)} : Anchor h' (φ i (v)) = ψ h' (Anchor h' (v)) Where h' is a variable that iterates through all ancestor levels of the power business object v (starting from the level where v itself is located, ascending level by level until the root node Sta), and Anchor is a variable that iterates through all ancestor levels of the power business object v. h'(v) is a function to obtain the unique identifier of the parent node of the power business object v at a specified ancestor level h'. For the transitive closure of Parent, ψ h' This is the alignment mapping already established at level h'. The above constraints ensure that the anchor points of the nodes to be aligned in different systems—the upper-level substations, lines, main transformers, or distribution areas—remain consistent, avoiding incorrect association of objects located in different physical topologies based solely on field name similarity.

[0046] Within the range of candidate nodes that satisfy the strong constraint of hierarchical anchor point consistency, the following scoring function, Score, can be used to determine the alignment target: Score(v, u) = λ H · H(v, u) + λ T · T(v, u) + λ A · A(v, u) Wherein, H(v, u) is the consistency term between the power business object v and the candidate node u, used to characterize the consistency between the upper-level anchor point of the business object v and the corresponding upper-level anchor point of the candidate node u in the fourth-level hierarchy; T(v, u) is the topological neighbor consistency term between the power business object v and the candidate node u; and A(v, u) is the attribute similarity term between the power business object v and the candidate node u based on the embedding vector of the Guangming Power large model. The weights of the three terms satisfy: λ H + λ T + λ A = 1 Where, λ H , λ T and λ A Let λ be the weights corresponding to H(v, u), T(v, u), and A(v, u), respectively. H , λ T and λ A The settings can be adjusted according to the data quality of different power proprietary business systems to adjust the impact of hierarchical structure, topological neighbors, and attribute semantics on the final alignment result.

[0047] As a preferred solution method, Figure 3The alignment process illustrated employs a hierarchical, two-stage solution algorithm. First, nodes are processed layer by layer in the order Sta → Line → Trans → Dist. The node with the highest score among the CIM-E candidate nodes at the same level that satisfies the strong constraint of hierarchical anchor point consistency is selected as the alignment target. After the anchor nodes at each level are established, the aligned nodes are used as seeds, and the alignment is propagated to adjacent nodes using a label propagation algorithm to complete the alignment of non-anchor nodes within the same level. This process embodies the solution sequence of first using the power grid physical topology for deterministic anchoring, and then performing heuristic diffusion within a limited neighborhood.

[0048] The complexity of traditional prior-free graph alignment algorithms is: O(n · m) Where n is the heterogeneous attribute graph G of the source power proprietary business system. i The number of nodes in the graph, i.e., the total number of power business objects within the power proprietary business system, where m is the number of nodes in the specification ontology graph G. CIM The number of nodes in the standard power grid model refers to the total number of power business objects in the model.

[0049] After adopting a four-level hierarchical anchor point structure and a two-stage hierarchical progressive solution, the complexity is reduced to: O(Σ h |V i h | · |V CIM h | + n · k) Among them, V i h In the heterogeneous attribute graph of the i-th source power proprietary service system, the attribute belonging to the th A collection of power business objects at the h-level (such as Sta, Line, Trans, Dist), V CIM h In the standard CIM-E ontology graph, k represents the set of power business objects belonging to level h, where k is the average node degree in the power topology graph.

[0050] In a real-world power grid scenario, this complexity is reduced to 1 / 10² to 1 / 10³ of the original. The anchoring phase searches only within the set of nodes at the same level, and the diffusion phase searches only within the neighbors of already aligned nodes, thus significantly narrowing the candidate space for cross-system object alignment.

[0051] After solving the alignment function, for each candidate node u in the standard ontology graph, all power business objects aligned to that node in each power proprietary business system are collected. A cross-system unified object identifier is generated for that node, and the cross-system unified object identifier, the local object identifiers of each power proprietary business system, and the correspondence between them are persisted to the cross-system semantic alignment graph. Subsequent candidate sub-operation sequences use the cross-system unified object identifier to represent power business objects. Before the system-specific intelligent agent calls the target system interface, the cross-system unified object identifier is converted into the corresponding local object identifier.

[0052] G i G CIM Together with the four-level hierarchical anchor point structure, it forms Figure 3 The input for the alignment process shown. Figure 3 The document demonstrates the process from data dictionary parsing, heterogeneous attribute graph and standardized ontology graph construction, hierarchical anchor point consistency determination, two-stage hierarchical progressive solution, to cross-system unified object identifier generation, following the order of data preparation, anchor point structure identification, constraint solving, and result output. Figure 4 This further demonstrates the parent-child containment relationship between the four levels and the Anchor. h The hierarchical anchor point identification function is applicable to each level. The parent-child relationship is used to limit the anchoring order from top to bottom. The anchor point identification function is used to compare whether objects in different power proprietary business systems belong to the same upper-level physical topology unit.

[0053] The formal knowledge base for power business procedures can be pre-constructed by power professionals based on business procedures such as the "Power System Dispatch Procedures," "Power Safety Work Procedures," "Substation Switching Operation Guidelines," and "Distribution Network Operation Procedures." Each natural language procedure is converted into hierarchical, quantified, linear temporal logic formulas, and a correspondence is established between these formulas and applicable business types, object types, and procedure clauses. After recognizing the task intent, the coordinating agent retrieves applicable business procedures based on the task type and target power business object, and the procedure verification agent obtains the corresponding formula set accordingly.

[0054] like Figure 5The diagram illustrates the HQ-LTL syntax and its three-term extension structure. HQ-LTL is based on Past-Future Linear Temporal Logic (PF-LTL), retaining Boolean operators, future tense operators X (Next, must occur in the next state / moment), F (Future, must occur at some future moment), G (Globally, must be true from the current moment until forever), U (Until, a property remains true until another property is true), and past tense operators Y (Yesterday, must have occurred in the previous state / moment), P (Past, occurred at some past moment), H (Historically, true from the initial moment until the present), and S (Since, true from some past moment until the present). It adds three extensions: hierarchical typed quantization operators and their scope mechanism, grid topology relation predicates, and bounded metric time operators. Atomic predicates and numerical predicates constitute the basic decidable elements in the formula.

[0055] The power grid model M used to explain the HQ-LTL formula is represented as: M = (O, τ, R, H) Where O is a set of limited power business objects; τ: O → T is an object type function used to map objects in the limited power business object set to the corresponding power business object type T; T includes at least Breaker, Disconnector, Line, Bus, Power Transformer, Distribution Transformer, Work Order, and Dispatch Command; R is a set of power grid topology relationships; H is a four-level hierarchy of the power grid, used to characterize the hierarchical position and hierarchical relationship of limited power business objects.

[0056] The syntax for the HQ-LTL formula φ is as follows: φ ::= q(o1, ..., o) n / / State predicate | val(o) c / / Numerical predicate | topo(o1, o2) / / Topological relation predicate | ¬φ | φ1∧φ2| φ1∨φ2| φ1→φ2 / / Boolean operator | Xφ | Fφ | Gφ | φ1Uφ2 / / Future temporal operators | Yφ | Pφ | Hφ | φ1Sφ2 / / Past tense operator | F ≤k φ | G ≤k φ | P ≤k φ / / Bounded time metric operator | o:T[σ].φ | o: T[σ].φ / / Hierarchical typification quantization Where q(o1, ..., o) n () is used to represent observable atomic states or atomic events in a power proprietary service system, o1, ..., o n This represents n power service objects. Status predicates can include equipment status, authorized status, verification action, cross-system coordination, and operational action predicates. Equipment status includes closed, open, grounded, energized, and de-energized; authorized status includes work order issuance, dispatch instruction issuance, and operation permit granting; verification actions include voltage verification and checking; cross-system coordination status is used to characterize whether the object statuses in multiple service systems are consistent; operational actions include closing, opening, and updating actions.

[0057] val(o) c is used to compare the value of a physical quantity of object o with a constant c. The physical quantities include voltage, current, active power, and reactive power. The comparison sign satisfies the following conditions: ∈ {<, ≤, =, ≥, >}.

[0058] topo(o1, o2) includes at least same_circuit(o1, o2), upstream(o1, o2), parallel_path(o1, o2), and belongs_to(o1, o2), which respectively express the same-circuit relationship, upstream-downstream relationship, parallel path relationship, and hierarchical affiliation relationship between o1 and o2. o1 and o2 represent two power business objects (i.e., specific equipment in the power grid, such as circuit breakers, lines, transformers, etc.), and φ1 and φ2 represent two HQ-LTL sub-formulas.

[0059] X, F, G, and U are used to express the relationships of next step, future time, always, and until, respectively; Y, P, H, and S are used to express the relationships of previous step, past time, always past, and since, respectively. ≤k G ≤k P ≤kUsed to express temporal constraints within k time units in the future or past. The time unit can be set as a time step according to the sampling method of the state trace, or it can correspond to seconds, minutes or other actual time scales.

[0060] o:T[σ].φ and o: T[σ].φ is used to perform full quantization or existence quantization on power business objects of type T that are located within the scope σ. The scope σ can be a hierarchical scope of the form Level(h), or a topological scope of the form circuit(x), substation(s), or parallel(x), or it can be a combination of multiple scopes through intersection or union. Among them, circuit(x) is used to limit the range of power business objects within the circuit where object x is located, substation(s) is used to limit the range of power business objects within the substation s, and parallel(x) is used to limit the range of power business objects that are in parallel path relationship with object x. Through the joint restriction of quantization type and scope, the procedure constraints can be limited to a finite set of objects directly related to the current task.

[0061] Define π = (s0, s1, ..., s...) m Let s be the state trace, where s0 is the initial state before the job starts, s1, ..., s2 are the states. m-1 This is the intermediate state of the operation, s m For the final state after the task is completed, and t is the time step, the key extended semantic representation of the HQ-LTL formula φ on (π, t, M) is as follows: (π, t, M) q(o1, ..., o n ) q(o1, ..., o n ) ∈ s t (π, t, M) val(o) c s t .val(o) c (π, t, M) topo(o1, o2) (o1, o2) ∈ R topo (π, t, M) o : T[σ]. φ o ∈ O ∩ τ 1 (T) ∩ σ : (π, t, M) φ[o] (π, t, M) o : T[σ]. φ o ∈ O ∩ τ 1 (T) ∩ σ : (π, t, M) φ[o] (π, t, M) F ≤k φ t' ∈ [t, t+k]: (π, t', M) φ (π, t, M) G ≤k φ t' ∈ [t, t+k]: (π, t', M) φ (π, t, M) P ≤k φ t' ∈ [max(0, t k), t] : (π, t', M) φ The semantics of the remaining operators follow the corresponding semantics of PF-LTL. Under the conditions of a finite set of power business objects, a finite state trace, and a finite upper limit of measurement time, the number of object instances contained in the quantization scope is finite, and the HQ-LTL formula can be reduced to a finite-scale propositional PF-LTL formula.

[0062] π = (s0, s1, ..., s m The sequence is generated from a cross-system candidate sub-operation sequence. The procedure verification agent obtains the power grid state and candidate sub-operation sequence before execution, simulates the changes in equipment state, service state and measurement value according to the execution order of each candidate sub-operation, records the initial state before the execution of the candidate sequence as s0, and records the state after the execution of subsequent sub-operations into the state trace in sequence.

[0063] like Figure 6 The diagram shows the five-step reduction flowchart of the HQ-LTL quantization-based verification algorithm. After the cross-system candidate sub-operation sequence is generated by state trace simulation, quantization-based transformation, topological relation predicate evaluation, numerical predicate evaluation, bounded metric temporal operator expansion, and PF-LTL bounded model detection are performed sequentially.

[0064] When T is used as the type of power business object limited by the hierarchical typification quantization operator, and σ is used as the quantization scope, a finite set of power business objects O ∩ τ is first determined in the power grid model M. 1 (T) ∩ σ, and recursively perform the following substitution on the quantized sub-formula in the formula: o : T[σ]. ψ(o) o ∈ O ∩ τ^{ 1}(T) ∩ σ ψ[o' / o] o : T[σ]. ψ(o) o ∈ O ∩ τ^{ 1}(T) ∩ σ ψ[o' / o] Where ψ(o) is a rule with placeholders, and ψ[o' / o] is the specific verification condition after replacing the placeholders with specific objects. Since O is finite and each scope σ can be deterministically computed on the power grid model M, universal quantization can be expanded into the conjunction of the sub-formulas corresponding to the corresponding power business object instances, and existential quantization can be expanded into the disjunction of the sub-formulas corresponding to the corresponding power business object instances. After quantization and basicization, a propositional formula φ′ of finite scale is formed.

[0065] After quantization and basising are completed, the topology predicates such as `same_circuit`, `upstream`, `parallel_path`, and `belongs_to` are evaluated based on the set of topology relationships `R` in the power grid model `M`. Here, `same_circuit` represents the same-circuit relationship between power business objects, `upstream` represents the upstream-downstream relationship between power business objects, `parallel_path` represents the parallel path relationship between power business objects, and `belongs_to` represents the hierarchical affiliation relationship between power business objects. The truth values ​​of the topology predicates are determined by the object relationships in the power grid model and are independent of the time steps in the state trace; therefore, they can be evaluated all at once and replaced with Boolean constants.

[0066] val(o) When c is a numerical predicate, the procedure verification agent reads the measured values ​​of the corresponding objects at each time step in the state trace, compares the measured values ​​with the constant c, and replaces the comparison result with the Boolean atomic propositions at the corresponding time step. Thus, continuous numerical constraints such as voltage, current, active power, or reactive power are converted into true and false state sequences that can participate in proposition-level model detection.

[0067] The bounded time operator is expanded as follows: F ≤k φ i=0..k X i φ G ≤k φ i=0..k X i φ P ≤k φ i=0..k Y i φ Among them, X i Let φ′′ represent the application of the future tense operator X i times consecutively, and let φ′′ be the propositional PF-LTL formula obtained after the above expansion. The procedure verification agent performs bounded model detection based on φ′′ and the state trace π. As an optional implementation, φ′′ can be compiled into the product of the future tense part of a Büchi automaton (a finite state automaton for handling infinitely long state sequences) and the past tense part of an inverse automaton, and synchronous product determination can be performed on the state trace π.

[0068] When |O| = N, the quantization nesting depth is d, the bounded metric time limit is K, the state trace length is m, and the original formula size is |φ|, the formula size after quantization-based is: O(|φ| · N d ) The formula size after temporal expansion is: O(|φ| · N d · K) The cost of bounded model detection on the state trace is: O(m · 2 |φ′′| ) Where, N d Let d be the total number of power business objects N in the power grid model raised to the power of d.

[0069] In real-world power grids, the scope σ of the topology significantly limits the scale of quantization. For example, the number of all disconnect switches in a given loop is typically on the order of O(10), allowing actual verification to be completed within an acceptable timeframe.

[0070] If the formal verification passes through bounded model testing, the cross-system candidate sub-operation sequence that satisfies the applicable business procedure is determined as an executable sub-operation sequence; if the formal verification fails, the procedure verification agent generates a structured counterexample (CE). CE = (t , φk (Instance, SubFormula) Among them, t φ is the earliest time step in which the procedure is violated on the state trace. k The violated procedure formula is represented by `Instance`, which is the specific object instance selected during quantization that caused the violation, and `SubFormula` is the specific subformula that was violated. By preserving the correspondence between time steps, procedure formulas, object instances, and subformulas, the specific actions that caused the procedure violation in the candidate sequence and their unmet preconditions can be identified, rather than simply returning a Boolean conclusion of verification failure.

[0071] t and φ k Structured counterexample information is transcribed into replanning hints by the coordinating agent. For example, a replanning hint might indicate at time step t... A certain closing operation was performed, but the procedure φ k The operation requires that a scheduling instruction has been obtained and a work ticket has been issued prior to the operation, but the corresponding preconditions in the current state trace are not met. The coordinating agent invokes the Guangming Power large model, generates a corrected cross-system candidate sub-operation sequence based on the replanning prompts, and resubmits the corrected sequence to the procedure verification agent. The above process is repeated until the formal verification passes or the preset maximum number of iterations is reached. A counterexample guides replanning, correcting the sub-operation sequence.

[0072] For example, before performing maintenance on line l, all disconnect switches in the circuit containing line l must be in the open position, and the corresponding formula is: G(q opMaintain (l) → d1: Disconnector[circuit(l)]. q opened (d1)) Where Disconnector is an isolating switch, q opened (d1) indicates that the disconnecting switch d1 is in the open position, q opMaintain (l) indicates that line l is undergoing (or will undergo) maintenance operations. circuit(l) limits the quantization scope to the circuit where line l is located. The hierarchical categorized quantization operator performs full quantization of the disconnect switches within the scope, eliminating the need to manually enumerate all disconnect switches in the entire power grid.

[0073] For example, any two circuit breakers in a parallel path cannot be in the closed position simultaneously; the corresponding formula is: G( b1, b2: Breaker. parallel_path(b1, b2) → ¬(q closed(b1) ∧ q closed (b2))) Where Breaker is a circuit breaker, b1 and b2 are two specific circuit breaker instances, and q closed The state predicate for the device being in the closed position, and parallel_path as a power grid topology relation predicate directly participates in the formula evaluation, in order to avoid the disallowed loop state in the parallel path.

[0074] After the voltage test is completed, the grounding wire must be connected within 30 seconds; otherwise, the voltage test must be repeated. The corresponding formula is: G(q verified (l) → F ≤30s (q grounded (l) ∨ ¬q verified (l))) Among them, F ≤30s A 30-second actual time boundary is reserved to verify whether a grounding operation or voltage detection failure occurs within the specified time. verified (l) indicates that line l has completed voltage testing; q grounded (l) indicates that line l has been grounded or has been connected to a grounding wire.

[0075] When the voltage of busbar b is greater than 1.05 or less than 0.95, corresponding adjustment measures need to be initiated within 5 seconds. The corresponding formula is: G((val(voltage(b))>1.05 ∨ val(voltage(b))<0.95) → F ≤5s q adjusted (b)) Where voltage(b) represents the voltage physical quantity corresponding to power business object b, and val is used to obtain the numerical value of the corresponding physical quantity. Therefore, val(voltage(b)) represents the voltage measurement value of power business object b. This formula evaluates the numerical predicate based on the voltage measurement value at each time step of the state trace, and uses a bounded metric time operator to verify whether the adjustment action occurs within the specified time range.

[0076] Before performing any closing operation within substation s, all relevant work permits for that substation must have been issued. The corresponding formula is: G( b: Breaker[substation(s)]. q opClose (b) → P( t: WorkOrder[substation(s)]. q ticket (t))) Where Breaker is the circuit breaker, substation(s) indicates the area within substation s, WorkOrder is the work order, and q opClose (b) indicates that a closing operation is performed on circuit breaker b, q ticket (t) indicates that the work order has been issued and is in effect. substation(s) defines the scope of the circuit breaker and the work order, and the past temporal operator P is used to check whether the corresponding work order issuance status has occurred before the closing operation. The above formula enables HQ-LTL to express object set quantization, grid topology relationships, numerical comparisons, past and future temporal relationships, and actual time boundaries in the same formal language.

[0077] After the procedure verification is passed, the coordinating agent executes the cross-system linkage operation sequentially according to the executable sub-operation sequence. The coordinating agent retrieves the current sub-operation from the executable sub-operation sequence, selects the corresponding system-specific agent according to the target power proprietary business system specified by the sub-operation, and assigns the sub-operation represented by the cross-system unified object identifier to the system-specific agent.

[0078] like Figure 7 The diagram shows the sequence of multi-agent collaborative execution and semantic alignment mapping. The coordinating agent first assigns sub-operation op1 (with a unified object identifier obj) to system-specific agent A. System-specific agent A initiates an object mapping request to the semantic alignment agent. The semantic alignment agent queries the cross-system semantic alignment graph and maps the cross-system unified object identifier in the sub-operation to the local object identifier in system-specific agent A. System-specific agent A constructs the operation request based on a pre-defined data dictionary and interface specifications, executes the read or write operation in its assigned power-specific business system, and returns the execution results, including whether the operation was successful, the returned data, and any exception information.

[0079] The object identifier in the execution result is the local object identifier of the system-specific intelligent agent A. The coordinating agent queries the cross-system semantic alignment graph through the semantic alignment agent, reverse maps the local object identifier to the cross-system unified object identifier, and updates the execution result after reverse mapping to the unified business object view, so as to record the state changes of the same power business object in different power proprietary business systems in a unified manner.

[0080] The coordinating agent determines whether the execution result of the current sub-operation meets expectations based on the updated unified business object view. When the execution result meets expectations, the next sub-operation is retrieved from the sequence of executable sub-operations. When the next sub-operation is executed by the system-specific agent B, the mapping from the cross-system unified object identifier to the local object identifier of system B, the local system execution, and the reverse mapping of the execution result are continued.

[0081] When the execution result of the current sub-operation is abnormal, the coordinating agent calls the Guangming Power large model to evaluate the impact of the abnormality on subsequent sub-operations. Abnormalities may include system interface call errors, discrepancies between the target object's state and the expected state, missing returned data, or the failure of preceding operations to produce the expected state change. Based on the evaluation result, the coordinating agent chooses to retry the current sub-operation, skip the current sub-operation and adjust subsequent sub-operations, or terminate the current execution and report the abnormality. If adjustments to subsequent sub-operations are required, the subsequent sequence can be regenerated based on the current unified business object view, and it can be determined whether formal verification needs to be reimplemented according to the applicable business procedures.

[0082] Figure 7 The above loop is presented in the order of unified identifier assignment, local identifier mapping, local system execution, result reverse mapping, view update, and next step decision. The semantic alignment agent undertakes bidirectional identifier mapping in each cross-system interaction, enabling the execution results from different power proprietary business systems to establish a correspondence with the same power business object in the unified business object view.

[0083] After all sub-operations in the executable sub-operation sequence have been executed, the coordinating agent summarizes the state changes in the unified business object view and generates an execution result report for the cross-system collaborative operation. The execution result report includes the execution status of each sub-operation, details of changes to the involved power business objects, and time consumption statistics, and can be submitted to the task initiator or upper-level business applications. This forms a complete autonomous collaborative process from natural language task parsing, object semantic unification, formal verification, negative example-guided replanning, cross-system execution, to result summarization.

[0084] In one application example of this application, the method is deployed in the internal network environment of a provincial power grid company. The operating support environment provides basic functions such as agent instantiation, intranet communication, management of power-specific business system interface calls, and on-demand loading of capability resources. In this operating environment, a coordination agent, multiple system-specific agents, semantic alignment agents, and procedure verification agents are instantiated, and the agents collaboratively call the Guangming Power Big Model.

[0085] This application example uses a formalized knowledge base of power business procedures pre-built by power professionals. The procedures in this knowledge base are sourced from the "Power System Dispatching Procedures," "Power Safety Work Procedures," "Substation Switching Operation Guidelines," and "Distribution Network Operation Procedures." This application example also uses power grid model instance data based on CIM-E or CIM-G standards to form a standardized ontology diagram, power grid topology, and a four-level anchor point structure of stations, lines, transformers, and distribution stations.

[0086] Taking a 10kV distribution line load transfer operation as an example, the business personnel input a natural language task: "Transfer the load of a certain 10kV line A1 to line B1, working hours are from 9:00 to 12:00 today." The coordinating agent parses the task, identifies the task intent as load transfer, and the target power business objects as 10kV lines A1 and B1. It then retrieves the relevant clauses in the "Substation Switching Operation Guidelines" and the "Power System Dispatch Regulations" related to load transfer. The semantic alignment agent queries the cross-system semantic alignment graph to obtain the cross-system unified object identifiers for lines A1 and B1, and associates the lines, circuit breakers, disconnectors, and related work objects in each power proprietary business system based on these unified object identifiers.

[0087] The coordinating agent generates a cross-system candidate sub-operation sequence based on the load transfer task. The procedure verification agent obtains the initial equipment state and measurement data related to the candidate sequence, simulates the state changes after the execution of the candidate sub-operations, and generates a power grid state trace. In this application example, before opening the disconnecting switch, all circuit breakers in the circuit containing the disconnecting switch must be in the detached position. The corresponding procedure formula is: G( d: Disconnector. q opAction_open (d) → b : Breaker[circuit(d)]. Yq opened (b)) Here, circuit(d) is used to define the circuit breaker objects in the circuit where the disconnecting switch d is located. The procedure verification agent determines the disconnecting switch and circuit breaker instances in the formula based on the topological relationship of the circuits where line A1 and line B1 are located, expands the universal quantization into the conjunction of the sub-formulas corresponding to each specific equipment instance, and performs bounded model detection on the candidate sub-operation sequences.

[0088] t When scheduling the opening time step for disconnector DS-101 in the candidate sequence, if the circuit breaker CB-201 in its circuit is still closed, the model detection result is a failure, and a structured counterexample is generated. This counterexample is recorded at time step t. Examples of violated procedural formulas, specific objects such as disconnector DS-101 and circuit breaker CB-201, and unsatisfied past tense sub-formulas.

[0089] The coordinating agent transcribes structured counterexamples into reprogramming hints, prompting the Guangming Power large-scale model to supplement the necessary operation of putting circuit breaker CB-201 in the detached position before opening disconnector DS-101. Based on this, the Guangming Power large-scale model generates a revised cross-system candidate sub-operation sequence, such as inserting the sub-operation of disconnecting CB-201 before opening DS-101. The revised sequence is then submitted to the procedure verification agent again until all applicable procedures pass formal verification.

[0090] After formal verification passes, the coordinating agent sequentially assigns sub-operations to the corresponding system-specific agents. The semantic alignment agent converts the cross-system unified object identifiers of objects such as line A1, line B1, disconnector DS-101, and circuit breaker CB-201 into local object identifiers in the target system. The system-specific agent executes the sub-operation in the corresponding power proprietary business system and returns the execution result. The semantic alignment agent performs a reverse mapping of the local object identifiers in the execution result, and the coordinating agent continues to execute the next sub-operation based on the updated unified business object view. After all sub-operations are completed, the coordinating agent generates an execution result report for the load transfer operation, recording the execution status of each sub-operation, the status changes of line A1 and line B1, the change details of related circuit breakers and disconnectors, and the time consumption statistics of each operation.

[0091] The above application examples can also be used in business scenarios such as power transmission, substation, distribution, and dispatching, and can be extended to collaborative operations between multiple power-specific business systems such as dispatching, operation and maintenance, and marketing. Depending on actual deployment needs, the method in this application can be used for the daily operation and emergency response of tens of thousands of power business objects such as lines, transformers, and circuit breakers.

[0092] This application also proposes a multi-agent autonomous collaborative decision-making system for a power multi-service system, comprising a multi-agent collaborative architecture consisting of a coordinating agent, several system-specific agents corresponding to different power-specific business systems, a semantic alignment agent, and a procedure verification agent. The coordinating agent, system-specific agents, semantic alignment agents, and procedure verification agents operate according to the aforementioned multi-agent autonomous collaborative decision-making method for a power multi-service system.

[0093] This application also provides an electronic device, which may include one or more processors, memory and communication interfaces.

[0094] The memory, communication interface, and processor are coupled together. For example, the memory, communication interface, and processor can be coupled together via a bus.

[0095] The communication interface is used for data transmission with other devices. The memory stores computer program code. This computer program code includes computer instructions, which, when executed by the processor, cause the electronic device to perform the steps of the multi-agent autonomous collaborative decision-making method described above for the power multi-service system.

[0096] The processor can be a central processing unit (CPU), a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. The processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The processor can be used to support an electronic device in performing the method steps provided in the above embodiments.

[0097] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. These buses can be categorized as address buses, data buses, control buses, etc.

[0098] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the multi-agent autonomous collaborative decision-making method for the above-mentioned power multi-service system.

[0099] The computer-readable storage media involved in this application include random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage media known in the art.

[0100] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A multi-agent autonomous collaborative decision-making method for a power multi-service system, characterized in that, Based on a multi-agent collaborative architecture including a coordinating agent, several system-specific agents corresponding to different power-specific business systems, a semantic alignment agent, and a procedure verification agent; including: By coordinating the intelligent agent to receive natural language descriptions of power business tasks, identifying the task intent and target power business objects of the power business tasks, retrieving applicable business procedures, and generating cross-system candidate sub-operation sequences; Based on a pre-built cross-system semantic alignment graph, the semantic alignment agent determines the cross-system unified object identifier corresponding to the target power business object, and represents the power business object in the cross-system candidate sub-operation sequence based on the cross-system unified object identifier. The procedure verification agent obtains the hierarchical quantized linear temporal logic formula corresponding to the applicable business procedure from the pre-built formal knowledge base of power business procedures. It generates the power grid state trace based on the cross-system candidate sub-operation sequence, reduces the hierarchical quantized linear temporal logic formula to the propositional level linear temporal logic formula with past tense, and performs bounded model detection to obtain the formal verification result. Based on the formal verification results, the final cross-system candidate sub-operation sequence is determined as the executable sub-operation sequence; the coordinating agent assigns sub-operations to the corresponding system-specific agents according to the executable sub-operation sequence. Based on the cross-system semantic alignment graph, the semantic alignment agent maps the cross-system unified object identifier in the sub-operation to the local object identifier in the corresponding power proprietary business system. The corresponding system-specific agent then executes the sub-operation in the corresponding power proprietary business system based on the local object identifier and returns the execution result. The semantic alignment agent reverse-maps the local object identifier in the execution result to a unified object identifier across the system. The coordination agent determines the next sub-operation in the executable sub-operation sequence based on the execution result after the reverse mapping, until the executable sub-operation sequence is completed.

2. The multi-agent autonomous collaborative decision-making method for a power multi-service system according to claim 1, characterized in that, The coordinating agent, system-specific agent, semantic alignment agent, and procedure verification agent all use the Guangming Power big model as the underlying inference engine. Each of the system-specific intelligent agents is pre-configured with the data dictionary and interface specifications of its corresponding power proprietary business system, which are used to perform read or write operations in the corresponding power proprietary business system.

3. The multi-agent autonomous collaborative decision-making method for a power multi-service system according to claim 2, characterized in that, The pre-constructed cross-system semantic alignment graph is obtained through the following methods: Analyze the data dictionary of each power proprietary business system and construct the heterogeneous attribute graph corresponding to each power proprietary business system; Construct a specification ontology diagram based on the CIM-E or CIM-G standards; Based on the substation level, transmission and distribution line level, main transformer level, and distribution transformer and distribution area level in the power grid physical topology, a four-level anchor point structure of station, line, transformer and distribution area is constructed. Based on the four-level hierarchical anchor point structure, the power business object nodes in the heterogeneous attribute graph and the candidate nodes in the standard ontology graph are hierarchically anchored and aligned. For each node in the specified ontology graph, collect the power business objects aligned to that node from each power proprietary business system, generate a cross-system unified object identifier for that node, and persist the cross-system unified object identifier and its correspondence with each local object identifier to the cross-system semantic alignment graph.

4. The multi-agent autonomous collaborative decision-making method for a power multi-service system according to claim 3, characterized in that, The power grid model corresponding to the hierarchical quantization linear temporal logic formula includes a limited set of power business objects, object type functions, a set of power grid topology relationships, and a four-level hierarchical structure of the power grid. The object type function is used to map each power business object in the limited power business object set to the corresponding power business object type. The set of power grid topology relationships is used to characterize the power grid topology relationships between power business objects in a limited set of power business objects. The four-level hierarchical structure of the power grid is used to characterize the hierarchical position and hierarchical affiliation of power business objects in the limited set of power business objects within the four-level anchor point structure of station, line, transformer, and distribution station.

5. The multi-agent autonomous collaborative decision-making method for a power multi-service system according to claim 4, characterized in that, The hierarchical quantization linear temporal logic formula is based on linear temporal logic with past tense, including state predicates, numerical predicates, power grid topology relation predicates, Boolean operators, future temporal operators, past temporal operators, bounded metric time operators, and hierarchical typified quantization operators; The hierarchical quantization linear temporal logic formula is reduced to a propositional linear temporal logic formula with past tense, and bounded model testing is performed, including: Based on the power business object type and quantization scope defined by the hierarchical typified quantization operator, the hierarchical quantization linear temporal logic formula is quantized and baseified, the universal quantization is expanded into the conjunction of the corresponding power business object instance, and the existential quantization is expanded into the disjunction of the corresponding power business object instance. The power grid topology relation predicates in the quantized and basicized hierarchical quantized linear temporal logic formula are evaluated based on the power grid model, and the evaluation results are replaced with Boolean constants. The numerical predicates are evaluated based on the measured values ​​at each time step in the power grid state trace, and the evaluation results are replaced with Boolean atomic propositions. The bounded time operator is expanded into a propositional formula consisting of future or past tense operators, resulting in the propositional linear temporal logic formula with past tense. Based on the propositional linear temporal logic formula with past tense and the power grid state trace, a bounded model test is performed to obtain the formal verification result.

6. The multi-agent autonomous collaborative decision-making method for a power multi-service system according to claim 5, characterized in that, Based on the formal verification results, the final cross-system candidate sub-operation sequence is determined as the executable sub-operation sequence, including: When the formal verification result is not passed, a structured counterexample is generated by the procedure verification agent; the structured counterexample includes the earliest time step in which the applicable business procedure is violated on the power grid state trace, the violated hierarchical quantized linear temporal logic formula, the power business object instance that caused the violation determined during quantization, and the violated sub-formula. The structured counterexamples are transcribed into replanning hints by a coordinating agent, and the Guangming Power big model is invoked to generate a corrected cross-system candidate sub-operation sequence based on the replanning hints. The corrected cross-system candidate sub-operation sequence is resubmitted to the procedure verification agent for formal verification until the formal verification is passed or the preset maximum number of iterations is reached. The cross-system candidate sub-operation sequences that pass formal verification are determined as the executable sub-operation sequences.

7. The multi-agent autonomous collaborative decision-making method for a power multi-service system according to claim 4, characterized in that, The next sub-operation in the sequence of executable sub-operations is determined by the coordinating agent based on the execution result after reverse mapping, including: When the execution result of the current sub-operation meets expectations, the next sub-operation is taken from the sequence of executable sub-operations; when the execution result of the current sub-operation is abnormal, the Guangming Power big model is called to evaluate the impact of the abnormality on subsequent sub-operations, and the current sub-operation is retried, the current sub-operation is skipped and subsequent sub-operations are adjusted, or the execution is terminated and the abnormality is reported based on the evaluation result. After all sub-operations in the executable sub-operation sequence have been executed, the following steps are also included: The system aggregates status changes in the unified business object view and generates an execution result report for cross-system collaborative operations. The execution result report includes the execution status of each sub-operation, details of changes to the power business objects involved, and time consumption statistics.

8. A multi-agent autonomous collaborative decision-making system for a power multi-service system, characterized in that, It includes a coordinating agent, several system-specific agents corresponding to different power proprietary business systems, a semantic alignment agent, and a procedure verification agent; The coordinating agent, system-specific agent, semantic alignment agent, and procedure verification agent operate in accordance with the multi-agent autonomous collaborative decision-making method for power multi-service systems as described in any one of claims 1 to 7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-agent autonomous collaborative decision-making method for a power multi-service system as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-agent autonomous collaborative decision-making method for the power multi-service system as described in any one of claims 1-7.

Images