A method for intelligently generating a device startup path based on dynamic analysis of microgrid topology fusion

Abstract

The present application relates to a kind of equipment starting path intelligent generation method of fusion microgrid topology dynamic analysis, to realize the fine path planning and intelligent scheduling of multiple equipment collaborative start task.For the problem of heterogeneous and complex dependence relationship of power grid operation path planning data, a three-dimensional data structure and four-dimensional association rule base are constructed based on primary equipment state data, logical node information and operation procedure record to generate power grid operation semantic atlas initialization base.Through the introduction of semantic enhancement, dynamic reasoning, dependence chain mining and compliance path screening, combined with time window and collaborative scheduling optimization, the compression and intelligent step of multiple equipment collaborative start path are realized.The scheme can dynamically adapt evolution and form a complete dependence propagation evidence chain, improve the safety, compliance and execution efficiency of scheduling.

Description

Technical Field

[0001] This invention relates to the field of power system automation and intelligent decision-making technology, and in particular to an intelligent method for generating device start-up paths that integrates microgrid topology dynamic analysis. Background Technology

[0002] Currently, in the field of power system automation and intelligent decision-making, the optimization of collaborative startup paths for multiple devices in the power grid, especially in microgrid scenarios, commonly employs path planning methods driven by topology analysis, static dependency tables, and manual operating procedures. Mainstream solutions rely primarily on device connection matrices, event-driven finite state machines, fixed dependency rule tables, and static electrical constraint models. By pre-setting the operation sequence between devices and basic safety interlock conditions, startup / shutdown steps are deduced according to the process business logic. In practical applications, these methods typically manifest in two modes: one directly sets the operation sequence for a finite set of devices using expert databases and human experience; the other combines SCADA / EMS systems to automatically determine device interlocking relationships and energy flow feasibility, thereby generating a step-by-step instruction set that meets basic safety procedures. In recent years, some research has also introduced process modeling, temporal automata, and graph theory methods to extend and optimize topology changes, device state constraints, and concurrent operation timing. The development trend is to attempt to automatically adapt to the needs of general scheduling processes, improving path generation speed and coverage.

[0003] The shortcomings of current technology are mainly reflected in the following aspects: Existing path planning tools fail to effectively and dynamically model the operational dependencies between multiple devices, remaining at the level of static manual rules and simple connection matrices.

[0004] When multiple devices need to be started simultaneously within the same time window, the system has difficulty automatically reasoning the logical relationship between the order of device operations, which can easily lead to security issues such as missing prerequisites, mutual exclusion logic conflicts, or lost procedure calls.

[0005] The lack of a unified modeling and closed-loop feedback mechanism for implicit dependencies, time-series scheduling, and security anchors in dynamic collaboration makes it difficult to support the actual scheduling operations' requirements for traceable, compliant, and efficient paths.

[0006] In scenarios where device topology or system status changes in real time, traditional methods cannot automatically evolve dependency logic based on real-time data, procedural requirements, and historical experience, resulting in insufficient flexibility and adaptability in path generation. Summary of the Invention

[0007] This application provides a method for intelligently generating device start-up paths that integrates dynamic analysis of microgrid topology, aiming to solve one of the problems or issues of the existing technology mentioned in the background.

[0008] This application provides a method for intelligently generating device start-up paths that integrates dynamic analysis of microgrid topology, specifically including: S1: Obtain the status data of primary equipment in the power grid, information of secondary logical nodes, and typical startup operation records. Based on the information, define a three-dimensional data structure that includes equipment entities, operation entities, and procedure entities. Construct a four-dimensional association rule base according to topology connection, electrical constraints, operation procedures, and historical collaboration patterns to generate the initialization basis of the power grid operation semantic graph.

[0009] S2: Based on the initialization base of the power grid operation semantic graph, perform semantic enhancement processing on the static connections between graph nodes, embed four types of semantic relationship edges: trigger dependency, mutual exclusion prohibition, timing enforcement, and procedure reference, and generate a power grid operation semantic graph with dynamic reasoning capabilities.

[0010] S3: Receive the set of multi-device collaborative startup tasks and initialize the target state of each device. Starting from all devices to be started, initiate parallel semantic path extension in the power grid operation semantic graph and generate an initial path candidate set containing candidate operation nodes and their prerequisites.

[0011] S4: For each candidate operation node in the initial path candidate set, traverse all its preconditions and retrieve the current system state satisfaction status in the power grid operation semantic graph. For unsatisfied preconditions, perform dynamic activation retrieval based on semantic relationship edges to generate implicit operation dependency chains.

[0012] S5: Based on the implicit operation dependency chain, calculate the semantic depth, procedure compliance score and historical collaboration frequency weight of each propagation path, eliminate operation combinations with incomplete propagation chains or triggering mutually exclusive prohibited semantic edges, and generate a compliant path candidate set.

[0013] S6: Map the compliant path candidate set to a time interval sequence and introduce the concept of collaborative window to jointly optimize the paths of each device and generate a multi-device collaborative startup path sequence containing compressed scheduling instruction steps.

[0014] S7: Based on the multi-device collaborative startup path sequence, force key operations to fall within the manual confirmation time window, and simultaneously extract the activated semantic relationship edge type, referenced procedure clauses and historical similar collaborative case numbers to generate the final startup path scheme with attached dependency propagation evidence chain.

[0015] S8: Output the final startup path scheme to the scheduling execution terminal, and update the weights of historical collaborative patterns in the power grid operation semantic graph based on actual execution feedback to complete the evolution loop.

[0016] This application provides a method for intelligently generating device startup paths that integrates dynamic analysis of microgrid topology, which has the following advantages: (1) To address the problems of equipment startup operations relying on manual experience, rigid rules that are difficult to adapt to topology changes, and low efficiency of multi-device collaboration in traditional power grids and microgrids, this solution constructs a power grid operation semantic graph that integrates three types of entities: equipment, operation, and procedures. This enables the formal expression and automated reasoning of dynamic dependencies in complex electrical systems. Compared to existing methods based on static dependency matrices or preset flowcharts, this solution models equipment status, operation actions, and scheduling procedures as a unified semantic network with constrained predicates. This not only supports the accurate characterization of strong temporal logic such as "phase verification must be performed before loop closure" and "load can only be carried after charging," but also achieves dynamic association identification between cross-device operations through semantic edges such as "trigger dependency," "mutual exclusion prohibition," and "temporal mandatory." During path planning, the system does not rely on a pre-configured operation sequence table. Instead, based on the current initial state of the power grid, it unfolds candidate operations layer by layer through a "semantic feasibility judgment" mechanism and uses a constrained multi-hop semantic propagation mechanism to trace whether the preconditions can be indirectly satisfied through other executable operations, thereby automatically discovering potential operation collaboration paths. This mechanism effectively overcomes the technical shortcomings of traditional methods, such as poor adaptability and slow response to replanning when faced with adjustments to non-standard wiring methods or temporary operating methods, and significantly improves the intelligence level and scenario generalization ability of startup scheme generation.

[0017] (2) To further improve the overall efficiency and safety controllability of the multi-device collaborative startup process, especially in the complex environment of frequent interaction between distributed power sources, energy storage and loads in microgrids, this scheme introduces an improved spatiotemporal constraint satisfaction optimization framework to achieve joint scheduling of operation sequences in the time dimension based on the semantic path candidate set. Unlike conventional graph search that only pursues the shortest path length, this scheme comprehensively considers the spatial distribution between operations, the difference in execution time, and the requirements of human-machine collaboration nodes, and proposes a "collaboration window" mechanism, which allows the parallel operations of multiple devices to be merged into the same instruction step under the premise of satisfying semantic dependencies, greatly reducing the frequency of scheduling instruction interaction and compressing the overall operation time to a response range of hundreds of milliseconds; at the same time, a "procedure anchor point" hard constraint is set to ensure that key risk operations (such as the first loop closure and main transformer impact) must fall within the manual confirmation time window, ensuring the effective integration of automated decision-making and human supervision. In addition, the path output not only includes the operation sequence of each device, but also generates a complete dependency propagation evidence chain, covering the activated semantic edge type, the referenced scheduling procedure clause number and historical similar case reference, enhancing the interpretability and audit traceability of the recommendation results. This design enables the system to have both efficient autonomous planning capabilities and compliance with the power system's "strict risk control and clear responsibility" operation and management standards, significantly outperforming traditional black-box recommendation models in terms of security and reliability.

[0018] In summary, this solution, by constructing a structured and reasonable semantic graph of power grid operation and combining dynamic dependency propagation with a spatiotemporal joint optimization mechanism, achieves a paradigm shift from "rule-driven" to "semantic-driven" approaches, overcoming multiple bottlenecks in flexibility, security, and collaborative efficiency inherent in traditional methods. The system can adaptively generate multi-device collaborative startup paths that conform to scheduling procedures and possess complete logical basis, without relying on manually written operation ticket templates or fixed device dependency tables. It exhibits excellent engineering practicality and promotional value, and is particularly suitable for complex scenarios involving frequent changes in operating modes, such as microgrids, substation upgrades and expansions, and renewable energy integration. This provides solid technical support for the closed-loop adaptive evolution of intelligent scheduling systems. Attached Figure Description

[0019] Figure 1 This is the main flowchart of a method for intelligently generating device startup paths that integrates dynamic analysis of microgrid topology.

[0020] Figure 2 This is a sub-flowchart of a method for intelligently generating device startup paths that integrates dynamic analysis of microgrid topology.

[0021] Figure 3 This is another sub-flowchart of a method for intelligently generating device start-up paths that integrates dynamic analysis of microgrid topology.

[0022] Figure 4 This is an application environment diagram of a device startup path intelligent generation method that integrates microgrid topology dynamic analysis in one embodiment.

[0023] Figure 5 An internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0024] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0025] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0026] like Figure 1As shown, this application provides an intelligent generation method for device start-up paths that integrates microgrid topology dynamic analysis, specifically including: S1: Obtain the status data of primary equipment in the power grid, information of secondary logical nodes, and typical startup operation records. Based on the information, define a three-dimensional data structure that includes equipment entities, operation entities, and procedure entities. Construct a four-dimensional association rule base according to topology connection, electrical constraints, operation procedures, and historical collaboration patterns to generate the initialization basis of the power grid operation semantic graph.

[0027] S2: Based on the initialization base of the power grid operation semantic graph, perform semantic enhancement processing on the static connections between graph nodes, embed four types of semantic relationship edges: trigger dependency, mutual exclusion prohibition, timing enforcement, and procedure reference, and generate a power grid operation semantic graph with dynamic reasoning capabilities.

[0028] S3: Receive the set of multi-device collaborative startup tasks and initialize the target state of each device. Starting from all devices to be started, initiate parallel semantic path extension in the power grid operation semantic graph and generate an initial path candidate set containing candidate operation nodes and their prerequisites.

[0029] S4: For each candidate operation node in the initial path candidate set, traverse all its preconditions and retrieve the current system state satisfaction status in the power grid operation semantic graph. For unsatisfied preconditions, perform dynamic activation retrieval based on semantic relationship edges to generate implicit operation dependency chains.

[0030] S5: Based on the implicit operation dependency chain, calculate the semantic depth, procedure compliance score and historical collaboration frequency weight of each propagation path, eliminate operation combinations with incomplete propagation chains or triggering mutually exclusive prohibited semantic edges, and generate a compliant path candidate set.

[0031] S6: Map the compliant path candidate set to a time interval sequence and introduce the concept of collaborative window to jointly optimize the paths of each device and generate a multi-device collaborative startup path sequence containing compressed scheduling instruction steps.

[0032] S7: Based on the multi-device collaborative startup path sequence, force key operations to fall within the manual confirmation time window, and simultaneously extract the activated semantic relationship edge type, referenced procedure clauses and historical similar collaborative case numbers to generate the final startup path scheme with attached dependency propagation evidence chain.

[0033] S8: Output the final startup path scheme to the scheduling execution terminal, and update the weights of historical collaborative patterns in the power grid operation semantic graph based on actual execution feedback to complete the evolution loop.

[0034] Step S1: Obtain primary power grid equipment status data, secondary logic node information, and typical startup operation records. Based on this information, define a three-dimensional data structure including equipment entities, operation entities, and procedure entities. Then, construct a four-dimensional association rule base based on topology connections, electrical constraints, operation procedures, and historical collaborative patterns to generate the initialization basis for the power grid operation semantic graph. Specifically, this includes: S1.1: Obtain primary power grid equipment status data, secondary logical node information, and typical startup operation records. Clean and standardize the multi-source heterogeneous data to generate a structured raw dataset containing equipment operating parameters, logical node attributes, and operation procedure text.

[0035] The system acquires electrical operating parameters from the primary equipment monitoring system, status signals from the secondary logic node monitoring system, and typical startup operation texts from the scheduling record database. It then precisely associates and matches these three types of data according to timestamps and equipment identifiers to form a multi-source heterogeneous raw data set.

[0036] The sensor range is verified for the primary equipment operating parameters in the original dataset, and records that are out of range or have zero values ​​are removed. At the same time, the numerical units are uniformly processed according to the preset physical unit conversion table.

[0037] Perform signal consistency detection on the secondary logical node information, use the pattern matching mechanism of node state change sequence to identify abnormal state transitions and record them as quality labels to ensure the reliability of logical node attribute descriptions in subsequent modeling.

[0038] The typical startup operation record execution procedure text is parsed, and the natural language processing interface is called to extract the operation actions, preconditions and execution environment descriptions, and a preliminary index is built according to the operation procedure clause number.

[0039] The primary equipment parameters, secondary logical node attributes, and procedure text features are mapped under a unified field template. A standardized index structure is established based on equipment coding rules and topology location coding rules, and normalization is performed on all numerical fields.

[0040] By using a multi-domain data fusion mechanism, the equipment operating parameters, logical node attributes, and procedure texts that have undergone quality verification and standardization are merged into a single structured raw dataset, thereby achieving input consistency for subsequent predicate modeling.

[0041] Through the above cleaning and standardization processes, the results of the previous step are transformed into a structured raw dataset that can be directly used for formal predicate definition and association rule mining, thus achieving the integrity and consistency of the data input.

[0042] S1.2: Based on the structured original dataset, define device entity predicates, operation action predicates and constraint condition predicates using formal modeling methods, and perform instantiation mapping processing on the predicates to generate a set of three-dimensional data structure objects containing electrical attributes, topology ports, preconditions and safety risk levels.

[0043] Based on the equipment operating parameters, logical node attributes, and operation procedure text in the structured raw dataset, a formal modeling interface is called to perform hierarchical parsing of different types of data, and to establish a parameter mapping relationship paradigm for equipment entity predicate templates, operation action predicate templates, and constraint condition predicate templates.

[0044] For primary equipment operating parameters, physical parameters such as rated voltage, rated current, phase sequence identifier, and grounding status are bound to the attribute slots of the equipment entity predicate according to the electrical attribute classification rules. The topology identifier information such as the bus segment to which the equipment belongs and the connection switch number is loaded into the topology port field, forming a combination structure of physical attributes and topology attributes in the equipment entity.

[0045] For the attributes of secondary logic nodes, the logic state abstraction method is called to establish a correspondence between the node signal type (input, output, remote signaling, telemetry) and the operation action predicate that it controls or reflects, and fill in the execution condition field and post-state field of the action predicate, so that the operation action predicate has a complete condition and effect mapping.

[0046] For the operating procedure text, the procedure clause parser is used to extract information such as procedure number, execution conditions, prohibition conditions, and safety risk level, and maps it to the clause reference field and risk level field of the constraint predicate. The risk level is then numerically encoded so that the constraint predicate can be associated with the equipment entity and the operation action predicate through the procedure number.

[0047] After the predicate template is defined, instantiation mapping is performed. The actual values ​​of the corresponding devices, actions and clauses in the structured original dataset are assigned to the template fields one by one. The consistency of references and the completeness of conditions between each predicate are verified. Finally, a set of three-dimensional data structure objects containing electrical attributes, topology ports, preconditions and security risk levels are generated.

[0048] By instantiating formal modeling methods, the structured raw dataset from the previous step is transformed into equipment entities, operation action entities, and procedure entities with semantic identifiers and logical constraints, thereby initializing the three-dimensional data structure and providing a precise entity definition and condition constraint foundation for subsequent construction of the association rule base and generation of the power grid operation semantic graph.

[0049] For example, for primary equipment status data collected from a 500kV substation, the electrical attribute slot of the equipment entity predicate template is assigned the values ​​of rated voltage 500, rated current 3150, phase sequence identifier ABC, and grounding status non-grounded; the topology port field is loaded with bus segment number "I segment" and connection switch number "QF1". In the logical node attributes, the input signal type "QF1 closing allowed" is mapped to the precondition field of the operation action predicate "execute QF1 closing operation", and the postcondition field is assigned the value "QF1 is closed and there is voltage on both sides". In the procedure clause text, the clause number "R-201" is parsed as the constraint condition predicate clause reference field, the execution condition is "bus voltage is normal", the prohibition condition is "load exceeds rated value", and the safety risk level is coded as 3. During the instantiation mapping process, the above parameters are assigned to the corresponding predicate fields respectively, and the consistency check logic is called to ensure that the clause reference field in the constraint condition predicate is consistent with the procedure number in the precondition field of the operation action predicate. The final generated set of three-dimensional data structure objects includes equipment entities with complete electrical attributes and topology ports, operation action entities with preconditions and postconditions, and procedure entities with clause references and risk codes. This achieves a tight constraint association between different entities. Verification results show that this set can support dynamic reasoning and path planning in the subsequent association rule base.

[0050] S1.3: Based on the set of three-dimensional data structure objects, extract topological connection relationships, electrical constraints, operating procedure clauses and historical collaboration mode features, and perform association rule mining and logical encoding processing on the features to generate a four-dimensional association rule library covering static connections and dynamic constraints.

[0051] When receiving a set of 3D data structure objects as input, the feature type recognition module labels the device entities, operation entities, and procedure entities in the set with their respective category tags, and calls the topology parsing engine to read the topology port attributes of the device entities to form a preliminary connectivity matrix. Based on the pairing relationships of each port in the connectivity matrix, the topology connection relationship extraction is performed, calculating the physical contact form and signal path direction between each pair of adjacent devices, and outputting the results as a topology connection feature set for subsequent rule mining. For the electrical attribute fields of the operation entities, the electrical constraint calculation module is called to perform numerical normalization processing on the constraints involving voltage, current, and power, and match them to the corresponding patterns in the standard constraint template library to form an electrical constraint feature set. For the text clause information of the procedure entities, the procedure clause parser is executed to convert the clause content into a structured predicate expression and map it to the applicable scenario label to generate a procedure clause feature set. Using the above three feature sets as input, the historical collaboration pattern analysis module is called to perform frequency statistics and pattern sequence abstraction on the start-up sequences involving the same device combination in the 3D object set, and outputs a historical collaboration pattern feature set.

[0052] The association rule mining engine processes four types of feature sets in parallel. A multi-dimensional Apriori algorithm is used to generate static association rules between topological connectivity features and electrical constraint features, clarifying the necessity of equipment connectivity structures for satisfying electrical conditions. Dynamic association rules are generated between procedure clause features and historical collaboration pattern features, describing the conditions under which operation sequences conform to procedure clauses and their evolutionary trends in historical patterns. Logical encoding is performed on the mined rule sets, mapping rule antecedents and consequents to unique rule identifiers, and attribute tags are added according to rule type to mark their static or dynamic properties. The rule structure optimization module is invoked to analyze the support and confidence of each rule, eliminating rules below a preset threshold and merging logically redundant rule branches to form a high-quality rule set. The optimized rule set is stored according to four dimensions: topological connectivity, electrical constraints, procedure clauses, and historical collaboration patterns, and a rule index table is constructed to enable fast rule retrieval and retrieval.

[0053] By using association rule mining and logical encoding, the three-dimensional data structure object set from the previous step is transformed into a four-dimensional association rule library that covers static connections and dynamic constraints. This enables the semantic indexability and high quality of the rules, providing a precise rule mapping foundation for subsequent graph node initialization and relation preloading.

[0054] S1.4: Based on the three-dimensional data structure object set and the four-dimensional association rule base, perform graph node initialization and edge relationship preloading operations, and perform semantic alignment and index construction processing on the nodes and edges to generate a power grid operation semantic graph initialization base with complete entity definition and rule mapping capabilities.

[0055] Step S2: Based on the initialization base of the power grid operation semantic graph, semantic enhancement processing is performed on the static connections between graph nodes, embedding four types of semantic relationship edges: trigger dependency, mutual exclusion prohibition, timing enforcement, and procedure reference, to generate a power grid operation semantic graph with dynamic reasoning capabilities. Specifically, this includes: S2.1: Obtain the device entity pairs and static connection edges in the initialization base of the power grid operation semantic graph, calculate the electrical coupling strength and signal transmission delay parameters between adjacent devices based on electrical topology connectivity analysis, and generate a set of basic associated edges that include physical connection attribute enhancements to establish the physical boundary conditions for semantic enhancement.

[0056] The input objects are pairs of equipment entities and static connection edges in the initialization base of the power grid operation semantic graph. Physical relationship parameter extraction based on electrical topology connectivity analysis is performed. For each pair of equipment entities with static connection edges, its topology port definition and electrical attribute parameters are read, and the topology ports are mapped to the coordinates of the system topology matrix nodes, forming a set of equipment-related coordinates that can be used for connectivity analysis. Connectivity analysis is invoked to calculate path impedance on the related coordinate set, using a complex impedance model to decompose it into active and reactive components, and the electrical coupling strength coefficient is obtained from the decomposition results. The coupling strength is obtained using the following formula: Where R is the active impedance of the path, X is the reactive impedance of the path, and K is the electrical coupling strength coefficient. Based on the physical length of the communication link and the characteristics of the conductor medium, the signal transmission delay parameter is calculated using the following formula: Where L is the signal transmission path length and v is the medium signal transmission speed. K and T are used as physical connection enhancement attributes, bound to the corresponding static connection edge data structure, and the basic association edge set is updated. A consistency check is performed on the basic association edge set to ensure there are no isolated nodes or outlier parameter values. Qualified enhancement attribute edges are recorded as physical boundary conditions for subsequent semantic enhancement relation embedding.

[0057] By calculating electrical topology connectivity and measuring signal delay, the static connection edges from the previous step are transformed into a set of basic associated edges containing physical coupling coefficients and time delay indices, thereby enabling the construction of physical boundary conditions required for semantic enhancement processing.

[0058] For example, in a 220kV substation scenario, the equipment entity is the main transformer low-voltage side circuit breaker QF1 and bus section II. The static connection edge is mapped to the topology matrix node coordinates (12,34) and (12,36). The measured path impedance R is 0.05Ω and X is 0.08Ω. The value of K is calculated using the formula: Calculations yielded K≈10.416. The physical length L of the signal link was measured to be 15m, the medium transmission speed v was 200,000,000 m / s, and the delay T was calculated using the formula: The result is 75ns. K and T are bound to the static connection edge of the QF1-bus II segment, and the basic associated edge set is updated. After consistency verification, this enhanced attribute is marked as a valid physical boundary condition, significantly improving the accuracy of causal inference and the reliability of temporal simulation in subsequent processes of embedding dependent and temporally enforced relation edges.

[0059] S2.2: Based on the aforementioned basic set of associated edges and the pre-set operation procedure knowledge base, the predicate logic matching mechanism is used to identify the pre-state requirements and post-state results of the operation action, construct trigger dependency edges that describe how the preceding operation automatically satisfies the conditions of the following operation, and generate a dynamic triggering link with causal inference capability.

[0060] Based on the aforementioned set of basic associated edges and the pre-set operation procedure knowledge base, state predicate logic matching is performed on the input device entity pairs and their static connection attributes to identify the pre-state requirements and post-state results that can form trigger dependencies. Between the two device entities corresponding to each basic associated edge, all operation action predicates associated with that edge are retrieved from the operation entity library, and their pre-condition sets and execution result sets are parsed. The state predicate matching mechanism is invoked to compare the pre-condition sets with the execution result sets of the other device one by one, generating successfully matched condition mapping pairs. For each condition mapping pair, a causal chain node pair from the pre-order operation to the post-order operation is constructed, and a trigger dependency edge type identifier is assigned. During the generation of trigger dependency edges, the constraints of the corresponding clauses in the pre-set operation procedure knowledge base are combined to filter out causal chains that conform to the same procedure grouping and whose execution order is compliant, while chains whose execution results are inconsistent with the procedure clauses are eliminated. Using the node sequence information of the causal chain, a dynamic trigger link index table is established, and the starting operation node, ending operation node, trigger type, and procedure number of each link are structurally encapsulated. By constructing causal chains and verifying procedures, the static connections of the basic set of related edges are extended into trigger dependency data with dynamic causal deduction capabilities, thereby realizing the dynamic generation of trigger links in the semantic graph of power grid operation.

[0061] S2.3: Using the dynamic triggering link and power grid security constraint rule set, execute resource conflict detection and electrical state mutual exclusion verification logic, identify operations that cannot be executed concurrently within the same time window to generate mutually exclusive prohibition relationship edges, and form a logical isolation mark to ensure operational security.

[0062] Based on the dynamic triggering link and the power grid safety constraint rule set, the system receives dynamic triggering link data with causal deduction capabilities and the corresponding set of device state entities from the output of the preceding sub-steps. Resource conflict detection and electrical state mutual exclusion verification logic are executed to identify operations that cannot be executed concurrently within the same time window. The resource occupancy information corresponding to each operation node in the dynamic triggering link is mapped to a resource occupancy matrix, where matrix rows represent operation nodes and matrix columns represent key physical resources and logical control resources. Resource conflict detection is achieved through resource occupancy vector comparison. When any two operation chains are executed simultaneously, the electrical topology analysis interface is called to map the state predicates of the two operations to the electrical state diagram. Combinations with voltage level conflicts, short-circuit risks, or signal reversals are retrieved according to the electrical state mutual exclusion rule set, generating a state conflict list. A time window parameter is introduced into the conflict list, and the time overlap calculation module is called to determine whether the conflicting operations are in the same scheduling time segment. After concurrent conflict determination, the conflicting operation node pair is marked as a mutually exclusive prohibition edge, and the attribute field of this edge is updated in the graph to include the resource type and state conflict level that triggered the conflict. Logical isolation markers are generated for mutually exclusive prohibited edges, and these markers are stored in the form of structured metadata for use in subsequent path planning processes, thus achieving a preliminary guarantee of operational security. This processing method transforms the dynamically triggered links from the previous step into semantically related edge data with mutually exclusive prohibited markers, enabling accurate identification and logical isolation of interoperability conflicts.

[0063] For example, a power grid startup task includes two operations: "Bus Section I commissioning" and "Bus Section II loop unblocking." The dynamic triggering link has established its preceding dependency chain. The resource occupancy matrix sets rows corresponding to operation nodes and columns corresponding to resources such as bus sections, circuit breakers, and monitoring and control devices. A matrix value of 1 indicates occupancy. Commissioning Bus Section I occupies Bus Section I and circuit breaker QF1, while unblocking Bus Section II occupies Bus Section II and circuit breaker QF2. When calling the electrical state mutual exclusion rule set for detection, it is found that the two operations will cause voltage imbalance at the bus tie point under the current topology, and the state conflict level is set to high. The scheduling time window is set to 5 seconds, Δt is calculated as the start time difference of the operation plan execution in 3 seconds, and the ratio calculation result is less than the threshold of 1, indicating a concurrent conflict. The system generates mutually exclusive prohibition edges in the graph, with attribute fields including resource type "bus section" and state conflict level "high," and generates logical isolation markers. During the subsequent path generation process, the two operations of blocking the mutual exclusion edge are executed in conjunction within the same time window. Verification shows that the safety of the startup path is significantly improved, and no voltage imbalance event occurs during the overall startup process.

[0064] S2.4: Based on the time series data in the logical isolation marker and typical startup operation records, apply the time series constraint mining method to extract the forced order rules between operations, construct the time series forced relationship edges describing the strict execution order, and generate a time constraint network that limits the operation execution window.

[0065] S2.5: Integrate the dynamic triggering link, logical isolation marker and time constraint network, establish a mapping link between operation nodes and procedure entities according to the scheduling procedure clause indexing mechanism, embed procedure reference relationship edges and update the graph adjacency matrix, and finally generate a power grid operation semantic graph with complete four-dimensional semantic reasoning capabilities.

[0066] Based on the completed construction results of the dynamic triggering link, logical isolation marker and time constraint network, the procedure entity index table of the loading scheduling procedure clause index mechanism is used as the input for mapping rules.

[0067] For each operation node involved in the dynamic triggering chain, its operation action predicate and precondition set are parsed, the procedure entity index table is called to retrieve the matching clause number and applicable scenario tag, and a direct mapping relationship from operation node to procedure entity is generated.

[0068] For mutually exclusive prohibited edges in the logical isolation marker, perform cross-validation of the procedure clauses to determine whether there are concurrent prohibition clauses in the procedure text. The matching results are then appended to the semantic attribute field of the mutually exclusive edge to ensure full coverage of the procedure constraints at the graph semantic layer.

[0069] For the temporal forced edges of the time-constrained network, a bidirectional link between the temporal forced edges and the procedure entities is established using the time anchor clauses defined in the procedure entity index table, and a manual confirmation time window parameter is added to the edge attributes.

[0070] The above-mentioned operation node-procedure entity mapping and semantic relationship edge-procedure entity link are uniformly written into the graph adjacency matrix. The matrix update method is called to perform index rearrangement and weight normalization processing on the newly added relations. The type identifier and clause number of the procedure reference relationship edge are embedded in the adjacency matrix.

[0071] By embedding the procedure reference relationship, the trigger link, isolation marker and time constraint network results of the previous step are transformed into graph storage data with a triple structure of device action - procedure entity - semantic edge, so as to realize the final generation of four-dimensional semantic reasoning capability.

[0072] like Figure 2 As shown, step S3 involves receiving a set of multi-device collaborative startup tasks and initializing the target state of each device. Starting with all devices to be started, parallel semantic path extension is initiated in the power grid operation semantic graph, generating an initial path candidate set containing candidate operation nodes and their preconditions. Specifically, this includes: S3.1: Obtain the set of multi-device collaborative startup tasks and the current operating status data of each device. Based on the device entity definition in the power grid operation semantic graph, perform mapping and matching processing on the device identifiers in the task set and the graph nodes to generate a standardized startup task queue with graph index attributes.

[0073] The input conditions include task description data for a multi-device collaborative startup task set and the corresponding current operating status data of each device. This data comes from the real-time acquisition interface of the power grid monitoring system and the task list of the dispatch instruction management platform, and is presented in a structured format. Based on the device entity definitions in the power grid operation semantic graph, a unique identifier parsing process is performed, splitting the device identifiers in the task set into three sub-fields: device type, topology location, and electrical attributes, forming a standardized identifier structure that can be used for graph retrieval. The graph node index construction module is called on the parsed standardized identifier structure to locate the device entity node in the semantic graph that matches the identifier through hash mapping and multi-key matching strategies, and its node index number and adjacency matrix position are recorded, realizing a direct mapping from the task set to the graph entity. After successful matching of entity nodes, the status synchronization interface is called to bind the current operating status data of each device to the corresponding device entity node attribute fields, including parameters such as circuit breaker opening / closing, bus voltage, and protection action status, ensuring that the task devices have real-time status semantics in the graph. For device entities already bound to a specific state, target state pre-processing is performed according to the task target configuration rules. Predefined target states (such as closing, operation under load, etc.) from the task set are written into the node's target state attributes, and state transition markers are generated to facilitate logical deduction during subsequent path expansion. The task queue builder is invoked to sort all matched and bound device entity nodes and their target states according to topology partitioning and electrical association priority, forming a standardized startup task queue with graph index attributes. Through this processing method, the results of the previous step are transformed into standardized task data that can be directly used for path search and dependency reasoning, achieving controllable initialization of multi-device collaborative startup targets.

[0074] For example, a 35kV substation collaborative startup task set includes three devices: bus section I, main transformer A, and medium-voltage side switch QF1. The input conditions include the target state of the task set: bus section I PT is in operation, main transformer A medium-voltage side is closed, and QF1 is closed; and the current operating state: bus section I PT is not in operation, main transformer A medium-voltage side is open, QF1 is open, and there is no voltage on both sides. The parsing process splits the bus section I identifier into device type "bus", location "I section", and electrical attribute "PT device"; main transformer A is split into device type "main transformer", location "A", and electrical attribute "medium-voltage side switch"; and QF1 is split into device type "circuit breaker", location "medium-voltage side", and electrical attribute "rated current 630A". The corresponding device entity nodes in the semantic graph are located through a multi-key matching strategy. The node indices are 12, 25, and 38, and the corresponding adjacency matrix positions are

[12] [],

[25] [], and

[38] [*]. The state synchronization interface marks the bus section I PT not being engaged as a voltage state of "zero voltage", the main transformer A medium-voltage side tripping as "open circuit", and the QF1 tripping as "open circuit and no voltage on both sides". The target state preset processing sets the bus section I PT being engaged as a voltage state of "normal", the main transformer A medium-voltage side closing as "closed", and the QF1 closing as "closed and operating under load", and generates state transition markers in the node attributes. The task queue builder sorts tasks according to the topology partition "bus – main transformer – circuit breaker" and electrical association priority, outputting a standardized startup task queue: [node index 12→25→38], which has a graph index attribute and can be called by subsequent path searches. Execution results show that this queue can accurately trigger dependency reasoning in subsequent path expansion, avoiding path breaks caused by inconsistent identifiers, and significantly improving the effectiveness and security of collaborative startup planning.

[0075] S3.2: Based on the target device identifier in the standardized startup task queue, traverse the operation entity library in the power grid operation semantic graph, retrieve the final state operation action predicate directly associated with each target device, and generate an initial target operation set containing the target operation node and the corresponding post-state description.

[0076] S3.3: For each target operation node in the initial target operation set, extract its preset precondition list, and call the graph state evaluation interface to compare with the current system real-time state data to generate a precondition verification result set marked with state Boolean values.

[0077] Based on the initial target operation set, the input conditions are the list of target operation nodes generated by S3.2 and their corresponding post-state descriptions. The execution objects are the set of state predicates in the power grid operation semantic graph and the current real-time system state data. For each target operation node, the node attribute interface is called to extract its preset precondition list, and the preconditions are recorded as independent verification entries in a structured format. For each precondition, the graph state evaluation interface is called to load the electrical state parameters and topology connection status output by the real-time monitoring system, and state comparison processing based on predicate logic matching rules is performed. If the match is successful, a Boolean value is marked as true in the result entry; if the match fails, it is marked as false. For numerical state conditions, the difference between the real-time acquired value and the condition threshold is calculated. For Boolean state conditions, the true value of the state identifier bit is directly compared with the preset condition value, and the result is used as a Boolean mark for satisfying the state. The precondition verification result set of all target operation nodes is encapsulated into structured data containing condition descriptions, measured values, threshold parameters, and Boolean marks for satisfying the state, serving as the input basis for subsequent reverse dependency retrieval. By extracting state predicates and performing logical matching, the target operation definition from the previous step is transformed into state verification result data, enabling accurate identification of feasible nodes in the multi-device collaborative startup path expansion.

[0078] For example, in a certain collaborative startup task, the target operation node is "execute QF1 closing operation," and its preconditions include "bus section I voltage is normal" and "bus PT is engaged." The real-time status data shows a bus voltage of 10.5kV, the condition threshold is 10.0kV, and the voltage deviation calculation yields a deviation of 0.5kV, falling within the allowable range, thus the voltage condition is met. The status evaluation interface is called to read the bus PT engagement flag, and the result is true, consistent with the condition definition, thus the PT condition is met. The two verification results are output as {"condition":"bus section I voltage is normal","measured value":10.5,"threshold":10.0,"metat":true} and {"condition":"bus PT is engaged","measured value":"true","threshold":"true","metat":true":true}, ultimately forming the precondition verification result set. In this scenario, all preconditions are met, and the path expansion process can directly enter the subsequent reverse dependency retrieval, improving the efficiency and security of startup path generation.

[0079] S3.4: Based on the precondition items that are not satisfied in the precondition verification result set, perform reverse semantic edge tracing in the power grid operation semantic graph, and use the trigger dependency relationship edge retrieval to find the upstream operation entity that can deduce the precondition, so as to generate a reverse dependency search space containing potential precondition nodes.

[0080] Based on the preconditions marked as unmet in the precondition verification result set, these preconditions are used as the starting objects for dependency tracing. The semantic edge reverse traversal interface is invoked in the power grid operation semantic graph, limiting the traversal to only the trigger dependency edge type to avoid introducing irrelevant constraint nodes. For upstream operation entities retrieved during the traversal, their operation action predicates and post-state descriptions are read. State matching is used to determine whether the post-state is completely consistent with the target precondition. If consistent, the operation entity is added to the potential precondition node set. For operation entities already added to the set, the topological connection attributes between them and the target device entity are analyzed, electrical path connectivity and signal transmission delay are calculated, and nodes with connection levels exceeding a preset threshold or delay parameters exceeding safety limits are removed to ensure the physical feasibility of the dependency chain. For the remaining potential nodes, their associated procedural reference semantic edges are retrieved to verify the applicability of the procedural clauses in the current task scenario. Verified nodes are marked as high-confidence dependency sources, and a reverse dependency search space is constructed to form an optional precondition operation set for subsequent path expansion. By using the above-mentioned reverse semantic edge tracing and node filtering processing method, the unmet preconditions of the previous step are transformed into potential pre-operation node data with physical feasibility and procedural adaptability, thus realizing the dynamic modeling basis of dependency relationships in multi-device collaborative startup path planning.

[0081] For example, in a multi-device collaborative startup task, for the target operation "closing the medium-voltage side switch of the main transformer," the precondition verification result set contains the unmet condition "normal medium-voltage side bus voltage." The system performs a reverse trigger dependency edge traversal in the semantic graph, finds the upstream operation "commissioning the medium-voltage side PT," reads the postcondition and finds a 100% match between the target condition and the node, and adds the node to the potential precondition operation set. Further calculation shows that the topological connection level between this node and the main transformer switch is level 2, the signal transmission delay is 0.05 seconds, which is below the delay threshold of 0.1 seconds, and the associated procedure clause number R-PT-002 is valid in the current startup task, thus marking it as a high-confidence dependency source. The dependency search space ultimately includes the node, its physical attributes, and procedure adaptation information for subsequent path tree expansion. In this scenario, reverse tracing significantly improves the responsiveness of path generation to implicit operational dependencies between devices, avoiding execution order conflicts that occur when modeling without dependencies.

[0082] S3.5: Integrate the initial target operation set with the potential preceding operation nodes in the reverse dependency search space, construct a parallel extended path tree from the current system state to the target state, and encapsulate the preceding conditions and topological connection attributes of all operation nodes on the path tree to generate an initial path candidate set containing candidate operation nodes and their preceding conditions.

[0083] Receive the inverse dependency search space generated by S3.4 and the initial target operation set defined by S3.2, perform set merging processing according to the graph node object identifier, and incorporate the target operation node and potential preceding operation node into the path generation input set.

[0084] The path tree construction engine is invoked, taking the current system state vector and the target state predicate set as inputs. Relying on the trigger dependency edges and temporal forced relationship edges of the power grid operation semantic graph, multiple parallel extended branches are generated and organized into a directed path tree structure.

[0085] During the path tree node generation process, the precondition encapsulation module is called for each operation entity to bind its precondition set with the current satisfied state boolean value as a condition attribute object.

[0086] The topology attribute extraction module is invoked synchronously to retrieve the topology port identifier, adjacent node list and electrical coupling parameters of the device entity associated with each operation node, generate topology connection attributes and encapsulate them into the node metadata.

[0087] Traverse the path tree branches and perform time window labeling and mutual exclusion tag binding on nodes with potential conflicts to ensure that the necessary information for timing conflict detection is available in subsequent optimization stages.

[0088] By encapsulating both the conditional attributes and topological connection attributes of node metadata, the logical relationship of the merged set is transformed into a set of candidate operation nodes that can be invoked by downstream dependency reasoning, forming an initial path candidate set with preconditions, thus realizing a structured mapping from system state to target state.

[0089] For example, in a 220kV substation, the initial target operation set includes two final state operations: "execute QF1 closing" and "close the medium-voltage side switch of the main transformer". The inverse dependency search space includes pre-operations such as "put into operation the medium-voltage side PT" and "close the PT secondary circuit breaker". In the system state vector, bus I section is de-energized, QF1 is in the open position, and the PT primary disconnector is open. The path tree construction engine generates three parallel branches. Branch one, according to the trigger chain, will execute "close the PT secondary circuit breaker" → "put into operation the medium-voltage side PT" → "close the medium-voltage side switch of the main transformer". Branch two will execute "close the PT secondary circuit breaker" → "put into operation the medium-voltage side PT" → "execute QF1 closing". Branch three will execute "close the PT primary disconnector" → "close the PT secondary circuit breaker" → "put into operation the medium-voltage side PT" → "execute QF1 closing". The preconditions for each node are marked as satisfied in the graph state evaluation. For example, if the precondition "PT primary disconnector is closed" for "PT is in operation" is not satisfied, the physical connectivity and coupling strength (0.87) between bus I segment and PT port is recorded in the topology connection attributes. The full path tree is labeled with time window data. For example, the earliest start time for "execute QF1 closing" is 35s and the latest end time is 60s. The mutual exclusion marker is associated with "execute QF2 closing", forming the final initial path candidate set containing all nodes. This set significantly improves logical consistency and collaborative security in subsequent dynamic dependency reasoning.

[0090] like Figure 3 As shown, step S4 involves: for each candidate operation node in the initial path candidate set, traversing all its preconditions and retrieving the current system state satisfaction status in the power grid operation semantic graph; performing dynamic activation retrieval based on semantic relationship edges for unsatisfied preconditions to generate implicit operation dependency chains. Specifically, this includes: S4.1: Perform a precondition traversal process on each candidate operation node in the initial path candidate set, extract all missing precondition sets that are not satisfied in the current system state, and generate a list of semantic gaps to be parsed as the execution object for subsequent dependency retrieval.

[0091] S4.2: Based on each missing precondition in the semantic gap list, perform reverse semantic edge matching operation in the power grid operation semantic graph, and use trigger dependency relationship edges and procedure reference relationship edges to locate the upstream operation entity that can generate the precondition, so as to generate a set of direct predecessor operation nodes as the starting anchor point for dependency propagation.

[0092] The input data includes a list of semantic gaps generated by step S4.1. Each element in the list is a precondition that is not met in the current system state, along with a condition type label and node index information.

[0093] For a single missing precondition in the semantic gap list, the reverse edge traversal interface of the power grid operation semantic graph is called, and the search direction is limited to the set of trigger semantic edges pointing from the target node to the source node.

[0094] During the reverse traversal, a condition matching operation is performed on each triggering dependency edge. The causal triggering condition predicate of the edge is compared with the predicate text and parameter value of the currently missing precondition. If the match is successful, the source operation node index of the edge is recorded.

[0095] The index of the source operation node that is successfully matched is searched again in the adjacency table of the procedure reference relationship to filter out the upstream operation nodes that are in the same procedure clause reference range and whose applicable scenario tags match the conditions of the current task startup.

[0096] For nodes that have passed both the trigger dependency edge and the procedure reference edge filtering, construct a set of direct predecessor operation nodes by node ID, and assign a starting anchor point identifier to each node in the set as the starting data object of the dependency propagation path.

[0097] By using the reverse semantic edge matching process described above, each item in the semantic gap list is transformed into a set of direct predecessor operation nodes that can be used for S4.3 recursive state verification, thereby realizing the starting point location of implicit dependency chain reasoning.

[0098] For example, in a multi-device collaborative startup task, the semantic gap list contains the conditional predicate "Bus Section I voltage is normal". During the search of the trigger dependency edges in the power grid operation semantic graph, a trigger edge is matched, pointing from the operation node "PT1 in" to the predicate "Bus Section I voltage is normal". This condition is supported by the procedure clause number GH-220 and applies to the scheduling scenario label "parallel startup". During the reverse traversal, the source node "PT1 in" of this edge is recorded as a candidate node, and its procedure reference is then confirmed to meet the current task condition in the procedure reference edge adjacency list. Dependency propagation is then performed using this node as the starting anchor. In performance verification, this approach significantly improves the accuracy of dependency chain starting point location. For example, in a certain test scenario, the trigger dependency matching time is only one-third of the original matrix retrieval, and the starting point location coverage is significantly improved.

[0099] S4.3: Perform recursive state verification processing on each node in the set of direct predecessor operation nodes. If the node's own prerequisites are still not met, continue to call the temporal forced relation edge and the mutual exclusion prohibition relation edge to trace upwards at multiple levels to generate the original dependency propagation path tree containing the complete causal logic sequence.

[0100] For each node in the set of direct precursor operation nodes, the power grid operation semantic graph state verification interface is called to compare all the preconditions defined by the node with the current system operation status data, and output the precondition satisfaction flag and the missing condition list.

[0101] For the missing condition list, a recursive dependency retrieval process is initiated. Taking the current node as the recursive entry point, the set of temporally enforced relation edges associated with it is traversed to generate a set of upstream operation candidate nodes with strict order constraints, and the order index value is recorded for multi-level path construction.

[0102] In the generated upstream operation candidate node set, perform mutual exclusion prohibition relationship edge retrieval, identify operation nodes that have resource conflicts or electrical state mutual exclusion with the current node, mark them as conflict nodes and add isolation flag parameters to ensure that conflict branches are excluded during the recursive tracing process.

[0103] The upstream operation nodes that pass the forced timing and mutual exclusion filtering continue to be recursively verified, and the precondition comparison and missing condition parsing operations are repeated, thereby continuously expanding the complete causal dependency chain in multi-level recursion.

[0104] The node sequences, relation edge types, and sequence index values ​​generated during the multi-level recursion process are aggregated into a single data structure to form the original dependency propagation path tree that contains all causal logic sequences and covers triggering, timing, and mutual exclusion relationships.

[0105] By using recursive state verification and multi-level relational edge tracing, the result of the direct predecessor operation node set in the previous step is transformed into the original dependency propagation path tree data with a complete causal chain, thus achieving the expected technical effect of dynamic emergence of dependency chains and complete modeling of multi-device collaborative logic in the path planning stage.

[0106] For example, in a multi-device collaborative startup scenario at a 220kV substation, the direct preceding operation node set includes the closing of the high-voltage side circuit breaker of the main transformer (node ​​A) and the activation of the medium-voltage busbar PT (node ​​B). The preconditions for node A include high-voltage busbar energization and the bus tie switch being in the open position. The state verification interface detects that the "bus tie switch open position" condition is not met and adds it to the missing condition list. The recursive dependency retrieval calls the timing-forced relation edge of node A to locate the upstream operation node C (bus tie switch open position operation), with the sequence index value set to 1. The mutual exclusion relation edge detection result shows that node C has no resource conflict, and proceeds to the next level of recursive state verification. The precondition for node C is "voltage balance on both sides of the bus tie switch," which is not met, triggering further recursion. The timing-forced relation edge is called again to locate the upstream node D (busbar segment phase verification operation), with the sequence index value set to 2, and the mutual exclusion relation edge verification confirms no conflict. The state verification of node D passes, with no missing conditions, and the recursion terminates. The final path tree sequence structure is: Node D → Node C → Node A, with sequential index values ​​of 2, 1, and 0 respectively. The relation edge types include both time-driven and trigger-dependent types, and the original dependency propagation path tree completely covers the causal logic chain. In this scenario, the generation of the path tree ensures that the high-voltage circuit breaker closing operation has all the necessary safety prerequisites, improving the safety and execution efficiency of the coordinated startup.

[0107] S4.4: Based on the original dependency propagation path tree, perform dynamic activation filtering to remove conflicting branches involving mutually exclusive prohibited semantic edges and merge duplicate parallel dependency branches to generate a logically consistent and time-inconsistent standardized implicit operation dependency chain.

[0108] Based on the node sequence of the original dependency propagation path tree and its associated semantic relationship edge information, the mutual exclusion prohibition semantic edge conflict detection module is called to perform branch scanning on the path tree, marking all branch node pairs with mutual exclusion prohibition relationships, forming a conflict branch identifier set.

[0109] For each conflict chain in the conflict branch identifier set, perform branch culling, remove the path branch corresponding to the conflict identifier from the original dependency propagation path tree, and maintain the index continuity of the remaining branches to ensure that the path tree structure does not become invalid due to branch culling.

[0110] After the conflict branches are removed, the parallel branch merging is called on the path tree. Branches with the same target state and completely identical precondition sets are merged, and the operation node sequence of the parallel branches is compressed into a single standardized path node sequence, reducing redundant dependency reasoning.

[0111] During the merging process, the execution order of the merged path node sequence is adjusted according to the directionality of the temporally forced semantic relationship edges, and the order of nodes with time dependencies is rearranged to eliminate temporal conflicts caused by branch merging.

[0112] The merged and rearranged path tree is subjected to logical consistency verification. The causal chain verification mechanism is invoked to ensure that the preconditions of all operation nodes are satisfied in their predecessor nodes and there are no temporal contradictions, thus forming a logically self-consistent data structure.

[0113] Through the above dynamic activation filtering process, the original dependency propagation path tree is transformed into a standardized implicit operation dependency chain with a regular structure, no mutual exclusion conflicts and causal integrity, so as to achieve highly reliable output of multi-device collaborative dependency chain.

[0114] For example, in a scenario of multi-device collaborative startup in a power grid, the original dependency propagation path tree includes a startup chain for device A and a startup chain for device B. Node opA3 of device A and node opB2 of device B have a mutually exclusive relationship within the same time window. After the conflict detection module detects this relationship, it marks the branch containing opB2 as a conflicting branch and removes it. After removal, the remaining path tree has two parallel branches: branch 1 is opA1→opA2→opA3, and branch 2 is opA1→opA2→opA4. The two branches have the same set of preconditions and the same target state, so they are merged into opA1→opA2→opA3 / opA4. During the merging process, based on the temporal mandatory relationship edge, it is found that opA4 can only be executed after opA3 is completed. The order is rearranged to opA1→opA2→opA3→opA4, eliminating potential time conflicts. The final verification process confirms that the preconditions of each node in this dependency chain have been satisfied in the preceding nodes, and there is no causal violation. For this scenario, the execution parameters for conflict detection and merging were set to a conflict scan depth of 4 layers, a merging criterion of identical preconditions and the same target state, and a rearrangement execution window of 1 minute. The standardized implicit operation dependency chain output after processing showed a significantly improved initialization success rate and startup stability in the scheduling execution verification, achieving the expected technical effect.

[0115] S4.5: Perform semantic encapsulation processing on the standardized implicit operation dependency chain, and map the activation order, dependency type and associated procedure clause of each operation node in the chain into structured dependency metadata to generate the final implicit operation dependency chain object that can be called by the subsequent compliance scoring and path optimization modules.

[0116] Step S5: Based on the implicit operation dependency chain, calculate the semantic depth, procedural compliance score, and historical collaboration frequency weight for each propagation path; eliminate operation combinations with incomplete propagation chains or those triggering mutually exclusive prohibited semantic edges; and generate a compliant path candidate set. Specifically, this includes: S5.1: Obtain node sequence data and semantic relationship edge type information in the implicit operation dependency chain, count the number of jump levels from the initial executable action to the current candidate operation node based on the graph traversal method, and generate semantic depth parameters that represent the complexity of logical derivation.

[0117] Data extraction is performed on the node sequence data and semantic relationship edge type information of implicit operation dependency chain objects to determine the complete link structure from the initial executable action node to the current candidate operation node. Based on the extracted link structure, the traversal interface of the power grid operation semantic graph is called to map the serialized node relationships into a traversable adjacency matrix representation, so as to accurately identify the jump path and its level during the traversal process. For each node jump during the traversal process, a level accumulation process is performed, and different accumulation weights are set for different types of jumps according to the semantic relationship edge type (trigger dependency, mutual exclusion prohibition, timing enforcement, procedure reference), thereby realizing a quantitative characterization of the logical deduction complexity. The semantic depth parameter is calculated, and the weights of each level in the traversal path are merged and accumulated to obtain a depth quantification result representing the logical deduction complexity. Through the semantic depth parameter quantification processing method, the implicit operation dependency chain of the previous step is transformed into deduction complexity data that can be called for subsequent procedure compliance calculation and path optimization, thereby optimizing the input parameters of the path compliance screening process.

[0118] For example, in a scenario of coordinated startup of multiple devices in a 500kV substation, the implicit operation dependency chain contains 8 nodes, covering three types of semantic edges: trigger dependency, timing enforcement, and procedure reference. During traversal, the weight of trigger dependency edges is set to 1.0, timing enforcement edges to 1.2, and procedure reference edges to 0.8. After the adjacency matrix mapping is completed, the traversal path crosses level indices 1 to 8. The semantic depth parameters are calculated using the above formula: Level 1 has a trigger dependency weight of 1.0, Level 2 has a trigger dependency weight of 1.0, Level 3 has a procedure reference weight of 0.8, Level 4 has a trigger dependency weight of 1.0, Level 5 has a timing enforcement weight of 1.2, Level 6 has a trigger dependency weight of 1.0, Level 7 has a procedure reference weight of 0.8, and Level 8 has a trigger dependency weight of 1.0. The cumulative result is 1.0 + 1.0 + 0.8 + 1.0 + 1.2 + 1.0 + 0.8 + 1.0 = 7.8, which is passed as the semantic depth parameter in this scenario to the procedure compliance scoring module. The verification results show that the high-scoring path selected after depth quantization has significantly improved execution stability and interlock procedure compliance during the actual startup process.

[0119] S5.2: Receive semantic depth parameters and the procedure entity numbers referenced in the implicit operation dependency chain, use the procedure clause matching engine to search the scheduling procedure knowledge base, compare the degree of conformity between the preconditions of each operation in the dependency chain and the procedure clauses, and generate a procedure compliance score that represents the legality of the operation.

[0120] S5.3: Extract historical collaboration pattern tags contained in the procedure compliance score and implicit operation dependency chain, call the historical case similarity calculation model to analyze the operation frequency of the same equipment combination in past successful startup records, and generate historical collaboration frequency weights that characterize the reliability of experience.

[0121] The system receives the procedure compliance score and historical collaboration pattern labels contained in the implicit operation dependency chain as input parameters to establish a retrieval entry point for matching device combinations. It then calls the historical case similarity calculation model to solve for the matching degree of the input device combination in historical successful startup records, and performs a one-to-one mapping process between the combination identifier and the historical case index.

[0122] Execution frequency statistics are performed on the matched historical case set to extract the frequency of occurrence of the same device combination and its corresponding time period in each case, forming an original frequency data matrix. Based on the original frequency data matrix, execution time normalization is performed to convert frequency values ​​of different period lengths to a unified time base to ensure frequency comparability between combinations.

[0123] A frequency weighting calculation formula is constructed. By introducing the procedure compliance score as a weighting coefficient, the historical execution frequency and procedure compliance are multiplied and fused to achieve quantitative calculation of empirical reliability.

[0124] Where W is the historical collaboration frequency weight, f is the normalized historical execution frequency value, c is the procedure compliance score of the corresponding combination, and N is the number of samples in the statistical period. Threshold filtering is applied to the calculated weight values ​​to remove combinations below a preset lower limit, ensuring that the weight results only cover paths with high empirical reliability.

[0125] By using weight calculation and filtering mechanisms, the procedural compliance score and historical collaboration pattern label are deduced and transformed into historical collaboration frequency weights that characterize the reliability of experience, thereby achieving quantitative reinforcement of the reliability of the path candidate set.

[0126] For example, in a scenario of multi-device coordinated startup in a substation, the input procedure compliance score is 0.85, and the historical coordination mode label combination of the implicit operation dependency chain is "Main Transformer A + Busbar Section I + Backup Feeder B". The historical case database shows that this combination has successfully started 15 times in the last 30 days. The statistical period N is set to 30 days, and the daily frequency f after normalization is 0.5. Multiplying f by the procedure compliance score c=0.85 and summing the results yields a weight value W=0.425. After filtering with a threshold of 0.3, this combination is retained and assigned a historical coordination frequency weight of 0.425. In practical applications, when this combination enters the path ranking stage, its high-frequency startup history and high procedure compliance work together to significantly improve its ranking in the comprehensive evaluation index, thereby increasing its probability of being selected into the compliant path candidate set.

[0127] S5.4: Integrate semantic depth parameters, procedural compliance scores, and historical collaboration frequency weights, and construct a comprehensive evaluation index for path quality using multi-objective weighted fusion. Quantitatively rank all propagation paths in the initial path candidate set to generate a path ranking list with a comprehensive score label.

[0128] Based on the semantic depth parameters, procedure compliance scores, and historical collaboration frequency weights already generated in the implicit operation dependency chain object set, multi-indicator integration processing is performed to construct a comprehensive path quality evaluation index. Unit normalization is applied to the semantic depth parameters and procedure compliance scores to map all indicators to the [0,1] interval while preserving the relative distribution characteristics of the original values, ensuring that indicators of different dimensions can participate in the same evaluation system. The normalized indicator set is initialized using a weight allocation matrix. The weight values ​​are extracted from a preset knowledge base based on system security requirements and scheduling efficiency priorities. A comprehensive score is calculated through multi-objective weighted fusion, with the specific calculation formula as follows: in, The normalized semantic depth parameter, This indicates the score for compliance with the normalization procedure. This represents the normalized historical collaboration frequency weight, where weight is... , , Provided by the system security level configuration file. Propagation paths in all initial path candidate sets are sorted in descending order according to the calculated comprehensive evaluation index. A stable sorting algorithm is used to preserve the original order of paths with the same score to prevent breakage of temporal dependency chains. Through this fusion processing method, the result of the previous step is transformed into a quantifiable path ranking list, achieving a path selection effect that balances high security and high efficiency.

[0129] S5.5: Read the path sorting list and the mutually exclusive prohibition semantic relationship edges defined in the power grid operation semantic graph, execute the conflict detection logic to determine whether there are operation combinations that trigger the mutual exclusion rule within the same time window, eliminate invalid paths with incomplete propagation chains or that violate mutual exclusion constraints, and generate a compliant path candidate set that only contains safe and feasible solutions.

[0130] Step S6: Map the compliant path candidate set to a time interval sequence and introduce the concept of a collaborative window. Execute a spatiotemporal constraint satisfaction method to jointly optimize the paths of each device, generating a multi-device collaborative startup path sequence containing compressed scheduling instruction steps. Specifically, this includes: S6.1: Obtain each candidate operation node and its associated implicit operation dependency chain from the candidate set of compliant paths. Based on the execution time attribute and security risk level of the operation entity, perform time interval mapping processing on each candidate operation node to generate an initial operation time window set containing the earliest start time and the latest end time.

[0131] Based on the sequence of operation nodes and their implicit operation dependency chains of the compliant path candidate set, the operation entity attribute interface in the power grid operation semantic graph is called to extract the execution time parameter and safety risk level value as input factors for time interval mapping.

[0132] The execution time parameter is vector-added with the time when the preconditions in the dependency chain are met to determine the earliest start time of each candidate operation node, and the execution time is accumulated to the end state time of the dependency chain to obtain the preliminary latest end time.

[0133] For operation nodes with a security risk level exceeding the threshold, a risk buffer coefficient is introduced, and the risk buffer calculation formula is called: Where R is the original execution time, K is the buffer multiplier corresponding to the risk level, and S is the safety risk level value. This formula is used to adjust the time interval to increase the operation fault tolerance window.

[0134] Perform interval normalization on the time intervals of all candidate operation nodes to map time units from different sources to the scheduling unit system, and use the temporal enforcement relationship in the dependency chain to correct the start and end boundaries of the time intervals.

[0135] The generated initial operation time window set is indexed and bound to the candidate operation node identifier, enabling bidirectional query capability from operation node to time interval.

[0136] This processing method transforms the candidate set of compliant paths from the previous step into a set of initial operation time windows with complete time boundary descriptions, enabling precise time input for subsequent spatiotemporal constraint model construction.

[0137] S6.2: Based on the initial operation time window set and the temporal forced semantic relationship edges in the graph, construct a spatiotemporal constraint model for multi-device collaborative startup, and use interval algebra operation rules to perform conflict detection processing on operation combinations with mutually exclusive prohibited semantic relationship edges to generate a conflict constraint matrix marked with potential temporal conflict points.

[0138] Based on the initial operation time window set and the temporal forced semantic relationship edges in the power grid operation semantic graph, a time constraint model for multi-device collaborative startup tasks is constructed. The earliest start time and latest end time of each time window are used as time interval elements, and the order constraint parameters of the temporal forced relationship edges are introduced to form a time interval dataset with logical dependency labels.

[0139] The time interval dataset is parsed using interval algebra operations. Intersection, union, and difference operations are performed on the time windows of each candidate operation to generate an interval relationship matrix representing time overlap and interval relationships. Each element in this matrix maps to the temporal relationship state between the corresponding operation nodes, supporting direct indexing for subsequent conflict detection.

[0140] The system invokes predefined mutually exclusive prohibition semantic relationship edges in the graph, and matches each element marked as temporally overlapping in the interval relationship matrix to identify combinations of operations with mutual exclusion constraints within the same time window. All combinations satisfying the mutual exclusion prohibition condition are recorded as potential conflict entries, along with metadata including the mutual exclusion relationship type and the source procedure clause number.

[0141] A conflict constraint matrix is ​​constructed based on a set of potential conflict entries. Time interval indices are bound to mutual exclusion tags, giving the matrix the dual semantic capability of simultaneously representing time constraints and security isolation. The conflict intensity parameter is calculated using the following formula: in, Let i be the conflict strength between operation i and operation j. The length of the intersection of time windows. and These represent the total time window lengths for the two operations. Based on the judgment logic that the conflict intensity parameter exceeds a preset threshold, the conflict flag status of the corresponding element is updated to a high conflict level. Through multi-matrix fusion, the time mapping result of the previous step is transformed into a conflict constraint matrix with the ability to locate potential temporal conflict points, thereby realizing the prediction and optimization of the safe execution window for multi-device collaborative startup path sequences.

[0142] S6.3: For the non-conflict operation subset in the conflict constraint matrix, according to the propagation direction of the trigger dependency semantic relationship edge, perform dynamic cooperative window merging, and perform overlapping compression processing on multiple independent operation time windows that satisfy the electrical connection topology adjacency and the historical cooperative frequency weight is higher than a preset threshold, so as to generate a coarse-grained cooperative scheduling instruction step sequence with parallel execution capability.

[0143] Based on the non-conflicting operation subset in the conflict constraint matrix, the trigger dependency semantic relationship edge propagation direction parsing module is invoked to perform directional labeling on the order of the end of the dependency chain of each operation node. The labeling results are then cross-compared with the device port adjacency table output by the electrical topology connectivity analysis to generate an operation pairing set that satisfies the electrical connection topology adjacency condition.

[0144] The historical collaboration mode frequency retrieval engine is invoked for the operation pairing set to read the collaboration frequency weight value of the corresponding pair in the historical collaboration mode weight matrix, and the pairings that are not lower than the preset collaboration weight threshold are marked as high collaboration potential combinations to form a candidate list of high collaboration operation time windows.

[0145] The candidate list of high collaborative operation time windows is used to determine the overlap of time windows. Based on the earliest start time and latest end time parameters of each operation time window, the overlap rate of time windows is calculated. Time window pairs with an overlap rate that reaches or exceeds a preset threshold are entered into the overlap compression processing queue.

[0146] Dynamic cooperative window merging is performed on the compression processing queue. The time windows are merged according to the priority order of the triggering dependency edge direction. The start and end times of the merged time window are taken as the union of the earliest start time and the latest end time of each operation, forming the initial sequence of coarse-grained cooperative scheduling instruction steps for this step.

[0147] Perform topology consistency checks on the initial sequence of coarse-grained cooperative scheduling instruction steps to verify that there are no broken chains, conflicts, or violations of timing-mandated semantic edges in the electrical topology and semantic dependency chains for all parallel execution operations. Output the sequence that passes the check as a coarse-grained cooperative scheduling instruction step sequence with parallel execution capability.

[0148] By using the dynamic collaborative window merging processing method described above, the time interval and conflict constraint matrix results of the previous step are transformed into parallel execution time window data that meet the conditions of topological adjacency and high collaborative frequency, thereby realizing instruction step compression and parallel execution capability improvement for multi-device operations.

[0149] S6.4: Based on the coarse-grained collaborative scheduling instruction step sequence, a rigid time anchor constraint defined by the procedure reference semantic relationship edge is introduced. An improved heuristic search algorithm is used to perform local reordering optimization on the operation order within the collaborative scheduling instruction step sequence, so as to generate the optimal multi-device collaborative startup path sequence that satisfies all preconditions and has the minimum total startup time.

[0150] S6.5: Obtain the optimal multi-device collaborative start-up path sequence, and in conjunction with the real-time refresh frequency of the power grid primary equipment status data, perform discretization encoding and instruction formatting encapsulation processing on each collaborative scheduling instruction step in the path sequence to generate a multi-device collaborative start-up path sequence containing compressed scheduling instruction steps that can be directly sent to the scheduling execution terminal.

[0151] Based on the input conditions of the optimal multi-device collaborative startup path sequence, the execution time attribute, safety risk level and topology association information of each collaborative scheduling instruction step in the path sequence are extracted, and the real-time refresh frequency of the power grid primary equipment status data is used as the time discretization benchmark.

[0152] The execution start and end time intervals of each coordinated scheduling instruction step are discretized and encoded. The refresh frequency value is used as the sampling interval to map the continuous time interval into a time slice index sequence with a fixed step size, so as to ensure that the scheduling execution terminal accurately loads and executes the operation within a fixed refresh period.

[0153] The path sequence node index mapping module is invoked to perform bidirectional binding encoding between the time slice index and the corresponding operation entity identifier, providing a verifiable time sequence unified identifier for subsequent instruction encapsulation.

[0154] Based on the parallel execution capability and mutual exclusion constraints of the multi-device collaborative startup path, the encapsulated time-slice sequence is formatted to construct an instruction data structure containing field information, including operation identifier, time-slice index, execution parameters, procedure reference number, risk level marker, etc.

[0155] A CRC checksum is used to generate a check field in the formatted instruction data structure and embed it at the end of the instruction to ensure that the scheduling execution terminal can detect data integrity and prevent timing misalignment during the receiving and parsing process.

[0156] By using discretization encoding and instruction formatting encapsulation, the optimal path sequence result from the previous step is transformed into multi-device collaborative startup path scheduling instruction data that can be directly issued, enabling the startup command to be executed instantly, accurately, and securely at the scheduling execution terminal.

[0157] Step S7: Based on the multi-device collaborative startup path sequence, key operations are forced to fall within the manual confirmation time window, and the activated semantic relationship edge types, referenced procedural clauses, and historical similar collaborative case numbers are extracted simultaneously to generate a final startup path scheme with an attached dependency propagation evidence chain. Specifically, this includes: S7.1: Perform risk level assessment on each operation node in the multi-device collaborative startup path sequence, identify high-risk operation nodes involving loop closing, loop opening, and load testing based on the pre-set power grid operation procedure knowledge base, and generate a set of key operations containing high-risk operation identifiers as input objects for subsequent procedure anchor point settings.

[0158] S7.2: Perform time window mapping transformation processing based on the set of key operations, calculate the minimum manual confirmation time required for each high-risk operation node using the scheduler's historical response delay statistical model, force the operation nodes in the set of key operations to be mapped to independent manual confirmation time window intervals, and generate a sequence of procedure anchor points with time constraint markers.

[0159] S7.3: Perform semantic backtracking retrieval processing for each anchor node in the procedure anchor node sequence, traverse the triggering dependency edges and timing-forced edges connected to the anchor node in the power grid operation semantic graph, extract all active semantic relationship edge types and referenced procedure clause numbers from the initial state to the current anchor state, and generate the original dependency propagation evidence fragment set.

[0160] For each anchor node in the procedure anchor sequence, the time constraint markers and node identifiers generated by the preceding S7.2 are read and used as the starting input objects for semantic backtracking retrieval. The power grid operation semantic graph retrieval interface is called, limiting the search scope to trigger-dependent semantic edges and time-forced semantic edges directly connected to the anchor node. A bidirectional traversal strategy based on node indexes is constructed to ensure that both forward and reverse links can be captured. Activation state verification is performed on the trigger-dependent edges matched during the traversal. State predicate matching is used to confirm that the dependency edge has been activated during path planning, and the edge's type code and associated causal logic sequence identifier in the graph are recorded. Temporal order consistency verification is performed on the matched time-forced edges. Edges that do not meet the anchor time window constraints are removed, while edges with reasonable order are retained, and their reference clause numbers in the procedure entity are extracted to form a mapping table between clause numbers and node indices. The aforementioned trigger-dependent edge types and procedure clause numbers are combined into structured evidence units and sequentially accumulated along the traversal path from the initial state node to the current anchor node, generating a set of original dependency propagation evidence fragments containing complete relationship chains. By using structured encapsulation and index mapping, the results of the previous step are transformed into raw dependency propagation evidence fragments that can be invoked by subsequent similarity matching processing, thus achieving the traceability of dependency relationships within the anchor time window.

[0161] For example, in a multi-device collaborative startup task of a power grid, the first type of procedure anchor node—the loop closing operation of section I of the 220kV busbar—has a manual confirmation time window of 180 seconds. This node is indexed as E-OP-220-IHX in the graph, and its directly connected trigger dependency edges are identified as TDEP-QF1-MX and TDEP-PT-IN, with corresponding causal logics of "circuit breaker QF1 closing triggers normal busbar voltage" and "PT activation triggers voltage detection is available," respectively. The timing-mandated edge is identified as TSEQ-QF1-BUS, and the reference clause number is R-OP-220-SEQ-01. The clause requires that the loop closing operation must be performed after the PT is activated, with an interval of no less than 30 seconds. Activation status verification results show that both trigger dependency edges are in an activated state, and time sequence verification confirms that TSEQ-QF1-BUS meets the anchor time window constraint. The accumulated set of original dependency propagation evidence fragments includes: (Trigger dependency edge type: TDEP, Node: QF1 closing, Clause number: R-OP-220-SEQ-01), and (Trigger dependency edge type: TDEP, Node: PT input, Clause number: R-OP-220-SEQ-01). Through this processing, the output data possesses causal chain integrity and clause citation consistency. Subsequent similarity matching can utilize this fragment set to locate similar loop initiation paths in the historical collaborative case library, thereby significantly improving the traceability of path schemes and the efficiency of security review.

[0162] S7.4: Perform similarity matching and fusion processing on the original dependency propagation evidence fragment set, use the vector similarity in the historical collaborative case library to retrieve the historical similar collaborative case number that is most similar to the current dependency propagation path topology, associate and encapsulate the original dependency propagation evidence fragment set with the historical similar collaborative case number, and generate a complete dependency propagation evidence chain data block.

[0163] The input object for performing similarity matching and fusion processing on the original dependency propagation evidence fragment set is the structured fragment set output by step S7.3, which includes multiple trigger dependency edge types, temporal forced edge types, and procedure clause numbers, as well as the topological path encoding of the corresponding anchor nodes. The semantic relationship sequence of each path in the fragment set is converted into a vector representation, and a preset semantic encoding model is used to map different types of relation edges and procedure numbers into fixed-length feature vectors, forming a standardized path representation matrix that can be used for similarity calculation.

[0164] Based on this standardized path representation matrix, the same semantic encoding process is performed on each stored path in the historical collaborative case library to obtain a set of historical path vectors. This set is then normalized to eliminate the influence of different path lengths or the number of procedures. The vector similarity calculation module is then invoked, using a cosine similarity-based index as the measure of path structure matching, to calculate the similarity value between the current path and historical paths. The formula is as follows: Where Sim represents the similarity value, v is the vector representation of the current path, and h is the vector representation of the historical path. Based on the similarity threshold, several historical collaborative case numbers that are closest to the current dependency propagation path topology are selected to generate a candidate case number set.

[0165] The candidate case number set is fused, and the semantic relationship edge records in the original dependency propagation evidence fragment set are associated and encapsulated with the corresponding historical case numbers. The matching results are recorded using a regularized metadata structure, including information such as the matching degree value, the relationship type of the high matching item, and the procedure number lookup table, so as to generate a complete dependency propagation evidence chain data block.

[0166] By using vector similarity calculation and associated encapsulation processing, the original dependency propagation evidence fragment set from the previous step is transformed into a complete dependency propagation evidence chain data with historical reference numbers and matching scores, thereby improving the traceability and verifiability of the current path scheme.

[0167] For example, in a multi-device collaborative startup task, the original dependency propagation evidence fragment set contains 5 paths, each with a semantic relation sequence of length 12. The semantic encoding model maps relation types to 6-dimensional sparse vectors and procedure numbers to 4-dimensional real vectors, combining them to form a 10-dimensional path feature vector matrix. A historical collaborative case library stores 200 paths, whose vector representations, after normalization, use the cosine similarity formula to calculate the matching degree between each historical path and the current path. For example, the similarity between the current path vector v and a historical path vector h is calculated as follows: This value is significantly higher than the set threshold of 0.7. The top 10 case numbers with the highest matching degree are selected from 200 historical paths and associated and encapsulated with the original evidence fragment sequence according to the path index. The generated data block contains the number of each matching path, the matching score between 0.74 and 0.82, the list of procedural clause references, and the activated semantic relationship types. When verifying the output in the scheduling and review system, this data block significantly improves the confirmability and traceability of the path scheme.

[0168] S7.5: Based on the procedure anchor sequence and the dependency propagation evidence chain data block, perform final scheme assembly processing, embed the procedure anchor sequence with time constraint mark into the corresponding time position of the multi-device collaborative startup path sequence, and attach the complete dependency propagation evidence chain data block as metadata to the final path scheme document to generate a final startup path scheme with dependency propagation evidence chain.

[0169] When assembling the final solution based on the procedure anchor sequence and dependency propagation evidence chain data block, the input conditions are the procedure anchor sequence with time constraint markers generated in S7.2 and the complete dependency propagation evidence chain data block generated in S7.4. Both are standardized and have unified time index and procedure reference fields. A time-series position parsing operation is performed on the procedure anchor sequence, calling the time index mapping table of the multi-device collaborative startup path sequence to locate the constraint marker of each anchor node to the corresponding scheduling instruction step position in the path sequence. An insertion operation is performed on the located scheduling instruction step, embedding the anchor's time window interval data and risk level parameters into the execution metadata structure of the operation node to achieve the binding relationship between the operation node and the manual confirmation time window. Metadata association processing is performed on the dependency propagation evidence chain data block, matching the operation nodes in the path sequence with the causal logic sequence numbers in the evidence chain. The successfully matched evidence chain segment is appended to the procedure reference field of the corresponding scheduling instruction step, forming a bidirectional mapping between path nodes and dependency chain evidence. A structural consistency check is performed on the entire path sequence to verify whether there are conflicts between the embedded procedure anchor point intervals and dependency chain metadata. For example, the same node may be marked with different time windows or reference inconsistent procedure clauses. Conflicts are corrected to unify procedure references and chain activation order. The verified and corrected path sequence is then encapsulated into a final path scheme document. Time constraint markers and dependency chain metadata are stored as additional data blocks in the document's metadata information area, enabling synchronized output of the startup path and dependency propagation evidence chain. Through the above processing, the results of the previous step are transformed into startup scheme data that can be directly invoked by the scheduling execution terminal, achieving dual protection of mandatory manual confirmation of key operations and dependency traceability during multi-device collaborative startup.

[0170] For example, in a multi-equipment coordinated startup task at a 500kV substation, the procedure anchor point sequence includes three high-risk nodes: closing the circuit breaker on the medium-voltage side of the main transformer (manual confirmation time window of 30 seconds), the loop closing operation of bus section I (manual confirmation time window of 45 seconds), and the standby bus load test (manual confirmation time window of 60 seconds). In the dependency propagation evidence chain data block, the dependency chain corresponding to the main transformer closing node includes the trigger dependency edge type "PT commissioning" and the procedure clause number "DL / T5678-2019 Clause 3.2.1", and the dependency chain corresponding to the bus closing node includes the time-series forced edge type "segmented loop unblocking" and the procedure clause number "DL / T 5678-2019 Clause 4.1.3". When performing time-series position parsing, the main transformer closing anchor point is located at the second step of the path sequence scheduling instruction, the bus loop closing is located at the fifth step, and the standby bus test is located at the eighth step. The insertion operation embeds each time window interval into the metadata of the scheduling instruction step node, with risk level parameters of 3, 4, and 5 respectively. During dependency chain appending, the second step node adds the dependency chain paragraph "PT commissioning → circuit breaker closing" and references clause "DL / T 5678-2019 Clause 3.2.1"; the fifth step adds the dependency chain paragraph "segmented loop breaking → bus loop closing" and references clause "DL / T 5678-2019 Clause 4.1.3". During consistency verification, an inconsistency was detected between the procedure clause number appended in the eighth step and the entry in the evidence chain; this was corrected and unified to "DL / T 5678-2019 Clause 5.1.4". The final output document includes a time constraint marker field, dependency chain metadata field, and path execution sequence, possessing the ability for manual confirmation of key nodes during startup and traceability of dependencies. Verification results show that this scheme can significantly improve the safety of the startup process and shorten the overall startup time in scheduling execution simulation.

[0171] Step S8: Output the final startup path scheme to the scheduling execution terminal, and update the historical collaborative mode weights in the power grid operation semantic graph based on actual execution feedback, completing the evolution loop. Specifically, this includes: S8.1: Obtain the actual execution log of the multi-device collaborative startup path sequence returned by the scheduling execution terminal, perform structured parsing processing on the actual execution log, and extract a multi-dimensional execution feature vector including the execution timestamp of the operation node, the compliance status of the procedure clauses, and the time spent on manual confirmation, so as to generate a standardized path execution feedback dataset.

[0172] S8.2: Based on the path execution feedback dataset, the actual execution trajectory is compared and analyzed with the preset multi-device collaborative startup path sequence using time alignment, and the trigger success rate and sequence deviation of each implicit operation dependency chain in actual operation are calculated to generate a set of quantitative evaluation dependency chain effectiveness indicators.

[0173] S8.3: Based on the dependency chain validity index set, the weighted sliding window statistical method is used to dynamically adjust the frequency weight of the historical cooperative patterns associated in the power grid operation semantic graph. The weight values ​​of highly valid dependency chains are increased, while the combined weight values ​​of low-validity or triggering mutually exclusive prohibition semantic edges are decreased, so as to generate an updated historical cooperative pattern weight matrix.

[0174] S8.4: Using the updated historical collaborative mode weight matrix, iteratively refresh the semantic relationship edge attributes in the power grid operation semantic graph, recalibrate the confidence scores of trigger dependencies and temporal forced edges, and generate a new version of the dynamic power grid operation semantic graph with adaptive evolution capabilities.

[0175] S8.5: Based on the new version of the dynamic power grid operation semantic graph, perform graph consistency verification and version archiving operations, and embed the updated historical collaborative mode weights as prior knowledge into the path planning initialization basis of the next round of multi-device collaborative startup task to complete the adaptive evolution closed loop and output the enhanced power grid operation semantic graph.

[0176] This application provides a method for intelligently generating device start-up paths that integrates dynamic analysis of microgrid topology, which can be applied to, for example... Figure 4 In the application environment shown, terminal 102 communicates with server 104 via a network. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices, and server 104 can be a standalone server or a server cluster consisting of multiple servers.

[0177] In one embodiment, a device startup path intelligent generation device integrating microgrid topology dynamic analysis is provided, comprising: a base construction module, a graph generation module, a path expansion module, a dependency chain generation module, a path filtering module, a time optimization module, a startup scheme generation module, and an evolution feedback module, wherein: The base construction module acquires primary equipment status data, secondary logic node information, and typical startup operation records of the power grid. Based on the information, it defines a three-dimensional data structure containing equipment entities, operation entities, and procedure entities. It also constructs a four-dimensional association rule base based on topology connections, electrical constraints, operation procedures, and historical collaboration patterns to generate a power grid operation semantic graph initialization base.

[0178] The graph generation module: Based on the initialization base of the power grid operation semantic graph, it performs semantic enhancement processing on the static connections between graph nodes, embeds four types of semantic relationship edges: trigger dependency, mutual exclusion prohibition, timing enforcement, and procedure reference, and generates a power grid operation semantic graph with dynamic reasoning capabilities.

[0179] Path extension module: Receives a set of multi-device collaborative startup tasks and initializes the target state of each device. Starting from all devices to be started, it initiates parallel semantic path extension in the power grid operation semantic graph and generates an initial path candidate set containing candidate operation nodes and their prerequisites.

[0180] Dependency chain generation module: For each candidate operation node in the initial path candidate set, it traverses all its preconditions and retrieves the current system state satisfaction status in the power grid operation semantic graph. For unsatisfied preconditions, it performs dynamic activation retrieval based on semantic relationship edges to generate implicit operation dependency chains.

[0181] Path filtering module: Based on the implicit operation dependency chain, calculate the semantic depth, procedure compliance score and historical collaboration frequency weight of each propagation path, eliminate operation combinations with incomplete propagation chains or triggering mutually exclusive prohibited semantic edges, and generate a compliant path candidate set.

[0182] Time optimization module: Maps the compliant path candidate set into a time interval sequence and introduces the concept of collaborative window to jointly optimize the paths of each device, generating a multi-device collaborative startup path sequence containing compressed scheduling instruction steps.

[0183] Startup scheme generation module: Based on the multi-device collaborative startup path sequence, it forces key operations to fall within the manual confirmation time window, and simultaneously extracts the activated semantic relationship edge type, referenced procedure clauses and historical similar collaborative case numbers to generate a final startup path scheme with an attached dependency propagation evidence chain.

[0184] Evolution Feedback Module: Outputs the final startup path scheme to the scheduling execution terminal, and updates the weights of historical collaborative patterns in the power grid operation semantic graph based on actual execution feedback, thus completing the evolution closed loop.

[0185] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for intelligently generating device startup paths that integrates dynamic analysis of microgrid topology.

[0186] In one embodiment, a computer device is provided, the computer device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of any of the methods in the above method embodiments.

[0187] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of any of the methods described in the above method embodiments.

[0188] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0189] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0190] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0191] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligently generating device start-up paths that integrates dynamic analysis of microgrid topology, specifically comprising: S1: Obtain the status data of primary equipment in the power grid, information of secondary logical nodes, and typical startup operation records. Based on the information, define a three-dimensional data structure containing equipment entities, operation entities, and procedure entities. Construct a four-dimensional association rule base based on topology connections, electrical constraints, operation procedures, and historical collaboration patterns to generate the initialization basis of the power grid operation semantic graph. S2: Generate a power grid operation semantic graph based on the initialization basis of the power grid operation semantic graph; S3: Receive the set of multi-device collaborative startup tasks and initialize the target state of each device. Starting from all devices to be started, initiate parallel semantic path extension in the power grid operation semantic graph and generate an initial path candidate set containing candidate operation nodes and their prerequisites. S4: For each candidate operation node in the initial path candidate set, traverse all its preconditions and retrieve the current system state satisfaction status in the power grid operation semantic graph. For unsatisfied preconditions, perform dynamic activation retrieval based on semantic relationship edges to generate implicit operation dependency chains. S5: Generate a compliant path candidate set based on implicit operation dependency chains; S6: Generate a multi-device collaborative startup path sequence based on the compliant path candidate set; S7: Based on the multi-device collaborative startup path sequence, force key operations to fall within the manual confirmation time window, and extract the activated semantic relationship edge type, referenced procedure clauses and historical similar collaborative case numbers to generate the final startup path scheme.

2. The intelligent generation method for device start-up paths integrating microgrid topology dynamic analysis according to claim 1, characterized in that, The process of generating a power grid operation semantic graph based on the power grid operation semantic graph initialization base involves performing semantic enhancement processing on the static connections between graph nodes based on the power grid operation semantic graph initialization base, embedding four types of semantic relationship edges: trigger dependency, mutual exclusion prohibition, timing enforcement, and procedure reference, and generating a power grid operation semantic graph with dynamic reasoning capabilities.

3. The intelligent generation method for device start-up paths integrating microgrid topology dynamic analysis according to claim 1, characterized in that, The process of generating a compliant path candidate set based on implicit operation dependency chains involves calculating the semantic depth, procedural compliance score, and historical collaboration frequency weight for each propagation path based on the implicit operation dependency chains, eliminating operation combinations with incomplete propagation chains or those triggering mutually exclusive prohibited semantic edges, and generating a compliant path candidate set.

4. The intelligent generation method for device start-up paths integrating microgrid topology dynamic analysis according to claim 1, characterized in that, The process of generating a multi-device collaborative startup path sequence based on a compliant path candidate set involves mapping the compliant path candidate set to a time interval sequence and introducing the concept of a collaborative window to jointly optimize the paths of each device, thereby generating a multi-device collaborative startup path sequence that includes compressed scheduling instruction steps.

5. The intelligent generation method for device start-up paths integrating microgrid topology dynamic analysis according to claim 1, characterized in that, Following S7, the following also includes: S8: Output the final startup path scheme to the scheduling execution terminal, and update the weights of historical collaborative modes in the power grid operation semantic graph based on actual execution feedback to complete the evolution loop.

6. The intelligent generation method for device start-up paths integrating microgrid topology dynamic analysis according to claim 1, characterized in that, S3 specifically includes: The system acquires a set of multi-device collaborative startup tasks and the current operating status data of each device. Based on the device entity definition in the power grid operation semantic graph, it performs mapping and matching processing on the device identifiers in the task set and the graph nodes to generate a standardized startup task queue with graph index attributes. Based on the target device identifier in the standardized startup task queue, the operation entity library in the power grid operation semantic graph is traversed to retrieve the final state operation action predicate directly associated with each target device, so as to generate an initial target operation set containing the target operation node and the corresponding post-state description. For each target operation node in the initial target operation set, its precondition list is extracted, and the graph state evaluation interface is called to compare the current real-time state data of the system to generate a precondition verification result set marked with state Boolean values. Based on the preconditions that are not satisfied in the precondition verification result set, reverse semantic edge tracing is performed in the power grid operation semantic graph. The upstream operation entity that can deduce the precondition is retrieved by using the trigger dependency relationship edge retrieval, so as to generate a reverse dependency search space containing potential precondition nodes. By integrating the initial target operation set with the potential preceding operation nodes in the reverse dependency search space, a parallel extended path tree from the current system state to the target state is constructed. The preceding conditions and topological connection attributes of all operation nodes on the path tree are encapsulated to generate an initial path candidate set containing candidate operation nodes and their preceding conditions.

7. The intelligent generation method for device start-up paths integrating microgrid topology dynamic analysis according to claim 1, characterized in that, S4 specifically includes: For each candidate operation node in the initial path candidate set, perform a precondition traversal process to extract all missing precondition sets that are not satisfied by the current system state, so as to generate a list of semantic gaps to be parsed as the execution object for subsequent dependency retrieval. Based on each missing precondition in the semantic gap list, a reverse semantic edge matching operation is performed in the power grid operation semantic graph. The upstream operation entity that can generate the precondition is located by using the trigger dependency relationship edge and the procedure reference relationship edge, so as to generate a set of direct predecessor operation nodes as the starting anchor point for dependency propagation. For each node in the set of direct predecessor operation nodes, recursive state verification is performed. If the node's own prerequisites are still not met, the temporal forced relationship edge and the mutual exclusion prohibition relationship edge are called to perform multi-level upward tracing to generate the original dependency propagation path tree containing the complete causal logic sequence. Based on the original dependency propagation path tree, conflicting branches involving mutually exclusive prohibited semantic edges are removed and duplicate parallel dependency branches are merged to generate a logically consistent and time-inconsistent standardized implicit operation dependency chain. The standardized implicit operation dependency chain is semantically encapsulated, and the activation order, dependency type and associated procedure clause of each operation node in the chain are mapped into structured dependency metadata to generate the final implicit operation dependency chain object that can be called by the subsequent compliance scoring and path optimization modules.

8. A device for intelligently generating equipment start-up paths that integrates microgrid topology dynamic analysis, characterized in that, The device includes: The base construction module acquires primary equipment status data, secondary logic node information, and typical startup operation records of the power grid. Based on the information, it defines a three-dimensional data structure containing equipment entities, operation entities, and procedure entities. It also constructs a four-dimensional association rule base based on topology connections, electrical constraints, operation procedures, and historical collaboration patterns to generate a power grid operation semantic graph initialization base. Graph generation module: Based on the initialization base of the power grid operation semantic graph, the static connections between graph nodes are semantically enhanced, and four types of semantic relationship edges are embedded: trigger dependency, mutual exclusion prohibition, timing enforcement and procedure reference, to generate a power grid operation semantic graph with dynamic reasoning capabilities. Path extension module: Receives a set of multi-device collaborative startup tasks and initializes the target state of each device. Starting from all devices to be started, it initiates parallel semantic path extension in the power grid operation semantic graph and generates an initial path candidate set containing candidate operation nodes and their prerequisites. Dependency chain generation module: For each candidate operation node in the initial path candidate set, it traverses all its preconditions and retrieves the current system state satisfaction status in the power grid operation semantic graph. For unsatisfied preconditions, it performs dynamic activation retrieval based on semantic relationship edges to generate implicit operation dependency chains. Path filtering module: Based on the implicit operation dependency chain, calculate the semantic depth, procedural compliance score and historical collaboration frequency weight of each propagation path, eliminate operation combinations with incomplete propagation chains or triggering mutually exclusive prohibited semantic edges, and generate a compliant path candidate set. Time optimization module: Maps the compliant path candidate set into a time interval sequence and introduces the concept of collaborative window to jointly optimize the paths of each device, generating a multi-device collaborative startup path sequence containing compressed scheduling instruction steps; Startup scheme generation module: Based on the multi-device collaborative startup path sequence, it forces key operations to fall within the manual confirmation time window, and simultaneously extracts the activated semantic relationship edge type, referenced procedure clauses and historical similar collaborative case numbers to generate a final startup path scheme with an attached dependency propagation evidence chain.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that... When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1-7.

Images