Semantized security verification method and system for power grid dispatching instruction

Abstract

The invention belongs to the technical field of power system automation, and particularly relates to a power grid dispatching instruction safety prevention and control method and system based on semantic analysis and multi-dimensional dynamic verification, and the method achieves the automatic semantic analysis and dynamic safety verification of a dispatching instruction through the technology of constructing a power grid knowledge graph and analyzing an instruction text to generate a structured semantic tree. The method supports the multi-dimensional conflict detection of the equipment state, the topological relation and the sequential logic, reduces the manual intervention proportion, and improves the verification accuracy and real-time performance.

Description

Technical Field

[0001] This invention belongs to the field of power system automation technology, specifically relating to a method and system for security control of power grid dispatching instructions based on semantic parsing and multi-dimensional dynamic verification. Background Technology

[0002] Against the backdrop of the accelerated evolution of new power systems, the traditional command system, centered on telephone dispatch, manual drafting, and centralized decision-making, has exposed systemic bottlenecks. These bottlenecks are primarily manifested in the high security risks associated with unidirectional dispatch command transmission. The traditional "dispatching end → execution end" command transmission heavily relies on telephone calls and manual repetition, resulting in inherent defects such as voice ambiguity, monitoring blind spots, and high rates of error. Statistics show that the misreading rate of commands due to non-standard terminology or accent differences during complex switching operations reaches as high as 4.7%. Furthermore, manual drafting and shift handover oversights contribute approximately 30% of the causes of power grid accidents. Moreover, when ambiguous commands require manual review, the average processing time is long, and the error interception rate does not reach 60%.

[0003] The semantic security verification method for power grid dispatching instructions is a core technical requirement for the safe operation of modern power systems. Its background stems from the limitations of traditional dispatching models, the escalating risks brought about by the increasing complexity of power grid structures, and the urgent need for intelligent decision-making in new power systems.

[0004] In power grid dispatching systems, although there are models like patent CN116054384A that rely on historical instructions for training, they cannot understand the semantics of the instructions and can only determine whether there is a misoperation by comparing the execution end status of the instructions. This only allows for retrospective judgment of misoperations and cannot prevent risks in advance. Patent CN120067952Aa, which is based on regular expressions or keyword matching, can only detect abnormal order of dispatching instructions in the database and cannot understand the semantics and real-time status. Traditional manual review has a misoperation interception rate of <60% and an average time of >3 minutes, which cannot meet the real-time requirements of new power systems.

[0005] Currently, there is a lack of a method to reduce the risk of misoperation by performing semantic parsing and dynamic security verification of power grid dispatching commands. By using dynamic semantic association, real-time risk prediction, and cross-system collaboration, the prevention and control of misoperation can shift from "post-event remediation" to "pre-event prevention". Summary of the Invention

[0006] The purpose of this invention is to provide a semantic security verification method and system for power grid dispatching instructions, addressing the problems existing in the prior art. By constructing a power grid knowledge graph and generating a structured semantic tree from the instruction text, the invention achieves automatic semantic parsing and dynamic security verification of dispatching instructions. It supports multi-dimensional conflict detection of equipment status, topology relationships, and timing logic, reduces the proportion of manual intervention, and improves verification accuracy and real-time performance.

[0007] The technical solution of this invention is: A semantic security verification method for power grid dispatching instructions includes the following steps: Constructing a dynamic power grid knowledge graph: Integrating SCADA real-time data with CIM power grid models to generate a knowledge graph that includes equipment entities, topology connection relationships, and real-time status attributes; Instruction semantic parsing: The BERT-BiLSTM model is used to parse natural language instructions and output a structured semantic tree, which is labeled with the operation object, action type and constraints. Multi-dimensional dynamic security verification, including: State consistency verification: compare the real-time state of the object operated by the instruction in SCADA with the state required by the instruction to see if there is any conflict. Topology safety verification: Construct node admittance matrix based on CIM model, calculate power flow distribution using Newton-Raphson method, and detect voltage / current overruns; Timing logic verification: Based on a predefined electrical mutual exclusion rule base, verify the state dependencies and operation order conflicts of multiple instruction sequences; Real-time risk blocking: When a conflict is detected in any verification dimension, the command is automatically intercepted and a visual alarm is generated, providing feedback on the cause of the conflict and correction suggestions.

[0008] Specifically, the equipment entities in the construction of the dynamic power grid knowledge graph include: circuit breakers, disconnectors, busbars and transformers; the topology connection relationships include: electrical connections, membership relationships and topology levels; the real-time status attributes include: real-time opening and closing status, load value and voltage level.

[0009] Specifically, the timing logic verification includes: Establish equipment state model: Define state transition rules for equipment such as circuit breakers and disconnectors, and record the real-time state (open / closed) of the equipment. Define electrical logic conflict rules; Conflict detection algorithm.

[0010] Specifically, the defined electrical logic conflict rules include: Mutual exclusion operation detection: There must be no opening / closing conflicts between adjacent commands for the same device; State-dependent detection: The associated circuit breaker must be in the open state before the disconnecting switch is operated; Grounding sequence test: Before grounding, it is necessary to verify that the equipment is under no voltage.

[0011] Specifically, the method of using the BERT-BiLSTM model to parse natural language instructions includes: The BERT layer generates word vector encodings to extract contextual semantic features; BiLSTM layers are used for sequence modeling to capture long-range dependencies; CRF layer annotation includes three types of semantic elements: device entity, operation type, and parameter conditions. The BERT layer uses a BERT model that has undergone domain-adaptive pre-training. The corpus used for the domain-adaptive pre-training includes at least one of the following: power grid dispatching procedures, historical operation tickets and electrical operation guidelines, State Grid Corporation of China power safety work procedures, or power grid operation guidelines. The domain-adaptive pre-training uses the Masked Language Modeling (MLM) task to further pre-train the basic BERT model on the domain corpus.

[0012] Specifically, the instruction semantic parsing also includes: associating semantic tree nodes with device topology relationships in the knowledge graph; verifying whether the operation action conforms to the current state of the device and domain rules, and discarding the instruction if there is a logical conflict. Specifically, a semantic security verification system for power grid dispatching commands includes: Real-time data interface: Accesses SCADA system equipment status and CIM power grid model; Semantic parsing module: Executes the BERT-BiLSTM model to convert natural language instructions into structured semantic trees; Knowledge graph database: a base of storage device topology relationships and security rules; Dynamic verification engine: Performs three-dimensional conflict verification and generates correction alarms, including a status verification unit, a topology analysis unit, and a timing detection unit; Visual feedback interface: An interactive interface that outputs conflict alarms and correction suggestions.

[0013] The topology analysis unit: The node admittance matrix is ​​generated by analyzing the CIM model, and the power flow is calculated by combining the Newton-Raphson method. Compare equipment safety limits to identify risks of overload, voltage exceeding limits, and short-circuit capacity exceeding standards.

[0014] The topology analysis unit performs predictive power flow limit detection, specifically including: simulating the future power grid topology formed after the execution of the scheduling command based on the current power grid model and real-time data; performing power flow calculation based on the future power grid topology to obtain predictive power flow distribution results; and comparing the predictive power flow distribution results with equipment safety limits to generate a limit risk warning.

[0015] The core of the aforementioned topology safety verification lies in predictive simulation: based on the real-time operation mode of the power grid (CIM model) and real-time data (SCADA), it simulates the new topology structure formed after the execution of dispatching instructions. Using this new topology as a basis, a node admittance matrix is ​​constructed, and numerical calculation methods such as the Newton-Raphson method are employed to perform full power flow calculations, obtaining the voltage, phase angle, and power flow distribution of each node and branch in the system after instruction execution. The calculation results are then prospectively compared with the safe operating settings of the equipment (such as upper / lower voltage limits, current thermal stability limits, and short-circuit capacity limits). This allows for the prediction and early warning of potential risks such as voltage exceeding limits, equipment overload, and short-circuit current exceeding limits before the actual issuance of instructions, achieving a leap from 'static verification' to 'dynamic prediction'. The beneficial effects of this invention are as follows: This invention achieves full-dimensional pre-emptive security control of power grid dispatching commands for the first time through a three-level collaboration of a semantic parsing engine (BERT-BiLSTM) → a dynamic verification engine (state / topology / timing) → a real-time blocking mechanism. It solves the inherent defects of traditional technologies such as "post-event remediation, static rules, and no topology prediction" and achieves pre-event risk blocking. It breaks through the integration of BERT-BiLSTM semantic understanding and CIM topology flow calculation, supporting multi-dimensional conflict detection. It constructs an electrical mutual exclusion rule base (including five-prevention rules) to solve the industry problem of timing logic conflicts. The error interception rate is improved: from the traditional 50% to 99.2% (actual test data of a provincial power grid), an improvement of ≥40%. The response speed is optimized: the verification time is <200ms, which is 900 times faster than manual review. It is compatible with complex commands: it supports unstructured commands (such as "disconnect switch A and isolate bus B"). It supports automatic processing of voice commands to text. Attached Figure Description

[0016] Figure 1 The system architecture diagram provided by this invention; Figure 2 : BERT-BiLSTM semantic parsing flowchart; Figure 3 : Schematic diagram of the collision detection algorithm. Detailed Implementation

[0017] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Example 1: This implementation provides a semantic security verification method for power grid dispatching commands, including the following steps: Constructing a dynamic power grid knowledge graph: Integrating SCADA real-time data and the CIM power grid model, a knowledge graph is generated containing equipment entities, topological connections, and real-time status attributes. The equipment entities in the dynamic power grid knowledge graph include circuit breakers, disconnectors, buses, and transformers; the topological connections include electrical connections, membership relationships, and topological hierarchy; and the real-time status attributes include real-time on / off status, load values, and voltage levels. The Neo4j knowledge graph is dynamically updated to ensure that the entity status is synchronized with the power grid in real time. The knowledge graph database also supports graph neural network-based reasoning computation for assessing the potential impact range of dispatch instructions. Therefore, this knowledge graph not only stores static relationships but also utilizes graph computation algorithms to reason about the diffusion range of operational impacts (e.g., which downstream loads and lines will be affected by disconnecting a switch), automatically discover hidden electrical connections, and assist in generating verification paths.

[0019] Instruction semantic parsing: The BERT-BiLSTM model is used to parse natural language instructions and output a structured semantic tree. The flowchart is as follows. Figure 2 As shown, the operation object, action type, and constraints are labeled. In fact, the semantic parsing of the instructions also includes: associating semantic tree nodes with device topology relationships in the knowledge graph; verifying whether the operation action conforms to the current device state and domain rules; and discarding the instruction if a logical conflict exists.

[0020] Multi-dimensional dynamic security verification, including: State consistency verification: compare the real-time state of the object operated by the instruction in SCADA with the state required by the instruction to see if there is any conflict. Topology safety verification: Based on the CIM model, a node admittance matrix is ​​constructed. Power flow distribution is calculated using the Newton-Raphson method to detect voltage / current exceedances. The CIM topology model predicts power flow distribution and detects exceedances by automatically constructing the computational topology (node ​​admittance matrix) of the power network through parsing standardized CIM power grid model files. Combined with given operating conditions (load, generation, equipment status), mature numerical algorithms (such as the Newton-Raphson method) are used to solve for the system's steady-state operating point (voltage, phase angle, power flow). Finally, the calculation results are compared one by one with the equipment's own safe operating limits (voltage, current, power, non-functional capacity), identifying and reporting all situations exceeding safe limits. The core value of CIM lies in providing an accurate and standardized source of power grid models, making the entire process more automated, standardized, and reliable. The topology safety verification unit of this invention does not simply query the current state, but simulates future states based on real-time topology and operating conditions (through power flow calculations), predicting whether operations will trigger potential risks such as voltage exceedances and equipment overloads, thereby achieving true pre-emptive prevention. Timing logic verification: Based on a predefined electrical mutual exclusion rule base, verify the state dependency and operation sequence conflict of multiple instruction sequences. Specifically, this includes establishing a device state model: defining state transition rules for devices such as circuit breakers and disconnect switches, recording the real-time state (open / closed) of the devices; and defining electrical logic conflict rules. Conflict detection algorithm. Real-time risk blocking: When a conflict is detected in any verification dimension, the command is automatically intercepted and a visual alarm is generated, providing feedback on the cause of the conflict and correction suggestions.

[0021] The meaning of "dynamic" is further specified in the following ways: 1) Knowledge graphs integrate SCADA data in real time; 2) Verification is based on real-time status; 3) Power flow calculation is based on the assumed topology after instruction execution for predictive calculation.

[0022] Furthermore, in this embodiment, the electrical logic conflict rules defined above include: Mutual exclusion operation detection: There must be no opening / closing conflicts between adjacent commands for the same device; State-dependent detection: The associated circuit breaker must be in the open state before the disconnecting switch is operated; Grounding sequence test: Before connecting the grounding wire, it must be verified that the equipment is uncharged; Five-prevention rule inspection: Preventing accidental separation or combination, operation under load, etc.

[0023] Furthermore, the method of using the BERT-BiLSTM model to parse natural language instructions in this embodiment includes: generating word vector encodings using the BERT layer and extracting contextual semantic features; BiLSTM layers are used for sequence modeling to capture long-range dependencies; CRF layer annotation includes three types of semantic elements: device entity, operation type, and parameter conditions.

[0024] The BERT layer uses a BERT model that has undergone domain-adaptive pre-training. The corpus used for the domain-adaptive pre-training includes at least one of the following: power grid dispatching procedures, historical operation tickets, and equipment manuals. The domain-adaptive pre-training uses the Masked Language Modeling (MLM) task to further pre-train the basic BERT model on the domain corpus.

[0025] The BERT model described is a domain-adaptive pre-trained version. The pre-training process includes: collecting power grid dispatching procedures, historical operation tickets, equipment technical manuals, and fault reports as domain corpora to construct a pre-training dataset; further pre-training the open-source BERT base model (such as BERT-base) using a masked language model (MLM) task on this dataset to make the model's lexical representation and contextual understanding capabilities more closely aligned with the power dispatching domain. Subsequently, supervised fine-tuning of the domain-specific BERT model is performed using manually annotated instruction sequence samples (labeled with equipment entities, operation types, and parameter condition tags), ultimately yielding the semantic parsing model used in this invention. By employing a domain-adaptive pre-trained BERT model, this invention significantly improves the accuracy of instruction entity recognition and relation extraction ('accuracy increased from 92% to 98.5%') compared to solutions using general models, greatly reducing the risk of misinterpretation due to semantic ambiguity and providing a reliable data foundation for subsequent security verification. During training, the BERT model employs domain-adaptive pre-training based on corpora such as power grid dispatching procedures and operation tickets, enabling the model to more accurately understand the semantics of power industry terminology. The BiLSTM-CRF model is trained on a manually annotated power instruction dataset. This dataset not only annotates entities and actions but also the logical relationships and parameter constraints between instructions, allowing the structured semantic tree output by the model to be directly used in downstream verification logic.

[0026] Example 2 This embodiment provides a semantic security verification system for power grid dispatching commands, including: Real-time data interface: Accesses SCADA system equipment status and CIM power grid model; Semantic parsing module: Executes the BERT-BiLSTM model to convert natural language instructions into structured semantic trees; Knowledge graph database: a base of storage device topology relationships and security rules; Dynamic verification engine: Performs three-dimensional conflict verification and generates correction alarms, including a status verification unit, a topology analysis unit, and a timing detection unit; Visual feedback interface: An interactive interface that outputs conflict alarms and correction suggestions.

[0027] Furthermore, the topology analysis unit: The node admittance matrix is ​​generated by analyzing the CIM model, and the power flow is calculated by combining the Newton-Raphson method. Compare equipment safety limits to identify risks of overload, voltage exceeding limits, and short-circuit capacity exceeding standards.

[0028] The topology analysis unit performs predictive power flow limit detection, specifically including: simulating the future power grid topology formed after the execution of the scheduling command based on the current power grid model and real-time data; performing power flow calculation based on the future power grid topology to obtain predictive power flow distribution results; and comparing the predictive power flow distribution results with equipment safety limits to generate a limit risk warning.

[0029] The execution logic of the topology analysis unit includes the following steps: 1. Topology Modeling: Analyze the latest CIM model and SCADA real-time status to generate the node admittance matrix for the current power grid.

[0030] 2. Instruction simulation: The scheduling instructions to be verified are transformed into network topology modification operations (for example, 'disconnecting switch S1' is simulated as disconnecting the branch corresponding to S1 in the matrix).

[0031] 3. New Matrix Generation: Based on the changed topology, a new node admittance matrix is ​​generated that represents the power grid structure after the command is executed.

[0032] 4. Power Flow Solution: Using the current generator output and load level as boundary conditions, substitute them into the new matrix, and use the Newton-Raphson method to calculate the power flow and solve for the future steady-state operating point of the system.

[0033] 5. Safety Scan: The results of solving the entire network bus voltage, line current, transformer power, etc., are compared one by one with the safety limit list of each device stored in the knowledge base.

[0034] 6. Risk Assessment: Identify and mark all out-of-limit devices and the degree of exceeding limits, and generate a predictive risk report. Compared to systems that only verify the current state, the system provided in this embodiment has a topology security verification unit with forward-looking risk prediction capabilities. It can discover hidden safety problems that may be caused after an operation (such as cascading overloads caused by 'load switching'), which is something that traditional static rule bases and state comparison methods cannot achieve, greatly improving the proactive safety of power grid operations.

[0035] Example 3 This embodiment uses the verification of line transfer maintenance command as an example for illustration.

[0036] Instruction: "Switch line L1 from operation to maintenance." The verification steps include: ① Semantic parsing and instructions: Extract the operation object (L1), target status (maintenance), and implicit action (the associated knife switches K1 and K2 must be disconnected first, and then the associated switch S1 must be disconnected). ② Knowledge graph query: Confirm that L1's current status is "Running" and the associated devices are K1 / K2 / S1; ③ Dynamic verification: If S1 is detected to be in the "protection lockout" state, generate an alarm "S1 is not ready, operation of L1 is prohibited". The purpose of semantic parsing and instruction step decomposition is to transform natural language instructions into a machine-understandable, structured sequence of operational tasks.

[0037] The specific process is as follows: 1. Extract key elements: Operation target: Clearly identify the target device as "Line L1".

[0038] Initial state: The identification command indicates that the line is currently in "running" state.

[0039] Target status: The identification command requires the line to be switched to "maintenance" status.

[0040] 2. Identify implicit actions: Based on power safety regulations and equipment operation logic (such as the "five-prevention" logic), system knowledge The library understands that "transferring a line from operation to maintenance" is not a single operation, but a sequence of operations. The core logic of this sequence is to first disconnect the load current path, and then isolate the line: The specific operating steps are as follows: Step 1: Disconnect the associated switch (S1): The circuit breaker (switch) directly connected to the line must be disconnected first to interrupt the load current. However, the switch operation may be affected by the protection status.

[0041] Step 2: Disconnect the associated disconnect switches (K1, K2): After the switch is opened, disconnect the disconnect switches (K1 and K2) on both sides of the line to achieve physical isolation between the line and the energized system. Only after the switch is opened and the disconnect switches are disconnected can the line be safely grounded and considered to be in a "maintenance" state.

[0042] The key steps by which the system parses the instruction are: Prerequisite action 1: Disconnect the associated disconnector K1 of line L1 (usually, the switch needs to be disconnected first).

[0043] Prerequisite action 2: Disconnect the associated disconnector K2 of line L1 (usually, the switch needs to be disconnected first).

[0044] Prerequisite action 3: Disconnect the associated switch S1 of line L1 (must be performed before disconnecting the disconnector, or at least ensure that it is in the open position and operable). The actual operation sequence may be strictly defined, such as "disconnect the switch first, then disconnect the load-side disconnector, and finally disconnect the power-side disconnector," which is simplified to the core logic here.

[0045] 3. Output structured operation sequence: The system internally generates an operation task chain to be verified. Task 0: Verify the overall feasibility of the instructions (main task); Task 1: Disconnect switch S1; Task 2: Disconnect knife switch K1`; Task 3: Disconnect switch K2`; Set line L1 to maintenance status (this status is automatically achieved upon successful completion of the preceding task).

[0046] The knowledge graph query and static verification steps are broken down, and the operation is verified using a power grid model (knowledge graph). To verify the correctness of the object's state and relationships, a static rule check is performed. The specific process is as follows: 1. Query the status of the operation object: The system queries the current remote signaling status of line L1 in the knowledge graph to confirm that its status is indeed "operating". If the status is inconsistent (such as being under maintenance or hot standby), an alarm will be issued at this step (e.g., "L1 is currently not in operation and cannot execute the switch to maintenance command").

[0047] 2. Query Related Equipment: The system queries the knowledge graph for key equipment that has a direct electrical connection and operational dependency with line L1. Related equipment includes: Switch (circuit breaker): S1 (responsible for interrupting load current); Disconnect switches (knife switches): K1, K2 (responsible for creating a clear disconnect point after the switch is opened, isolating the circuit). (This may also include related grounding switches, protection devices, etc., but this case focuses on the core switchgear.) 3. Static rule check (basic): Based on the connection relationships and device attributes in the graph, perform preliminary verification: confirm that S1, K1, and K2 are indeed connected to L1 in the topology.

[0048] Confirm that the state transition from "operation" to "maintenance" is defined as permissible. (This stage typically does not involve the real-time dynamic status of the equipment, such as protection signals); 4. Output static verification results: Confirm that the line status and associated equipment meet the basic conditions for executing instructions at the model level.

[0049] The dynamic status verification and real-time alarm process is broken down into steps to check the current and real-time operating status and signals of critically related devices, determining whether each step in the operation sequence can be safely executed immediately. This is a crucial step in preventing misoperation. The specific process is as follows: 1. Focus on key prerequisite equipment: The system knows that the first step (and the most critical step) of the operation sequence is to disconnect switch S1.

[0050] 2. Query the real-time status of the equipment: The system obtains the latest remote signaling, telemetry and protection signals of switch S1 through a real-time data interface (such as SCADA / monitoring system).

[0051] 3. Detection of Obstruction Conditions: The system detected a critical signal from S1: "Protection Lockout". "Protection Lockout" means that the switch's protection device (such as relay protection) has prohibited the switch's opening operation for some reason (such as device malfunction, maintenance pressure plate activation, communication interruption, etc.). This is an important safety lockout state to prevent the switch from being disconnected when the device is malfunctioning, which could lead to malfunction or failure of protection.

[0052] 4. Logical reasoning and alarm generation: Reasoning: The first step (disconnecting S1) is a necessary prerequisite for all subsequent operations (disconnecting K1 and K2, setting L1 to maintenance).

[0053] Reasoning: S1 is in the "protection lockout" state, which means that it cannot be operated (opened) at present.

[0054] Conclusion: Therefore, the entire operation command to "switch L1 from operation to maintenance" cannot be safely started and executed.

[0055] 5. Generate precise alerts: The system generates a clear alarm message indicating: Root cause device: S1 Problem status: Not ready (Specific reason: protection interlock) Impact: Operation of target device L1 is prohibited. Recommended action: The protection lockout state of S1 must be released first, and the command can only be executed again after confirming that S1 has returned to a normal and operable state.

[0056] Example of an alarm: "Critical alarm: Associated switch S1 is in the 'protection lockout' state and is not ready. The operation of 'switching line L1 from operation to maintenance' is prohibited! Please check and handle the S1 status." This verification process achieves intelligent, efficient, and accurate verification of the security and executability of the complex operation command "line to maintenance" by intelligently parsing the command intent, using the power grid knowledge graph to verify model relationships, and combining real-time dynamic signal detection. When it detects that the critical equipment (switch S1) is not ready (protection lockout), it promptly generates a clear alarm and locks the operation, ensuring the safety and reliability of power grid operation.

[0057] Example 4: This embodiment uses multi-instruction timing conflict detection as an example for illustration. The detailed process is the same as in Embodiment 3. The conflict detection algorithm flow is as follows: Figure 3 As shown, the specific implementation steps of this embodiment are as follows.

[0058] Step 1: Instruction semantic parsing (corresponding patent) Figure 2 process) 1. Input and Preprocessing: The system receives a sequence of instructions [e.g., "Close switch K1", "Disconnect bus M1"]. These instructions may be in text form or text converted from voice commands.

[0059] 2. Analysis of the BERT-BiLSTM-CRF model: BERT layer: Converts each word in the instruction (such as "close", "switch", "K1", "disconnect", "bus", "M1") into a word vector rich in contextual information.

[0060] BiLSTM layer: Models the instruction sequence and understands the verb-object relationship between "closing" and "switch K1", and the verb-object relationship between "disconnecting" and "bus M1".

[0061] CRF layer: Performs sequence labeling to accurately extract the semantic elements of each instruction. Command 1: Close switch K1: Device entity: K1 (and linked to the "Circuit Breaker" entity in the knowledge graph) Operation type: Closing (shutting down) Parameter conditions: None Command 2: Disconnect bus M1: Device entity: M1 (linked to the "bus" entity in the knowledge graph) Operation type: Disconnect (Severe) Parameter conditions: None 3. Generate structured semantic trees: The system generates two semantic trees, each corresponding to one of the two instructions, and specifies their execution order (instruction 1 first, then instruction 2).

[0062] 4. Knowledge Graph Association: Associate device nodes K1 and M1 in the semantic tree with entities in the knowledge graph to prepare for subsequent verification.

[0063] Step 2: Knowledge Graph Query and Static Verification 1. Query device attributes and status: The system queries from the dynamic knowledge graph: K1: Type is "circuit breaker", current real-time status is open (hypothetically).

[0064] M1: Type is "bus", current real-time status is energized.

[0065] 2. Querying Topology Relationships: The system queries the topology connections in the knowledge graph and confirms that bus M1 is the upstream power source of circuit breaker K1. This means that the power flow is: M1 -> K1 -> (downstream load). This relationship is key to triggering timing conflicts.

[0066] Step 3: Multi-dimensional dynamic security verification (core: timing logic verification & topology security verification) The conflicts in this embodiment mainly occur in the temporal logic dimension, and are verified through topology simulation.

[0067] ③ Timing logic verification: Call the rule base: The system calls the predefined "electrical mutual exclusion rule base", especially the switching operation sequence rule (i.e., the refinement of the "power supply side priority operation rule"): "When de-energizing, the operation must be carried out in the order from the load side to the power supply side (first disconnect the switch, then disconnect the bus); the order is reversed when energizing. Rule matching and analysis: The objective of the instruction sequence is: first close K1 (power supply operation), then disconnect M1 (power outage operation).

[0068] The rule requires that for the power outage operation of disconnecting M1, it must be ensured that all load-side equipment (including K1) is in the disconnected state.

[0069] Conflict point: In the current sequence, before disconnecting M1, K1 was not disconnected, but instead closed, which directly contradicts the rule requirements. The system determines that there is a timing logic conflict.

[0070] ② Topology simulation (an extended application of topology security verification): Building the simulation environment: The topology analysis unit in the dynamic verification engine constructs the node admittance matrix based on the latest CIM model and SCADA real-time data (such as node voltage and load) and initializes the power flow calculation environment.

[0071] Simulation of instruction sequence execution: 1. Simulation Step 1 [Close Switch K1]: Simulate closing circuit breaker K1.

[0072] Power flow calculation: Since upstream M1 is energized, after closing K1, power flows from M1 through K1 to its downstream load (if the load exists, it will carry current).

[0073] Status update: K1 status changes to closed, and its downstream line is energized.

[0074] 2. Simulation Step Two [Disconnect Bus M1]: Simulate disconnecting bus M1.

[0075] Power flow calculation: Disconnecting M1 means removing the power supply to the entire system. The calculation results show that K1 and all its downstream devices instantly lose power, and the current drops sharply to 0.

[0076] Simulation results analysis: The simulation clearly demonstrates a violation of operating procedures—power outage under load. Directly disconnecting the power bus M1 under load will generate a huge electric arc on its circuit breaker, seriously endangering equipment and personal safety, and causing an unplanned power outage.

[0077] Conflict confirmation: The results of topology simulation and the conclusions of timing logic analysis corroborate each other, jointly confirming the high risk of this instruction sequence.

[0078] Step 4: Real-time risk blocking and visual alerts 1. Risk Decision: When both the timing detection unit and the topology analysis unit report a conflict, the dynamic verification engine makes an interception decision after comprehensive judgment.

[0079] 2. Generate alarm information: The system generates structured alarms, including the following content: Conflict types: Sequential logic conflict / Topology security conflict Conflict rule: Violation of the power supply side priority operation rule (switching operation principle) Conflict Description: The instruction sequence requires closing the load switch K1 first, followed by disconnecting its upstream power bus M1. This operational sequence is violated; if executed, it will cause abnormal power loss to K1 and its downstream loads, posing a risk of power outage under load.

[0080] Correction suggestion: Please adjust the operation sequence. The correct power outage sequence should be: first disconnect all downstream load switches (such as disconnecting K1), and after confirming that everything is correct, disconnect the power bus (M1).

[0081] 3. Visual Feedback: This alarm information is presented to the dispatcher through a visual feedback interface. The interface will highlight the conflicting command and the affected equipment (K1, M1), and display the above information in red alarm form, prompting the operator to handle it immediately.

[0082] Example 4 This embodiment uses voice command processing as an example for explanation.

[0083] Specific instruction: "Disconnect all circuit breakers on the 35kV busbar." Verification process: ① Speech recognition → Text conversion; ② Semantic parsing → Locating 35kV busbar associated equipment; ③ Verifying busbar energization status → Intercepting unauthorized operations. Specific operation steps are as described in Example 3, and will not be detailed here.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A semantic security verification method for power grid dispatching commands, characterized in that, Includes the following steps: Constructing a dynamic power grid knowledge graph: Integrating SCADA real-time data with CIM power grid models to generate a knowledge graph that includes equipment entities, topology connection relationships, and real-time status attributes; Instruction semantic parsing: The BERT-BiLSTM model is used to parse natural language instructions and output a structured semantic tree, which is labeled with the operation object, action type and constraints. Multi-dimensional dynamic security verification, including: State consistency verification: compare the real-time state of the object operated by the instruction in SCADA with the state required by the instruction to see if there is any conflict. Topology safety verification: Construct node admittance matrix based on CIM model, calculate power flow distribution using Newton-Raphson method, and detect voltage / current overruns; Timing logic verification: Based on a predefined electrical mutual exclusion rule base, verify the state dependencies and operation order conflicts of multiple instruction sequences; Real-time risk blocking: When a conflict is detected in any verification dimension, the command is automatically intercepted and a visual alarm is generated, providing feedback on the cause of the conflict and correction suggestions.

2. The verification method according to claim 1, characterized in that, The equipment entities in the construction of the dynamic power grid knowledge graph include: circuit breakers, disconnectors, busbars and transformers; the topology connection relationships include: electrical connections, membership relationships and topology levels; the real-time status attributes include: real-time opening and closing status, load value and voltage level.

3. The verification method according to claim 1, characterized in that, The timing logic verification includes: Establish equipment state model: Define state transition rules for equipment such as circuit breakers and disconnect switches, and record the real-time state (open / closed) of the equipment; Define electrical logic conflict rules; Conflict detection algorithm.

4. The verification method according to claim 3, characterized in that, The defined electrical logic conflict rules include: Mutual exclusion operation detection: There must be no opening / closing conflicts between adjacent commands for the same device; State-dependent detection: The associated circuit breaker must be in the open state before the disconnecting switch is operated; Grounding sequence test: Before grounding, it is necessary to verify that the equipment is under no voltage.

5. The verification method according to claim 1, characterized in that, The method of parsing natural language instructions using the BERT-BiLSTM model includes: The BERT layer generates word vector encodings to extract contextual semantic features; BiLSTM layers are used for sequence modeling to capture long-range dependencies; CRF layer annotation includes three types of semantic elements: device entity, operation type, and parameter conditions.

6. The verification method according to claim 5, characterized in that, The BERT layer uses a BERT model that has been pre-trained with domain adaptability. The corpus used for the domain-adaptive pre-training includes at least one of the following: power grid dispatching procedures, historical operation tickets and electrical operation guidelines, State Grid Corporation of China power safety work procedures, or power grid operation guidelines. The domain-adaptive pre-training uses the Masked Language Modeling (MLM) task to further pre-train the basic BERT model on the domain corpus.

7. The verification method according to claim 2, characterized in that, The instruction semantic parsing also includes: associating semantic tree nodes with device topology relationships in the knowledge graph; verifying whether the operation action conforms to the current state of the device and domain rules, and discarding the instruction if there is a logical conflict.

8. A semantic security verification system for power grid dispatching commands, characterized in that, include: Real-time data interface: Accesses SCADA system equipment status and CIM power grid model; Semantic parsing module: executes the BERT-BiLSTM model to convert natural language instructions into structured semantic trees; Knowledge graph database: stores device topology relationships and security rule base; Dynamic verification engine: Performs three-dimensional conflict verification and generates correction alarms, including a status verification unit, a topology analysis unit, and a timing detection unit; Visual feedback interface: An interactive interface that outputs conflict alarms and correction suggestions.

9. The verification system according to claim 8, characterized in that, The topology analysis unit: The node admittance matrix is ​​generated by analyzing the CIM model, and the power flow is calculated by combining the Newton-Raphson method. Compare equipment safety limits to identify risks of overload, voltage exceeding limits, and short-circuit capacity exceeding standards.

10. The verification system according to claim 8 or 9, characterized in that, The topology analysis unit performs predictive power flow limit detection, specifically including: Based on the current power grid model and real-time data, the future power grid topology formed after the execution of the scheduling command is simulated; power flow calculation is performed based on the future power grid topology to obtain predictive power flow distribution results; The predictive power flow distribution results are compared with the equipment safety limits to generate an over-limit risk warning.

Images