A power grid operation mode adaptive adjustment method and system

Abstract

A method and system for adaptive adjustment of power grid operation modes involves acquiring multi-source heterogeneous power grid data and constructing a knowledge base based on this data. Based on the knowledge base, deterministic rule verification is performed on operation instructions to be verified, filtering out instructions that violate preset rules. A large-scale operation instruction model is constructed and trained using the knowledge base to obtain a trained model, outputting dispatch instructions that comply with safety constraints. For operation instructions that pass the deterministic rule verification, power flow calculation verification, N-1 security analysis verification, and a three-level verification for extreme power grid scenarios and major events are performed sequentially. If any level of verification fails, the dispatch instruction is reconstructed by the large-scale operation instruction model, and the three-level verification is re-executed until all verifications pass. This invention achieves automatic generation and closed-loop optimization of safe and compliant dispatch instructions, significantly improving the automation level and safety assurance capability of power grid operation mode adjustment.

Description

Technical Field

[0001] This invention belongs to the field of power system technology, and in particular relates to a method and system for adaptive adjustment of power grid operation mode. Background Technology

[0002] In recent years, with the deepening of smart grid construction, the power grid industry has been committed to improving the level of intelligent operation, creating a green, safe, efficient, and open smart grid, continuously upgrading smart grid infrastructure, and promoting the deep integration of advanced information and communication technologies and control technologies with all aspects and businesses of the power grid. Against this backdrop, achieving unified integration of power grid data and collaborative business connectivity, and improving the power grid's planning, engineering construction, production operation, and intelligent control capabilities, has become a key objective of the industry's digital transformation. At the same time, power grid control systems are developing towards online interconnection, real-time interaction, and intelligent decision-making, aiming to build a control technology support system for the safe, stable, and coordinated operation of power generation, grid, load, and storage, and actively introducing emerging technologies such as artificial intelligence to optimize control professional workflows and business data, comprehensively improving the efficiency and effectiveness of control operations.

[0003] The formation of ultra-high-voltage (UHV) AC / DC hybrid power grids has enhanced my country's ability to optimize the allocation of power resources on a large scale, supporting a significant expansion of the power grid. However, this has also led to a surge in power grid operation modes. On the other hand, the large-scale integration of new power grid components such as renewable energy and energy storage has introduced significant randomness and volatility into grid operation, resulting in a substantial increase in the frequency and methods of grid mode adjustments. These changes have placed unprecedented pressure on dispatching personnel, significantly increasing the difficulty of dispatching operations and posing certain potential risks to the safe and stable operation of the power grid. These risks manifest in several ways: first, the ineffective use of dispatching and operation data; second, the difficulty in accurately analyzing mode adjustment schemes; and third, the difficulty in quickly verifying power flow risks.

[0004] Existing technology 1 ("Research on Typical Scenarios for Power Grid Operation Mode Extraction," Sun Xiaolei, North China Electric Power University) uses an improved DBSCAN algorithm to select feature variables, combines it with improved K-means clustering to extract typical scenarios, and introduces permutation and combination entropy to measure the degree of difference between scenarios. However, this method cannot cover all possible typical operation scenarios, leading to increasing difficulties in identifying operation modes. Another example is existing technology 2, patent publication number CN120999566A, which provides a method for intelligent generation of power grid operation modes based on historical data. Based on a pre-built artificial intelligence model, it extracts operation mode features from historical power grid operation data; generates target features based on these features; performs power flow calculations on the first operation mode data of the power grid; verifies the calculation results with the target features; and checks the current operation mode. However, this method lacks explicit modeling of power grid topology, equipment constraints, and rule-based knowledge, and lacks a knowledge accumulation and updating mechanism.

[0005] Therefore, there is an urgent need for an adaptive adjustment method for power grid operation mode based on operation commands, in order to adapt to the complex and ever-changing characteristics of power grid operation mode and unknown scenarios, and to quickly ensure the security and compliance of dispatch commands. Summary of the Invention

[0006] To reduce the workload of dispatchers and improve the intelligence level of dispatching operations, it is urgent to analyze power grid operation mode adjustment schemes using natural language processing technology, construct an operation instruction knowledge base, and refine, extract, associate, and integrate power grid control operation experience and knowledge, such as business processes, operating procedures, stability limits, and historical dispatch operation data, to achieve online adjustment and safety verification. This paper designs an adaptive adjustment system for power grid operation modes. First, a knowledge base storage unit is constructed, utilizing a power grid scenario equipment operation instruction knowledge extraction network to establish a structured standard library and an unstructured operation instruction knowledge base for the knowledge unit. Second, a deterministic reasoning unit is constructed, performing deterministic verification of equipment status and parameter compliance based on the structured standard library and the power grid topology knowledge base. Then, a large model training and reasoning unit is constructed, using a reward function mechanism and randomly generated fault scenarios to train the agent, forming an intelligent agent for online adjustment and verification of power grid operation mode operation instructions. Finally, an online adjustment and verification unit is constructed to perform power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events on operation commands, and to adjust the verification results online. This enables the autonomous mining and analysis of implicit knowledge in regulation, overcoming the shortcomings of traditional dispatching operations that rely entirely on manual operation, adapting to the complex and ever-changing requirements of power grid operation, and improving dispatching efficiency and safety.

[0007] The present invention adopts the following technical solution.

[0008] This invention proposes an adaptive adjustment method for power grid operation mode, comprising: Acquire multi-source heterogeneous power grid data and construct a knowledge base based on the multi-source heterogeneous power grid data; Based on the knowledge base, deterministic rule verification is performed on the operation instructions to be verified, and instructions that violate preset rules are filtered out; Construct a large model of operation instructions, train it using a knowledge base, obtain a well-trained large model of operation instructions, and output scheduling instructions that meet safety constraints. For operation instructions that pass the deterministic rule verification, power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events are executed sequentially. If any level verification fails, the operation instruction big model automatically reconstructs the scheduling instruction based on the over-limit information or fault information fed back by the verification, and re-executes the three-level verification until all three-level verifications are passed.

[0009] More preferably, the training process of the large model of operation instructions includes: The operation instruction model is trained by reinforcement learning, and guided to generate safe and compliant operation instructions based on multi-dimensional reward signals that include physical feasibility, safety constraints, and scheduling procedure compliance.

[0010] More preferably, the multidimensional reward signal includes: The physical feasibility reward is used to evaluate the convergence of power flow calculations after the command is executed. Safety constraint rewards are used to evaluate the degree to which voltage and line thermal stability safety constraints are met after the command is executed; The scheduling procedure compliance reward is used to evaluate the degree to which instructions conform to expert experience and rules.

[0011] More preferably, in the three-level verification, the first-level power flow is used to determine whether voltage over-limit, branch overload, or transformer overload occurs after the execution of the operation command. The specific steps include: The complete topology and power flow data are obtained from the knowledge base to form a current operating profile that includes all node voltages, branch parameters, reactive power compensation, etc. Based on the type of operation instruction verified by the deterministic rules, the current operating section is modified to obtain the power grid status section after the operation; Calculate the power flow on the grid state profile after the operation, obtain the power flow calculation results, and perform limit checks on node voltages, branch currents, and transformer loads item by item: If there are no violations, the operation instruction is determined to have passed the first level of verification and is sent to the second level. If any of the following conditions exist: voltage overload, branch overload, or transformer overload, the power flow overload verification result under the operation command is determined to be unsuccessful. The overload information, including the overloaded device and its overload quantity, as well as the topology, is transmitted to the trained operation command big model. The operation command big model is adjusted online based on the overload information, and the adjusted operation command is re-executed for the first-level power flow calculation verification until the output passes.

[0012] More preferably, in the three-level verification, the second-level N-1 security analysis verification further verifies the power grid security after any critical equipment failure based on the operation instructions that have passed the first-level verification. The specific steps include: By combining equipment importance labels with topological adjacency relationships, an N-1 set of security analysis faults is generated; For each faulty component in the N-1 security analysis fault set, perform a break simulation and recalculate the power flow on the new topology after the break. If all faulty components meet the conditions, then the N-1 safety criterion is satisfied under the operation command; the conditions include power flow convergence after interruption, no voltage over-limit, and no branch overload. If any faulty component causes any of the above conditions to be unmet, the operation is deemed unsuccessful. The failure information is output to the large operation instruction model to obtain the adjusted operation instruction, and the first and second level verifications are performed again until the operation is successful.

[0013] More preferably, in the three-level verification, the third-level verification for extreme power grid scenarios and major events performs a final verification on the operation instructions that have passed the first two levels of verification. The specific steps include: In extreme weather scenarios, meteorological factors are introduced to verify the dynamic current carrying capacity of the line; the overload capacity of the transformer is verified under extreme high temperature conditions; and the instantaneous failure rate of each equipment component under the current weather scenario is calculated; equipment components with instantaneous failure rates higher than a set threshold are formed into a failure chain. Calculate the product of the instantaneous failure rates of each device component in the fault chain under the current extreme scenario, as the probability of the fault chain occurring; calculate the total load loss after the fault chain is triggered, count the number and magnitude of nodes that exceed voltage or frequency limits after the fault, and determine whether it causes system instability; normalize the results after the above fault chain is triggered, and sum the normalized values ​​by weight to obtain the severity of the consequences. Multiply the probability of occurrence by the severity of the consequences to obtain the overall risk value; If no limit exceedance, instability, or load loss occurs under any extreme weather scenario, and the overall risk value is below the threshold, then the third-level verification is passed. If an over-limit, instability, or loss of load occurs, or if the comprehensive risk value is not lower than the threshold, the third-level verification is deemed to have failed. The fault chain list and the preset preventive control measures are then transmitted to the large-scale operation instruction model for online adjustment. The adjusted operation instructions are then re-executed through the third-level verification process until the third-level verification passes.

[0014] This invention also proposes an adaptive adjustment system for power grid operation modes, comprising a knowledge base storage unit, a deterministic reasoning unit, a large model training and reasoning unit, and an online adjustment and verification unit: The knowledge base storage unit acquires multi-source heterogeneous power grid data and constructs a knowledge base based on the multi-source heterogeneous power grid data; The deterministic reasoning unit performs deterministic rule verification on the operation instructions to be checked, and filters out instructions that violate preset rules; The large model trains the inference unit using a knowledge base to obtain a trained large model of operation instructions, which outputs scheduling instructions that comply with safety constraints. The online adjustment and verification unit, for operation instructions that have passed the determination rule verification, sequentially performs power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events. If any level verification fails, the operation instruction big model automatically reconstructs the scheduling instruction based on the over-limit information or fault information fed back by the verification, and re-executes the three-level verification until all three-level verifications are passed.

[0015] Furthermore, in the online adjustment and verification unit, the first-level power flow is used to determine whether voltage over-limit, branch overload, or transformer overload occurs after the operation command is executed; The second-level N-1 security analysis and verification further verifies the power grid security after the failure of any critical equipment by using the operation instructions verified by the first-level verification. The third level of verification, which targets extreme power grid scenarios and major events, performs a final verification based on the operation instructions from the first two levels of verification.

[0016] The present invention also proposes a terminal, including a processor and a storage medium: The storage medium is used to store instructions; The processor is used to perform the steps of the above method according to the instructions.

[0017] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The formation of ultra-high-voltage AC / DC hybrid power grids has significantly enhanced my country's ability to optimize the allocation of power resources on a large scale and supported the continuous expansion of the power grid. However, it has also led to increasingly complex and numerous power grid operation modes. Simultaneously, the large-scale integration of new energy sources and energy storage into the grid has further increased the randomness and volatility of system operation, resulting in more frequent adjustments to operation modes and increasingly diversified control methods. These changes have placed unprecedented pressure on dispatchers, significantly increasing the difficulty of dispatching operations and bringing new potential risks to the safe and stable operation of the power grid.

[0019] 2. This invention addresses the problems of insufficient data utilization, inaccurate scheme analysis, and untimely power flow risk verification in power grid control and operation by designing an adaptive adjustment system for power grid operation modes. First, a knowledge base storage unit is constructed, utilizing a power grid scenario equipment operation instruction knowledge extraction network to establish a structured standard library and an unstructured operation instruction knowledge base. Second, a deterministic reasoning unit is constructed, performing deterministic verification of equipment status and parameter compliance based on the structured standard library and the power grid topology knowledge base. Then, a large-scale model training and reasoning unit is constructed, using a reward function mechanism and randomly generated fault scenarios to train the agent, forming an online adjustment and verification agent for power grid operation mode operation instructions. Finally, an online adjustment and verification unit is constructed, performing power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events on the operation instructions, and implementing online adjustments based on the verification results.

[0020] 3. This invention introduces the GraphRAG method to deeply integrate the power grid topology knowledge graph with the operation mode command knowledge graph. Upon receiving a control target, it first retrieves relevant knowledge subgraphs based on the graph structure to capture the topological and semantic relationships between entities. Then, it combines a large language model to accurately deduce the sequence of operation commands that meet safety constraints. Subsequently, the deduced commands are deployed in a real power grid or a high-precision simulation environment for verification. Based on the verification results, the knowledge units in the structured standard library are iteratively corrected and optimized, forming a closed-loop mechanism of "reasoning-verification-feedback". Attached Figure Description

[0021] Figure 1 This is a flowchart of a method for adaptive adjustment of power grid operation mode according to the present invention; Figure 2 This is a flowchart of the training process for the large-scale operation instruction model of Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of an adaptive adjustment system architecture for power grid operation modes according to the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0023] like Figure 1 As shown, this invention proposes an adaptive adjustment method for power grid operation mode, comprising the following steps: Acquire multi-source heterogeneous power grid data and construct a knowledge base based on the multi-source heterogeneous power grid data; Based on the knowledge base, deterministic rule verification is performed on the operation instructions to be verified, and instructions that violate preset rules are filtered out; Construct a large model of operation instructions, train it using a knowledge base, obtain a well-trained large model of operation instructions, and output scheduling instructions that meet safety constraints. The training process of the large-scale operation instruction model includes: The operation instruction model is trained by reinforcement learning, and guided to generate safe and compliant operation instructions based on multi-dimensional reward signals that include physical feasibility, safety constraints, and scheduling procedure compliance.

[0024] The multidimensional reward signal includes: The physical feasibility reward is used to evaluate the convergence of power flow calculations after the command is executed. Safety constraint rewards are used to evaluate the degree to which voltage and line thermal stability safety constraints are met after the command is executed; The scheduling procedure compliance reward is used to evaluate the degree to which instructions conform to expert experience and rules.

[0025] For operation instructions that pass the deterministic rule verification, power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events are executed sequentially. If any level verification fails, the operation instruction big model automatically reconstructs the scheduling instruction based on the over-limit information or fault information fed back by the verification, and re-executes the three-level verification until all three-level verifications are passed.

[0026] In the three-level verification, the first-level power flow is used to determine whether voltage over-limit, branch overload, or transformer overload occurs after the execution of the operation command. The specific steps include: The complete topology and power flow data are obtained from the knowledge base to form a current operating profile that includes all node voltages, branch parameters, reactive power compensation, etc. Based on the type of operation instruction verified by the deterministic rules, the current operating section is modified to obtain the power grid status section after the operation; Calculate the power flow on the grid state profile after the operation, obtain the power flow calculation results, and perform limit checks on node voltages, branch currents, and transformer loads item by item: If there are no violations, the operation instruction is determined to have passed the first level of verification and is sent to the second level. If any of the following conditions exist: voltage overload, branch overload, or transformer overload, the power flow overload verification result under the operation command is determined to be unsuccessful. The overload information, including the overloaded device and its overload quantity, as well as the topology, is transmitted to the trained operation command big model. The operation command big model is adjusted online based on the overload information, and the adjusted operation command is re-executed for the first-level power flow calculation verification until the output passes.

[0027] In the three-level verification, the second-level N-1 security analysis verification further verifies the power grid security after the failure of any critical equipment by the operation instructions verified by the first-level verification. The specific steps include: By combining equipment importance labels with topological adjacency relationships, an N-1 set of security analysis faults is generated; For each faulty component in the N-1 security analysis fault set, perform a break simulation and recalculate the power flow on the new topology after the break. If all faulty components meet the conditions, then the N-1 safety criterion is satisfied under the operation command; the conditions include power flow convergence after interruption, no voltage over-limit, and no branch overload. If any faulty component causes any of the above conditions to be unmet, the operation is deemed unsuccessful. The failure information is output to the large operation instruction model to obtain the adjusted operation instruction, and the first and second level verifications are performed again until the operation is successful.

[0028] In the three-level verification, the third level, which is for extreme power grid scenarios and major events, performs a final verification on the operation instructions that have passed the first two levels of verification. The specific steps include: In extreme weather scenarios, meteorological factors are introduced to verify the dynamic current carrying capacity of the line; the overload capacity of the transformer is verified under extreme high temperature conditions; and the instantaneous failure rate of each equipment component under the current weather scenario is calculated; equipment components with instantaneous failure rates higher than a set threshold are formed into a failure chain. Calculate the product of the instantaneous failure rates of each device component in the fault chain under the current extreme scenario, as the probability of the fault chain occurring; calculate the total load loss after the fault chain is triggered, count the number and magnitude of nodes that exceed voltage or frequency limits after the fault, and determine whether it causes system instability; normalize the results after the above fault chain is triggered, and sum the normalized values ​​by weight to obtain the severity of the consequences. Multiply the probability of occurrence by the severity of the consequences to obtain the overall risk value; If no limit exceedance, instability, or load loss occurs under any extreme weather scenario, and the overall risk value is below the threshold, then the third-level verification is passed. If an over-limit, instability, or loss of load occurs, or if the comprehensive risk value is not lower than the threshold, the third-level verification is deemed to have failed. The fault chain list and the preset preventive control measures are then transmitted to the large-scale operation instruction model for online adjustment. The adjusted operation instructions are then re-executed through the third-level verification process until the third-level verification passes.

[0029] This invention also proposes an adaptive adjustment system for power grid operation modes, comprising a knowledge base storage unit, a deterministic reasoning unit, a large model training and reasoning unit, and an online adjustment and verification unit: The knowledge base storage unit acquires multi-source heterogeneous power grid data and constructs a knowledge base based on the multi-source heterogeneous power grid data; The deterministic reasoning unit performs deterministic rule verification on the operation instructions to be checked, and filters out instructions that violate preset rules; The large model trains the inference unit using a knowledge base to obtain a trained large model of operation instructions, which outputs scheduling instructions that comply with safety constraints. The online adjustment and verification unit, for operation instructions that have passed the determination rule verification, sequentially performs power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events. If any level verification fails, the operation instruction big model automatically reconstructs the scheduling instruction based on the over-limit information or fault information fed back by the verification, and re-executes the three-level verification until all three-level verifications are passed.

[0030] Furthermore, in the online adjustment and verification unit, the first-level power flow is used to determine whether voltage over-limit, branch overload, or transformer overload occurs after the operation command is executed; The second-level N-1 security analysis and verification further verifies the power grid security after the failure of any critical equipment by using the operation instructions verified by the first-level verification. The third level of verification, which targets extreme power grid scenarios and major events, performs a final verification based on the operation instructions from the first two levels of verification.

[0031] The present invention also proposes a terminal, including a processor and a storage medium: The storage medium is used to store instructions; The processor is used to perform the steps of the above method according to the instructions.

[0032] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0033] Example 1 This invention proposes the following technical solution, specifically a method for adaptive adjustment of power grid operation mode, the steps of which are as follows: Step 1: Construct a knowledge base based on multi-source heterogeneous power grid data; Step 1.1: Obtain multi-source heterogeneous power grid data, including multi-source heterogeneous power grid data and data used to construct an unstructured knowledge base: The data used to build the structured knowledge base mainly includes unstructured texts such as power grid dispatch plans, power grid dispatch procedures, and operating regulations. These texts contain primary equipment entities (such as circuit breakers, disconnectors, grounding switches, lines, transformers, busbars, etc. and their attributes), secondary equipment entities (such as protection devices, automatic transfer switches, etc. and their attributes), and rule entities containing rich logical relationships, which constitute the core knowledge of power grid operation. Primary equipment entities include circuit breakers, disconnectors, grounding switches, lines, transformers, busbars, etc. Equipment information such as equipment ID, rated voltage, rated current, active power, and reactive power is extracted from primary equipment, and the topological relationships between equipment are established. Secondary equipment entities include protection devices, automatic transfer switches, etc. Equipment information such as equipment ID, protection type, and protection range is extracted from secondary equipment. Rule entities are extracted from unstructured text such as power grid dispatching plans, power grid dispatching procedures, and operating regulations.

[0034] A power grid dispatching plan is a pre-formulated, standardized operational procedure designed to address specific changes in the operating status of the power grid. It includes key information such as power grid equipment, power grid operating modes, power grid operation instructions, and power grid stability limits.

[0035] Power grid dispatching regulations are a set of mandatory technical and management standards formulated by the state or power grid enterprises. They systematically stipulate the principles, procedures, standards, and safety criteria for power grid dispatching operations. They include entities such as equipment operation constraints, safety boundaries, and stability limits, and contain rich logical relationships.

[0036] Operating procedures are standardized operating steps, abnormal handling methods, and safety precautions specified for specific equipment, systems, and operating scenarios. Operating procedures include standard operating procedures for specific equipment, operating parameters, abnormal and defect handling, and operational risk warnings.

[0037] Data used to build a structured knowledge base: The system receives telemetry data, remote signaling data, operational status, limits, and operational priorities from the SCADA system via a forward isolation device, and acquires meteorological data from a meteorological interface. An unstructured knowledge base is then constructed based on this acquired data. The telemetry data includes voltage, current, active power, reactive power, units, and timestamps; the remote signaling data includes location information, protection actions, alarm signals, and status values; and the meteorological information includes temperature, humidity, wind speed, and light intensity.

[0038] Step 1.2, Power Grid Data Preprocessing: To extract effective information from multi-source heterogeneous power grid data, this invention cleans the power grid data based on the characteristics of multi-source heterogeneous power grid text data in the field of power grid dispatching.

[0039] First, format standardization is performed by using text parsing tools to convert DPF, Word, and image information into a unified text format, eliminating format differences. Next comes noise removal, which primarily employs a dual filtering model based on information entropy and domain rules. The information entropy of the text fragment is defined as:

[0040] in, Represents a text fragment The information entropy value indicates that the higher the entropy value, the more chaotic and uncertain the character distribution. Character exist The probability of a character appearing in a text is usually estimated by dividing the frequency of the character's appearance by the total number of characters in the text.

[0041] Its decision function is defined as:

[0042] in, The threshold is dynamically adjusted based on historical data. Based on the power grid professional terminology dictionary The rule matching function is used to filter generic text or advertising information that is not related to the domain.

[0043] The text then enters the text standardization stage. Through a constructed standard terminology mapping table and numerical unit conversion rules, the text is systematically corrected. For example, "main transformer" is standardized to "main transformer transformer," and "220 kV" is unified to "220 kV," ensuring consistency in expression. Next, information completion is performed, mainly based on simple context and domain common sense, filling in obviously missing commonly used units and standard terminology abbreviations. Finally, the cleaned documents are scored and filtered using preset quality assessment indicators, laying a reliable foundation for downstream tasks. In the preferred embodiment of this invention, the preset quality assessment indicators are calculated by using the BM25 algorithm to calculate the similarity score between terms related to power grid operation modes and the cleaned text. High-quality text is selected based on the score threshold, while low-scoring text is excluded.

[0044] Step 1.3, Construction of Knowledge Base Storage Units Based on preprocessed text data and real-time access to SCADA and meteorological data, two types of knowledge bases are constructed in parallel: Construction of a structured knowledge base: Extract power grid equipment entities, their attributes (such as equipment ID and rated parameters), topological connections between equipment, operating rules and constraints, etc. from preprocessed unstructured text (dispatch plans, procedures, etc.) to form a structured knowledge set with entities, relationships and attributes as the core.

[0045] When building a structured knowledge base, each device entity must be associated with its core attribute of voltage level, and the graph needs to establish clear electrical connection point entities to accurately describe the terminal connection relationships between devices, thereby forming a complete and computable physical connection network of the power grid.

[0046] Construction of unstructured knowledge base: Telemetry, teleindication, status, limit and meteorological data that are accessed in real time from SCADA system and meteorological interface are organized, stored and managed according to their spatiotemporal attributes and measurement point information to form a data set reflecting the real-time operating status of the power grid and the external environment.

[0047] When constructing an unstructured knowledge base, the connections not only represent the physical connectivity between two devices, but also need to embed rich dynamic attributes, including attributes such as "current operational status" (operating / hot standby / cold standby / maintenance), "standby type," "application priority," "switching time," "limit value," "limit type," and "effective conditions." These attributes embedded in the relationships transform static network connections into a dynamic, rule-based, and constraint-filled operational sandbox, enabling the knowledge graph to understand the security implications and operational context behind the instructions, truly achieving the leap from "describing the power grid" to "instruction operation."

[0048] Step 2: Perform rapid knowledge verification on the knowledge base based on deterministic rules: After the knowledge base storage unit is constructed, the received operation instructions are subjected to rapid and interpretable compliance verification based on deterministic rules. This step aims to utilize the pre-stored structured knowledge and rules in the knowledge base, combined with real-time runtime data. Its execution process is divided into five sub-steps: loading the operation device, recalling relevant rules, obtaining real-time binding values, condition judgment, and comprehensive output.

[0049] First, during the loading of the operating device phase, the device entity and all its attributes, including rated parameters, current status, and associated devices, are retrieved from the knowledge graph based on the target device ID, providing a complete device context for subsequent rule matching. Then, the relevant rule recall phase begins. A graph traversal algorithm is used, starting from the operating device node and performing a breadth-first search along predefined relational edges in the knowledge graph, with a search depth limited to 2. Taking a closing circuit breaker as an example, this algorithm can recall interlocked grounding switch rules, rules governing permissible closing conditions for connected lines, rules governing the voltage range of the busbar, rules governing protection device coordination, and rules limiting the number of device operations.

[0050] After successfully recalling the relevant rules, the process proceeds to the real-time binding value acquisition stage. For each atomic condition in a rule, the unstructured knowledge base is queried to obtain the valid value for that measurement point. If the data is invalid or the data update time exceeds 5 seconds, the atomic condition is marked as "undecidable" to avoid making dangerous decisions based on outdated or erroneous data. Subsequently, condition judgment is performed, comparing the real-time value with the rule threshold. Only when all atomic conditions in a rule are satisfied is the rule judged as passed; otherwise, a violation is recorded, including the device, expected value, actual value, and the corresponding rule ID.

[0051] Finally, a comprehensive output determination is performed: If all rules pass and there are no uncertainties, output "Pass" along with a summary of all passed rules, and push the passed operation instructions to the subsequent online adjustment and verification steps for further three-level verification; If at least one rule is violated and there are no uncertainties, the output is "Fail," and a structured cause tree is generated. The root node of this tree represents the prohibited operation, and each child node corresponds to specific information about a rule violation. If a rule is incorrect, it is modified based on expert experience and then pushed to subsequent online adjustment and verification steps for further three-level verification. Specific information may include errors in the execution order of verification operation instructions, incomplete device objectification information for operation instructions, etc.

[0052] Through the above design, the deterministic verification of knowledge graphs can quickly output clear conclusions in typical scenarios and provide a clear decision basis for subsequent branch processing.

[0053] Step 3, Construction and training of the large-scale operation instruction model: To handle complex scheduling reasoning problems that go beyond deterministic rules, a large-scale operational instruction derivation model needs to be constructed and trained. The training of this model relies on the power grid domain knowledge contained in the structured standard library and unstructured knowledge base constructed in step 1. The training process is as follows: Figure 2 As shown.

[0054] The model building and training process includes the following stages: Base model fine-tuning and domain adaptation: The Qwen3-32B model was selected as the base model. Data related to power grid operation commands were used, and LoRA (Low-Rank Adaptation) technology was employed to fine-tune the base model. The advantage of this method is that it maintains the generality of the pre-trained model while adapting to downstream tasks with minimal parameters. It also supports efficient switching and deployment between multiple tasks. The LoRA calculation formula is as follows:

[0055] in, This refers to the original weight matrix frozen in the pre-trained model. For input feature variables, and These are the trainable downstream and upstream parameter matrices in the low-rank adapter, respectively. This allows for optimization only during fine-tuning. and With only a few parameters, efficient task adaptation can be achieved without changing the original model architecture.

[0056] Decision-making ability optimization based on reinforcement learning: Building upon fine-tuning, reinforcement learning techniques were incorporated to further enhance the model's decision-making ability to generate safe and compliant operating instructions through trial and error in interaction with the simulation environment. Training was conducted in a power grid simulation environment integrated with a power flow calculator, and the model's learning was guided by setting three dimensions of reward signals: A physical feasibility reward is given to assess whether the power flow calculation converges after the command is executed. A positive reward is given for convergence, and a negative reward is given for divergence. Safety constraint rewards quantify whether safety constraints such as voltage and line thermal stability are violated after the command is executed. The more violations, the greater the negative reward. The scheduling procedure compliance reward combines rule base matching and a small discriminative model trained on historical data to evaluate the similarity of instructions to expert experience in natural semantics and logic.

[0057] By exploring randomly generated fault scenarios, the model continuously optimizes its strategy (i.e., the way instructions are generated) through the aforementioned reward function, thereby learning to simulate the decision-making process of a scheduling expert. This method significantly improves the inference accuracy of complex operation instructions.

[0058] Step 4, Online Level 3 Security Verification and Adjustment: The operation commands undergo online security verification and dynamic adjustment. This step ensures the security of the commands through three progressively higher levels of verification: the first level is power flow calculation verification, the second level is N-1 security analysis verification, and the third level is a three-level verification for extreme power grid scenarios and major events.

[0059] Step 4.1, Perform the first level: power flow calculation verification After initial screening in step 2, the operational instructions in the contingency plan proceed to this step. Power flow calculations are performed on the instructions under actual power grid operation to determine if any over-limit events (such as voltage over-limit, branch overload, or transformer overload) have occurred.

[0060] Specifically, the latest complete topology and power flow data are obtained from the knowledge base constructed in step 1 to form a current operating profile that includes all node voltages, branch parameters, reactive power compensation, etc.

[0061] Furthermore, the above-mentioned cross-sections are precisely modified according to the type of operation instruction (such as closing circuit breakers, opening circuit breakers, adjusting transformer taps, switching capacitors) to generate a post-operation power grid status cross-section.

[0062] Specific modification methods include: Circuit breaker: Directly removes the target branch from the network topology. The program will also delete the admittance matrix elements of the branch and adjust the node relationships.

[0063] Adjusting transformer tap change: Changing the turns ratio in the transformer equivalent model according to instructions.

[0064] Switching capacitors: Modifies the reactive power injection of a specified bus. When switching on a capacitor bank, a set of rated reactive power is added to the reactive power injection node of that bus; when switching off, the corresponding capacity is subtracted.

[0065] After completing the above modifications, a post-operation power grid state profile is generated, which includes the complete topology, parameters, and initial values ​​of variables.

[0066] Furthermore, power flow calculations are performed on the power grid state section after the operation. After obtaining the power flow calculation results for the section after the operation, limit judgments are made for each item, including node voltage, branch current, and transformer load. Each limit is not a fixed constant, but changes dynamically with equipment type, environmental conditions, and operating procedures.

[0067] Specifically, if there are no violations, the instruction is determined to have passed the first level of verification and is sent to the second level (N-1 security analysis verification). If any of the following conditions exists: voltage overload, branch overload, or transformer overload, the power flow overload check result under this operation is determined to be "fail". The overload information (including the overloaded device and its overload amount) and topology are transmitted to the large-scale operation command model trained in step 3, which then makes online adjustments based on this constraint information. The adjusted new command re-executes the first-level power flow calculation and check until the output is "pass". The adjustment and verification are cyclical and continuous. If the process fails more than five times, the operation command is sent to the dispatcher for adjustment.

[0068] Step 4.2, perform the second level: N-1 security analysis and verification: For instructions that pass the first level of verification, further verification is performed on their power grid security after any critical equipment failure (N-1).

[0069] Specifically, by combining equipment importance labels with topological adjacency relationships, an N-1 security analysis fault set is generated, such as all incoming and outgoing lines of the bus where the operating equipment is located, and critical lines adjacent to the bus. For each faulty element in the N-1 security analysis fault set, an interruption simulation is performed, and power flow calculations are recalculated on the new topology after the interruption.

[0070] Furthermore, if all faulty components meet the conditions (i.e., power flow convergence after interruption, no voltage over-limit, and no branch overload), then the system is deemed to meet the N-1 safety criterion under this operation command. If any faulty component causes any of the above conditions to be unmet, the result is "failed". The reason for failure (such as a fault causing an overload on a line) is output to the large operation instruction model for adjustment. The adjusted instruction is then re-verified at the first and second levels until it passes.

[0071] After the second-level verification is passed, the operation instructions are sent to the third-level verification, which is a three-level verification for extreme power grid scenarios and major events.

[0072] Step 4.3, Perform Level 3: Verification for extreme power grid scenarios and major events: For instructions that pass the first two levels of verification, a final verification is performed under more stringent and extreme conditions.

[0073] Specifically, the verification covers peak summer demand as well as extreme weather scenarios such as typhoons, ice storms, and lightning. Meteorological factors such as wind speed and icing are introduced to verify the dynamic current carrying capacity of the lines; the overload capacity of the transformers is verified under extreme high-temperature conditions; and the instantaneous failure rate of each component under the current weather scenario is calculated, marking equipment with a failure rate higher than 70%. Equipment with a failure rate higher than 70% is grouped into a failure chain.

[0074] Calculate the overall risk value of a failure chain consisting of high-failure-rate components under the current extreme scenario. Simultaneously, simulate the simultaneous failure (NK) of multiple components triggered by preset major events (such as DC blockage, large unit tripping, etc.) to determine whether the system is unstable.

[0075] Furthermore, under the current extreme scenario, the probability of occurrence of each fault chain is multiplied by the severity of its consequences to obtain a comprehensive risk value. The probability of occurrence of a fault chain can be simplified to the product of the instantaneous failure rates of each component in the chain under the current extreme scenario, or estimated using conventional reliability engineering methods such as fault tree analysis and Monte Carlo simulation. For example, for a simple fault chain consisting of components A and B, its probability of occurrence can be estimated as the product of the instantaneous failure rate of component A and the conditional failure rate of component B after A's failure. The severity of the consequences is used to measure the degree of impact on power grid safety after the fault chain occurs. Its quantification can be based on the assessment of the consequences triggered by the fault chain, including but not limited to: Calculate the total load loss (MW) caused by the fault chain; The number and magnitude of nodes exceeding voltage or frequency limits after a fault are statistically analyzed. Determine whether it will cause system instability (such as transient power angle instability or voltage collapse); One or more of the above physical quantities are mapped to a standardized severity score between 0 and 1 using preset weights. For example, the severity S can be defined as: S = (load loss / total system load) * α + (number of nodes exceeding limits / total number of nodes) * β, where α and β are preset weight coefficients, and α + β = 1.

[0076] If the system does not experience any limit exceedances, instability, or load loss under all extreme weather scenarios, and the overall risk value is below the threshold, then it is deemed to have passed the third-level verification. At this point, the operation command has passed three levels of rigorous verification and can be confirmed to have a high degree of safety, capable of handling normal, faulty, and even extreme operating conditions. Furthermore, if the conditions are not met or the overall risk value is not lower than the threshold, the third-level verification is deemed "failed," and the high-risk failure chain list and preset preventive control measures are transmitted to the operational instruction big data model for online adjustment. The adjusted instructions need to be re-executed through the complete three-level verification process until the third-level verification passes. The threshold can be determined based on historical accident statistics, risk tolerance, or industry guidelines.

[0077] The adjusted operating instructions have now passed power flow calculation, N-1 security analysis, and three-level verification for extreme power grid scenarios and major events. This ensures the safe and stable operation of the power grid under normal operating conditions, occasional faults, and even extreme disasters, significantly reducing operational risks and enhancing the power grid's ability to cope with complex uncertainties.

[0078] Example 2 like Figure 3 As shown, this invention also proposes an adaptive adjustment system for power grid operation modes, including a knowledge base storage unit, a deterministic reasoning unit, a large model training and reasoning unit, and an online adjustment and verification unit: The knowledge base storage unit acquires multi-source heterogeneous power grid data and constructs a knowledge base based on the multi-source heterogeneous power grid data; The deterministic reasoning unit performs deterministic rule verification on the operation instructions to be checked, and filters out instructions that violate preset rules; The large model trains the inference unit using a knowledge base to obtain a trained large model of operation instructions, which outputs scheduling instructions that comply with safety constraints. The online adjustment and verification unit, for operation instructions that have passed the determination rule verification, sequentially performs power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events. If any level verification fails, the operation instruction big model automatically reconstructs the scheduling instruction based on the over-limit information or fault information fed back by the verification, and re-executes the three-level verification until all three-level verifications are passed.

[0079] Furthermore, in the online adjustment and verification unit, the first-level power flow is used to determine whether voltage over-limit, branch overload, or transformer overload occurs after the operation command is executed; The second-level N-1 security analysis and verification further verifies the power grid security after the failure of any critical equipment by using the operation instructions verified by the first-level verification. The third level of verification, which targets extreme power grid scenarios and major events, performs a final verification based on the operation instructions from the first two levels of verification.

[0080] Example 3 The present invention also proposes a terminal, including a processor and a storage medium: The storage medium is used to store instructions; The processor is used to perform the steps of the above method according to the instructions.

[0081] Example 4 The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0082] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0083] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0084] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0085] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0086] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for adaptive adjustment of power grid operation mode, characterized in that, include: Acquire multi-source heterogeneous power grid data and construct a knowledge base based on the multi-source heterogeneous power grid data; Based on the knowledge base, deterministic rule verification is performed on the operation instructions to be verified, and instructions that violate preset rules are filtered out; Construct a large model of operation instructions, train it using a knowledge base, obtain a well-trained large model of operation instructions, and output scheduling instructions that meet safety constraints. For operation instructions that pass the deterministic rule verification, power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events are executed sequentially. If any level verification fails, the operation instruction big model automatically reconstructs the scheduling instruction based on the over-limit information or fault information fed back by the verification, and re-executes the three-level verification until all three-level verifications are passed.

2. The adaptive adjustment method for power grid operation mode according to claim 1, characterized in that: The training process of the large-scale operation instruction model includes: The operation instruction model is trained by reinforcement learning, and guided to generate safe and compliant operation instructions based on multi-dimensional reward signals that include physical feasibility, safety constraints, and scheduling procedure compliance.

3. The adaptive adjustment method for power grid operation mode according to claim 2, characterized in that: The multidimensional reward signal includes: The physical feasibility reward is used to evaluate the convergence of power flow calculations after the command is executed. Safety constraint rewards are used to evaluate the degree to which voltage and line thermal stability safety constraints are met after the command is executed; The scheduling procedure compliance reward is used to evaluate the degree to which instructions conform to expert experience and rules.

4. The adaptive adjustment method for power grid operation mode according to claim 1, characterized in that: In the three-level verification, the first-level power flow is used to determine whether voltage over-limit, branch overload, or transformer overload occurs after the execution of the operation command. The specific steps include: The complete topology and power flow data are obtained from the knowledge base to form a current operating profile that includes all node voltages, branch parameters, reactive power compensation, etc. Based on the type of operation instruction verified by the deterministic rules, the current operating section is modified to obtain the power grid status section after the operation; Calculate the power flow on the grid state profile after the operation, obtain the power flow calculation results, and perform limit checks on node voltages, branch currents, and transformer loads item by item: If there are no violations, the operation instruction is determined to have passed the first level of verification and is sent to the second level. If any of the following conditions exist: voltage overload, branch overload, or transformer overload, the power flow overload verification result under the operation command is determined to be unsuccessful. The overload information, including the overloaded device and its overload quantity, as well as the topology, is transmitted to the trained operation command big model. The operation command big model is adjusted online based on the overload information, and the adjusted operation command is re-executed for the first-level power flow calculation verification until the output passes.

5. The adaptive adjustment method for power grid operation mode according to claim 1, characterized in that: In the three-level verification, the second-level N-1 security analysis verification further verifies the power grid security after the failure of any critical equipment by the operation instructions verified by the first-level verification. The specific steps include: By combining equipment importance labels with topological adjacency relationships, an N-1 set of security analysis faults is generated; For each faulty component in the N-1 security analysis fault set, perform a break simulation and recalculate the power flow on the new topology after the break. If all faulty components meet the conditions, then the N-1 safety criterion is satisfied under the operation command; the conditions include power flow convergence after interruption, no voltage over-limit, and no branch overload. If any faulty component causes any of the conditions mentioned above to be unmet, the operation is deemed unsuccessful. The failure information is output to the large operation instruction model to obtain the adjusted operation instruction, and the first and second level verifications are performed again until the operation is successful.

6. The adaptive adjustment method for power grid operation mode according to claim 1, characterized in that: In the three-level verification, the third level, which is for extreme power grid scenarios and major events, performs a final verification on the operation instructions that have passed the first two levels of verification. The specific steps include: In extreme weather scenarios, meteorological factors are introduced to verify the dynamic current carrying capacity of the line; the overload capacity of the transformer is verified under extreme high temperature conditions; and the instantaneous failure rate of each equipment component under the current weather scenario is calculated; equipment components with instantaneous failure rates higher than a set threshold are formed into a failure chain. Calculate the product of the instantaneous failure rates of each device component in the fault chain under the current extreme scenario, as the probability of the fault chain occurring; calculate the total load loss after the fault chain is triggered, count the number and magnitude of nodes that exceed voltage or frequency limits after the fault, and determine whether it causes system instability; normalize the results after the above fault chain is triggered, and sum the normalized values ​​by weight to obtain the severity of the consequences. Multiply the probability of occurrence by the severity of the consequences to obtain the overall risk value; If no limit exceedance, instability, or load loss occurs under any extreme weather scenario, and the overall risk value is below the threshold, then the third-level verification is passed. If an over-limit, instability, or loss of load occurs, or if the comprehensive risk value is not lower than the threshold, the third-level verification is deemed to have failed. The fault chain list and the preset preventive control measures are then transmitted to the large-scale operation instruction model for online adjustment. The adjusted operation instructions are then re-executed through the third-level verification process until the third-level verification passes.

7. A power grid operation mode adaptive adjustment system utilizing the method of any one of claims 1-6, comprising a knowledge base storage unit, a deterministic reasoning unit, a large model training and reasoning unit, and an online adjustment and verification unit, characterized in that: The knowledge base storage unit acquires multi-source heterogeneous power grid data and constructs a knowledge base based on the multi-source heterogeneous power grid data; The deterministic reasoning unit performs deterministic rule verification on the operation instructions to be checked, and filters out instructions that violate preset rules; The large model trains the inference unit using a knowledge base to obtain a trained large model of operation instructions, which outputs scheduling instructions that comply with safety constraints. The online adjustment and verification unit, for operation instructions that have passed the determination rule verification, sequentially performs power flow calculation verification, N-1 security analysis verification, and three-level verification for extreme power grid scenarios and major events. If any level verification fails, the operation instruction big model automatically reconstructs the scheduling instruction based on the over-limit information or fault information fed back by the verification, and re-executes the three-level verification until all three-level verifications are passed.

8. The adaptive adjustment system for power grid operation mode according to claim 7, characterized in that: In the online adjustment and verification unit, the first level of power flow is used to determine whether voltage over-limit, branch overload or transformer overload occurs after the operation command is executed; The second-level N-1 security analysis and verification further verifies the power grid security after the failure of any critical equipment by using the operation instructions verified by the first-level verification. The third level of verification, which targets extreme power grid scenarios and major events, performs a final verification based on the operation instructions from the first two levels of verification.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-6.

Images